Operations and

Services
Generative AI
Health
Education

MARCH 2025 www.computer.org

 PUBLISH WITH THE
IEEE COMPUTER SOCIETY

Break
Free.
You Have
Choices.
It’s Author’s Choice:
IEEE Computer Society publishes
fully open access journals as well
as hybrid journals and magazines
to accommodate the unique
needs of all researchers.

www.computer.org/cfp
 IEEE COMPUTER SOCIETY computer.org

STAFF
Editor Periodicals Portfolio Senior Managers Director, Periodicals and Special Projects
Lucy Holden Carrie Clark and Kimberly Sperka Robin Baldwin

Production & Design Artist Periodicals Operations Project Specialists Senior Advertising Coordinator
Carmen Flores-Garvey Priscilla An and Christine Shaughnessy Debbie Sims

Circulation: ComputingEdge (ISSN 2469-7087) is published monthly by the IEEE Computer Society. IEEE Headquarters, Three Park Avenue, 17th
Floor, New York, NY 10016-5997; IEEE Computer Society Publications Office, 10662 Los Vaqueros Circle, Los Alamitos, CA 90720; voice +1 714 821 8380;
fax +1 714 821 4010; IEEE Computer Society Headquarters, 2001 L Street NW, Suite 700, Washington, DC 20036.
Postmaster: Send address changes to ComputingEdge-IEEE Membership Processing Dept., 445 Hoes Lane, Piscataway, NJ 08855. Periodicals Postage
Paid at New York, New York, and at additional mailing offices. Printed in USA.
Editorial: Unless otherwise stated, bylined articles, as well as product and service descriptions, reflect the author’s or firm’s opinion. Inclusion in
ComputingEdge does not necessarily constitute endorsement by the IEEE or the Computer Society. All submissions are subject to editing for style,
clarity, and space.
Reuse Rights and Reprint Permissions: Educational or personal use of this material is permitted without fee, provided such use: 1) is not made for
profit; 2) includes this notice and a full citation to the original work on the first page of the copy; and 3) does not imply IEEE endorsement of any third-
party products or services. Authors and their companies are permitted to post the accepted version of IEEE-copyrighted material on their own Web
servers without permission, provided that the IEEE copyright notice and a full citation to the original work appear on the first screen of the posted copy.
An accepted manuscript is a version which has been revised by the author to incorporate review suggestions, but not the published version with copy-
editing, proofreading, and formatting added by IEEE. For more information, please go to: http://www.ieee.org/publications_standards/publications
/rights/paperversionpolicy.html. Permission to reprint/republish this material for commercial, advertising, or promotional purposes or for creating new
collective works for resale or redistribution must be obtained from IEEE by writing to the IEEE Intellectual Property Rights Office, 445 Hoes Lane,
Piscataway, NJ 08854-4141 or [email protected]. Copyright © 2025 IEEE. All rights reserved.
Abstracting and Library Use: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy for private use of patrons,
provided the per-copy fee indicated in the code at the bottom of the first page is paid through the Copyright Clearance Center, 222 Rosewood Drive,
Danvers, MA 01923.
Unsubscribe: If you no longer wish to receive this ComputingEdge mailing, please email IEEE Computer Society Customer Service at help@
computer.org and type “unsubscribe ComputingEdge” in your subject line.
IEEE prohibits discrimination, harassment, and bullying. For more information, visit www.ieee.org/web/aboutus/whatis/policies/p9-26.html.

2025 IEEE Computer Society Magazine Editors in Chief

Computer IEEE Intelligent Systems IEEE Security & Privacy

Jeff Voas, NIST Bo An, Nanyang Sean Peisert, Lawrence
Technological University Berkeley National Laboratory
Computing in Science and University of California,
& Engineering IEEE Internet Computing Davis
Jeffrey Carver, Weisong Shi, University of
University of Alabama Delaware IEEE Software
Sigrid Eldh, Ericsson,
IEEE Annals of the History IEEE Micro Mälardalen University,
of Computing Hsien-Hsin Sean Lee, Sweden; Carleton
Troy Astarte, Intel Corporation University, Canada
Swansea University
IEEE MultiMedia IT Professional
IEEE Computer Graphics Balakrishnan Prabhakaran, Charalampos Z.
and Applications University of Texas at Dallas Patrikakis, University of
Pak Chung Wong, Trovares West Attica
and Bill & Melinda Gates IEEE Pervasive Computing
Foundation (Interim EIC) Fahim Kawsar, Nokia Bell Labs
and University of Glasgow

2469-7087/25 © 2025 IEEE Published by the IEEE Computer Society March 2025 1
MARCH 2025 • VOLUME 11 • NUMBER 3

8
Building the
14
Convergence
34
Algorithms That
Factory of the of IT, OT, and Can Deny Care,
Future With the the IoT and a Call for AI
Industrial Internet Explainability
of Things
 Operations and Services
8 Building the Factory of the Future With the Industrial
Internet of Things
JEFF DANIELS

14 Convergence of IT, OT, and the IoT

CHRISTOF EBERT

Generative AI
20 Nothing Is Harder to Resist Than the Temptation of AI
ANDREW PARK, JAN KIETZMANN, JAYSON KILLORAN, YUANYUAN CUI,
PATRICK VAN ESCH, AND AMIR DABIRIAN

28 Unleashing the Potential: Integrating ChatGPT and

the Internet of Things for Enhanced User Experiences
and Automation
MOHAMAD KASSAB AND JOANNA F. DeFRANCO

Health
34 Algorithms That Can Deny Care, and a Call for
AI Explainability
DOMENICO TALIA

38 Lifeblood of Health is Data

RAMESH JAIN

Education
46 The Future of Education: Generative Artificial
Intelligence’s Collaborative Role With Teachers
NIR KSHETRI

52 Leverage Biology to Learn Rapidly From Mistakes

Without Feeling Like a Failure
LAUREN E. MARGULIEUX, JAMES PRATHER, MASOUMEH RAHIMI, AND
GOZDE CETIN UZUN

Departments
4 Magazine Roundup
7 Editor’s Note: Merging and Mastering IT, OT, and the IoT
60 Conference Calendar

Subscribe to ComputingEdge for free at

www.computer.org/computingedge
 Magazine Roundup

T he IEEE Computer Society’s lineup of 12 peer-reviewed technical magazines covers cutting-edge topics rang-
ing from software design and computer graphics to Internet computing and security, from scientific appli-
cations and machine intelligence to visualization and microchip design. Here are highlights from recent issues.

Brazil, that aim to increase wom- research, technical development,

A Vision for Transformative en’s access, skills, and leadership and ultimately commerce.
Intersections in technology.

The rapid evolution of connected

and autonomous vehicles (CAVs) A Chronicle of Quantum
is ushering in a transformative era Computer-Aided Design in Technologies in Game and
for transportation, challenging the United States, 1949–1984: Software Development
the adequacy of traditional inter- Designing in a Closed World
section designs. To support the While more than 300 quantum
rapid development, validation, and Drawing together four intercon- games have been developed by
deployment of adaptive intersec- nected contexts or sites—radar enthusiasts and commercial par-
tion systems that meet the com- defense, numerical control, tech- ties, a comprehensive research
plex demands of CAVs, and to opti- nology transfer programs, and the program or state-of-the-art review
mize both traffic efficiency and gradual formation of a software is notably absent. This Septem-
road safety, the authors of this market—this article, featured in ber/October 2024 IEEE Computer
article, featured in the Decem- the October–December 2024 issue Graphics and Applications article
ber 2024 issue of Computer, of IEEE Annals of the History of addresses this gap by conducting
introduce a framework called Computing, offers a history of com- a thorough literature review, pre-
OpenIntersection. puter-aided design in the United senting current advancements
States. It focuses on this technol- and examples in quantum games
ogy’s formative period, beginning and interactive designs, and
in the Second World War’s after- exploring future prospects for QT
Bridging the Gender Gap in math and ending in the transi- in game development practices.
Computing: Insights From Latin tional decade of 1980 when—with
America With a Focus on Brazil the rise of personal computing—
computer-aided design expanded
The gender gap in digital skills beyond research labs and indus- Explaining Imitation Learning
and computing persists glob- try niches into a multibillion-dollar Through Frames
ally despite initiatives to include business. It also traces the role of
women and girls. This October– political doctrines and economic As one of the prevalent meth-
December 2024 Computing in Sci- anxieties in shaping both aca- ods to achieve automation sys-
ence & Engineering article dis- demic and industry developments tems, imitation learning (IL) pres-
cusses recent initiatives in Latin in computer-aided design, foster- ents a promising performance in
America, with a special focus on ing it as a domain of speculative a wide range of domains. However,

4 March 2025 Published by the IEEE Computer Society  2469-7087/25 © 2025 IEEE
 despite the considerable improve- multimodal scene recognition net-
ment in policy performance, the work (SAD-MSR).
corresponding research on the Synchronous, Low-Latency,
explainability of IL models is still Off-Module Interface for the
limited. Inspired by the recent IBM z16 Telum Processor
approaches in explainable AI, the
authors of this November/Decem- The IBM Telum Processor1 is the Trigger-Action Programming
ber 2024 IEEE Intelligent Systems microprocessor chip for the lat- for Wellbeing: Insights From
article propose a model-agnostic est generation IBM z16 and Linux- 6590 iOS Shortcuts
explaining framework for IL mod- One Emperor 4 mainframe hybrid
els called Remove and Retrain cloud systems. At the outset of Trigger-action programming (TAP)
via Randomized Input Sampling the Telum Processor design, there platforms allow users to personal-
for Explanation (R2RISE). R2RISE were key challenges related to off- ize their digital ecosystems through
aims to explain the importance of chip connectivity. In this article, the definition of trigger-action
frames with respect to the overall featured in the November/Decem- rules. Yet, little is known about
policy performance. ber 2024 issue of IEEE Micro, the whether such a paradigm can be
authors describe some of those used to support users’ wellbeing. To
challenges and demonstrate the bridge this gap, the authors of the
innovations that allowed for the July–September 2024 issue of IEEE
Alignment Studio: Aligning success of the IBM z16. Pervasive Computing scraped 6590
Large Language Models trigger-action programs from iOS
to Particular Contextual shortcuts, and analyzed the data-
Regulations set to understand what aspects of
The Multimodal Scene their wellbeing users are already
The alignment of large language Recognition Method Based on programming and what opportuni-
models is usually done by model Self-Attention and Distillation ties remain untapped.
providers to add or control behav-
iors that are common or univer- Scene recognition is a challenging
sally understood across use cases task in computer vision because
and contexts. By contrast, in of the diversity of objects in scene Adoption Challenges for
this article, from the September/ images and the ambiguity of object Cryptographic Protocols
October 2024 issue of IEEE Inter- layouts. In recent years, the emer-
net Computing, the authors pres- gence of multimodal scene data In this article, featured in the Novem-
ent an approach and architecture has provided new solutions for ber/December 2024 issue of IEEE
that empowers application devel- scene recognition, but it has also Security & Privacy, the authors inter-
opers to tune a model to their par- brought new problems. To address viewed cryptography experts from
ticular values, social norms, laws, these challenges, the authors in academia and industry to learn from
and other regulations and orches- this October–December 2024 IEEE their experiences with the design
trate between potentially conflict- MultiMedia article propose the self- and deployment of cryptographic
ing requirements in context. attention and distillation-based protocols. They present adoption

www.computer.org/computingedge 5
MAGAZINE ROUNDUP

challenges, including misaligned to support an ML research and derived through academic research
incentives in academia and stan- development project at Jefferson from applications in large Internet-
dardization, mismatched assump- Lab. They describe a composable facing companies. To assist practi-
tions, low-quality reference imple- ML workflow, a custom CGroupV2 tioners in a broader range of compa-
mentations, and usability issues. exporter, and the implementation of nies, the authors of this September/
Prometheus, MLFlow, and Grafana. October 2024 IT Professional article
present guidelines for CE, based on
empirical data from interviews with
Establishing Machine Learning 27 practitioners at 12 companies of
Operations for Continual Evidence-Based Guidelines varying sizes and CE maturity.
Learning in Computing for Advancing Continuous
Clusters: A Framework for Experimentation
Monitoring and Optimizing
Cluster Behavior Continuous experimentation (CE) is Join the IEEE
used by many internet-facing com- Computer Society
The authors of this article from panies to improve the value of their
the January/February 2025 issue products based on user feedback
computer.org/join
of IEEE Software focus on build- gathered, e.g., through online exper-
ing a machine learning (ML) opera- iments using A/B testing. Frame-
tions continual learning capability works and theories for CE have been

ADVERTISER INFORMATION

Advertising Coordinator Central US, Northwest US, Southeast US, Asia/Pacific:

Eric Kincaid
Debbie Sims Email: [email protected]
Email: [email protected] Phone: +1 214-553-8513 | Fax: +1 888-886-8599
Phone: +1 714-816-2138 | Fax: +1 714-821-4010 Cell: +1 214-673-3742

Midwest US:
Advertising Sales Contacts Dave Jones
Email: [email protected]
Mid-Atlantic US, Northeast, Europe, the Middle East Phone: +1 708-442-5633 | Fax: +1 888-886-8599
and Africa: Cell: +1 708-624-9901
Dawn Scoda
Email: [email protected]
Jobs Board (West Coast and Asia), Classified Line Ads
Phone: +1 732-772-0160
Cell: +1 732-685-6068 | Fax: +1 732-772-0164 Heather Buonadies
Email: [email protected]
Southwest US, California:
Phone: +1 623-233-6575
Mike Hughes
Email: [email protected]
Cell: +1 805-208-5882 Jobs Board (East Coast and Europe), SE Radio Podcast

Marie Thompson
Email: [email protected]
Phone: +1 714-813-5094

6 ComputingEdge March 2025

 Editor’s Note

Merging and Mastering IT,

OT, and the IoT

T o remain competitive, it is
important for companies to
adapt to developing technological
“Building the Factory of the Future
With the Industrial Internet of
Things,” describes the technology
health care depending on how it
is being used. The author of “Algo-
rithms That Can Deny Care, and
changes that are rapidly transform- required to develop a factory of the a Call for AI Explainability,” from
ing the business world and influenc- future through combined IT and OT Computer, reveals flaws in the use
ing many aspects of life, from global design. Computer’s article, “Conver- of big data and machine learn-
markets to food services, manufac- gence of IT, OT, and the IoT,” outlines ing to make health care decisions
turing to smart cities, health to edu- the hurdles companies will need to and calls for algorithm explainabil-
cation. Maintaining relevance in the overcome to keep pace with the ity. In “Lifeblood of Health is Data,”
industry requires prioritizing inno- new business models and ecosys- from IEEE MultiMedia, the author
vation, which can include gathering tems created by IT, OT, and the IoT. explains how tapping into health
resources and making use of emerg- While GAI is a useful tool, it data can improve lifestyles and
ing tools to improve business opera- is still relatively new and risky. health care.
tions and practices. Failing to keep Companies must take a cautious Educators are always searching
up can result in a loss of relevance approach when incorporating GAI for ways to improve learning, from
and can have dire consequences for into their operations. In IT Profes- using GAI to biology. IT Professional
security, consumers, and bottom sional article “Nothing Is Harder article “The Future of Education:
lines. This issue of ComputingEdge to Resist Than the Temptation of Generative Artificial Intelligence’s
discusses new advancements— AI,” the authors review the current Collaborative Role With Teach-
and related challenges—at the con- state of GAI in business and poten- ers” analyzes how GAI enhances
vergence of information technology tial risks associated with its use. teaching and learning by evaluat-
(IT), operational technology (OT), “Unleashing the Potential: Inte- ing its impact on tasks performed
the Internet of Things (IoT), and GAI grating ChatGPT and the Internet by teachers. The article, “Leverage
(Generative AI). The articles also of Things for Enhanced User Expe- Biology to Learn Rapidly From Mis-
explore the effects of data, machine riences and Automation,” from takes Without Feeling Like a Fail-
learning, and GAI on health care Computer, examines use cases, ure,” from Computing in Science
and education, as well as efforts to challenges, and future directions & Engineering, shows how neu-
improve learning. for ChatGPT’s integration within rochemicals in the brain impact
IT and OT are converging to the IoT. learning and explains how to lever-
jump-start advancements in the Evolving technology has the age biology to improve education
IoT. The author of Computer article, potential to both hurt and harm and problem solving.

2469-7087/25 © 2025 IEEE Published by the IEEE Computer Society March 2025 7
 EDITOR: Joanna F. DeFranco, The Pennsylvania State University, [email protected]

This article originally

appeared in
DEPARTMENT: INTERNET OF THINGS
vol. 56, no. 8, 2023

Building the Factory of the

Future With the Industrial
Internet of Things
Jeff Daniels , Lockheed Martin

The future intelligent factory is centered on information technology and operational

technology (IT/OT) convergence. The IT, OT, engineering and operations skills,
data, and process needed to transform operations are discussed.

T
he convergence of information technology and already uses an event-driven architecture to improve
operational technology (IT/OT) systems has experiences. Hitachi Vantara, the maker of industrial
combined to enable capabilities not available machines, cites a case study where an amusement
in traditional organizational and systematic silos. The park is using its products for predictive maintenance
Industrial Internet of Things (IIoT) is a specific applica- on several rides in their parks across the globe.3 Kone
tion of the IoT in manufacturing and heavy industries, Corporation has more than 160,000 elevators and
enabling the integration of data, communication, escalators using its KONE 24/7 Connected Services,
and automation technologies to optimize production which includes IIoT services.4
processes and improve efficiency. The IIoT allows con-
nectivity in industrial, manufacturing, production, and THE CASE FOR CONNECTING
field support and in harsh environments. This article MACHINES, DEVICES, ROBOTS,
reviews why we need a factory of the future, the tech- AND TOOLS
nology required, where to connect IIoT devices, and the Why would we connect factory machines, devices,
skills required to design, deploy, maintain, and sustain robots, and tools? Long before the IIoT, manufactur-
an integrated factory of the future while transforming ing execution systems, supply chain systems, and
and modernizing systems and practices. resource planning systems have been capturing, cre-
Business transformation through IT/OT conver- ating, and storing massive amounts of data. Many of
gence is occurring throughout global markets in aero- the data sources had specific and proprietary sche-
space, food services, and manufacturing. Lockheed mas, access controls, and processes to get to the
Martin is laying the groundwork for faster and more data. Analysts had to create custom widgets, merge
agile production operations with a new, cybersecure, data, and reformat, export/import, and wrangle the
and standards-based network that can automatically data into a usable format. In short, these systems were
predict maintenance needs, analyze production per- “islands” of information that had to be bridged by man-
formance, and monitor quality. The platform is called draulic work, informal processes, and customization.
the Intelligent Factory Framework.1 In food services, The factory of the future integrates these data
McDonald’s recently introduced an automated store in streams into a consistent standardized format to
Fort Worth, TX, USA, where food is ordered by a kiosk allow predictive, prescriptive, and in situ analysis. The
and delivered by a conveyor.2 The global franchise factory operation is illuminated, rather than isolated
on an “island” with no integration with other functions.
The visibility of systems, state of operations, plans,
Digital Object Identifier 10.1109/MC.2023.3260419 work instructions, quality, and availability of data have
Date of current version: 26 July 2023 become actionable and are used to make decisions.

8 March 2025 Published by the IEEE Computer Society  2469-7087/25 © 2025 IEEE
 Rather than responding to a maintenance action on
a machine, a change in supply chain logistics, or a
production schedule change, operators now have
the data to take action ahead of the event. Instead of
responding days after a supply chain interruption, we
can pivot manufacturing lines to use different materi-
als or sources to mitigate a stop-work situation. If a
job is running on a machine that may cause a quality
defect, in situ analysis will allow operators to take
action and avoid costly material scrap.
Kone manages hundreds of thousands of elevators FIGURE 1. The Spot Robot at Grand Prairie, TX, USA. (Source:
with millions of actions each day. By capturing sensor Courtesy of Lockheed Martin.)
data from these devices, Kone uses an IIoT data lake
and artificial intelligence/machine learning (AI/ML)
models to make decisions. Maciej Kranz, Kone’s chief Within the technology stack, there are key ele-
technology officer, shares4 ments around connected things, data analytics, AI,
cybersecurity, industry standards, and the cloud com-
“…through data-driven predictability and insights, puting platform.
we can reduce [elevator] repair call-outs by more
than 40%. We can also proactively identify up to 70%
of faults and reduce entrapments by up to 50%.” THE IIOT ALLOWS CONNECTIVITY
IN INDUSTRIAL, MANUFACTURING,
PRODUCTION, AND FIELD SUPPORT
TECHNOLOGY STACK AND IN HARSH ENVIRONMENTS.
The components in the technology stack must be iden-
tified with several considerations. There are many
choices in the IIoT landscape and some pitfalls. We Connected machines, devices,
have seen early entrants in the market shutter services, robots, and sensors
such as Google, and cease product development and These provide real-time data on the status and per-
support, while others, such as Amazon Web Services formance of machines, allowing for the monitoring
(AWS), have excelled and continue to release new fea- and optimization of production processes. Increas-
tures on a regular basis.5 Michael Mackenzie, the gen- ingly, we are seeing mobile machines such as auton-
eral manager of AWS IoT, recently shared a new product omous mobile robots and guided vehicles in fac-
release called TwinMaker, which allows customers to6 tories. Lockheed Martin recently introduced the
Boston Dynamics Spot Robot into its Grand Prairie,
“…derive previously unavailable insights about their TX, USA, factory (Figure 1). McDonald’s is testing a
operations that inform real-time improvements to conveyor system for food delivery at its test restau-
their buildings, factories, industrial equipment, and rant. The human–machine interface will be key to
production lines and make accurate predictions understanding where connected machines, tools,
about system behavior with minimal effort.” and systems are most effective: essentially, how

www.computer.org/computingedge 9
INTERNET OF THINGS

operators, engineers, and analysts are using the IIoT pipeline is the critical component to maintain the
to improve their operations. chain of custody and accuracy of the data from the
“shop to the top.”
Data Analytics and AI
The vast amounts of data generated by connected PROCESS IS THE KEY
devices and sensors should be analyzed using data Process harmonization is essential to the factory of
analytics to identify patterns and trends, leading to the future. As common processes are realized, collab-
improved efficiency and decision making. The AI/ML oration will occur. Quality expert Dr. W. Edwards Dem-
models must align with insights required for decision ing stressed that10,11,12
making. For example, discrete knowledge of the type
of manufacturing or maintenance action is required to “the vast majority, 85 to 94%, of the time, the
create models that are meaningful. answer is found in the processes or systems in place
are not up to the task of handling all the variations
Cybersecurity that exist in today’s business climate, and as a
Connecting systems securely is challenging given the result, customer expectations are not met.”
heterogeneous products in the industrial landscape.
A consistent integration methodology and cyberse- Designers are becoming increasingly confident in
curity practices in defining the best security archi- applying multiple technology advancements to solve
tecture that aligns with the Confidentiality, Integrity, volatile, uncertain, complex, and ambiguous chal-
Assurance (CIA) model are required. As IIoT technol- lenges. When we consider where we insert the factory
ogy architectures become more integrated, Ahmad et of the future and IIoT technologies, the process must
al.7 identify the need to “increase the involvement and determine where to connect, what to connect, and
advancement of cybersecurity education at all levels.” how to efficiently prioritize the most valuable systems
We must maintain a vigilant position with security and and platforms.
privacy risks. To help prioritize, a modern data- and analytics-
focused approach is Job Task Analysis (JTA), a compre-
Standards hensive description of the duties and responsibilities
Normalized data are essential to gain output. The of a profession, occupation, or specialty area.13 The
long-held principle of “garbage in, garbage out” is Institute for Operations Research and the Manage-
applicable. Many systems run MQTT for messaging ment Sciences, the leading international association
and time-series data structure for capturing teleme- for operations research and analytics professionals,
try and operational data. Lockheed Martin is advanc- has identified seven domains of JTA, which include
ing the digital twin through standards such as STEP business problem (question) framing, analytics prob-
AP242 (ISO 10303-242) (“Managed Model-Based 3D lem framing, data, methodology (approach) selection,
Engineering”). 8 For the cybersecurity foundation, model building, deployment, and model lifecycle
the National Institute of Standards and Technology deployment.
800-171 and Cybersecurity Maturity Model Certifica- Popa et al.14 found that an important aspect of the
tion standards, which govern information safeguards JTA model is the first three domains: business problem
and dissemination controls, are applicable for data (question) framing, analytics problem framing, and
movement, machine updates, and authorization/ data (Figure 2). Spending the majority of time framing
authentication. 9 the problem and identifying the data needs is key. For
example, by collecting IIoT data, what problem do we
Cloud Computing hope to solve? And what data are available, and where
The intelligent factory relies on the ability to quickly might we get the data? IIoT sensor data can seem
and easily access and analyze large amounts of data. infinite with tens of thousands of tags being sent with
Cloud computing provides the necessary infrastruc- each operation: temperature, axis, rotation, revolu-
ture and processing power to enable this. The data tion, duration, etc.

10 ComputingEdge March 2025

 INTERNET OF THINGS

In the 1993 book, Reengineering the Corporation,

Hammer and Champy 15 challenge organizations to
identify key processes and make them lean and effi-
cient as possible. Technological changes in the 19th
century around material and energy sources becom-
ing available helped fuel the Industrial Revolution with
process changes, including the factory model, assem-
bly lines, division of labor, and communications. In the
late 20th century, the Digital Revolution introduced
technological advancements in pervasive connectiv-
ity, digital systems, cloud computing, cybersecurity,
and data analytics. These advancements, including
the IIoT, give us the opportunity to redesign and har-
monize our business processes to achieve dramatic
improvements in new ways, affecting outcomes
related to performance, cost, quality, service, and FIGURE 2. The JTA model. (Source: Popa et al.14)
speed to market.

SKILLS TRANSFORMATION technology, and the application of problem-solving

Skills transformation will be needed to design, sup- through critical thinking. Organizations that embrace
port, and leverage data from the factory of the future. continuous learning and a vigorous program of edu-
IT/OT skills are centered on a full-stack model, includ- cation will lead future transformation and maintain
ing a number of proficiencies, technologies, and tech- industrial leadership.
niques. Creating a competency center of skills around
IT/OT convergence, automation, and DevOps is a stra-
tegic approach. Lockheed Martin created the Auto-
mation Center of Excellence, which includes the
T he IIoT introduces an inflection point in manu-
facturing and heavy industries, enabling the
integration of data, communication, and automation
automation services, development, robotic process technologies to optimize production processes and
automation, and workflow enablement, and its IT/OT improve efficiency. In Deming’s 14 Points on Qual-
convergence initiative, the Intelligent Factory (Figure ity Management10 or the Deming Model of Qual-
3). A curriculum based on industry practice and inter- ity Management, we are challenged to “improve
nal practice was curated and made available to all constantly and forever every process for planning,
employees on its Digital Academy
platform. The IIoT track includes
critical skills such as DevOps,
automation, industrial engineer-
ing, production operations, AI and
ML, and cloud computing.
Many universities, such as
the University of Central Florida,
Pennsylvania State University, the
University of Florida, and Tarleton
State University, are introducing
AI and ML concepts to their engi-
neering and computer science
curriculum. Graduates are learning FIGURE 3. Lockheed Martin’s Intelligent Factory Team working at a production
critical technical skills, emerging facility.

www.computer.org/computingedge 11
INTERNET OF THINGS

production and service.” The IIoT allows us to reengi- 6. R. Pell. “Digital twins ‘made easy’ with new AWS IoT
neer and reimagine how we design, deploy, and man- service.” EE News 2.0. Accessed: Jan. 24, 2023. [Online].
age operations. https://www.eenewseurope.com/en/digital-twins
The future factory concept identifies why connec- -made-easy-with-new-aws-iot-service/
tivity is critical as well as securing and governing the 7. N. Ahmad, P. A. Laplante, J. F. DeFranco, and M.
operational data. The technology stack may vary, but Kassab, “A cybersecurity educated community,”
the core components include machines, tools, robots, IEEE Trans. Emerg. Topics Comput., vol. 10, no. 3, pp.
sensors, data analytics and AI/ML, cybersecurity, 1456–1463, Jul./Sep. 2022, doi: 10.1109/TETC.2021
standards, and cloud computing. While no single tech- .3093444.
nology will solve today’s volatile, uncertain, complex, 8. “STEP AP242 (ISO 10303-242) ‘Managed model-based
and ambiguous challenges, the ability to adapt and 3D engineering.’” Accessed: Jan. 24, 2023. [Online].
adjust will keep organizations ahead of ready. Available: http://www.ap242.org/
Leading researchers and practitioners in the field 9. “Protecting controlled unclassified information in
realize that professionalism and ethics are keys to nonfederal information systems and organizations,”
ensuring safety, security, and responsibility in the NIST, Gaithersburg, MD, USA, SP 800-171 Rev. 1, 2016.
solutions and systems. It is our duty to safeguard our [Online]. Available: https://csrc.nist.gov/publications
communities and future factories while providing /detail/sp/800-171/rev-1/final
guidance for the application of emerging technologies 10. “W. Edwards Deming’s 14 points for total quality
and platforms. management,” American Society of Quality, Milwaukee,
WI, USA, 1982. [Online]. Available: https://asq.org
REFERENCES /quality-resources/total-quality-management/deming
1. L. Martin. “New intelligent factory network comes -points
to life.” Accessed: Jan. 24, 2023. [Online]. Available: 11. W. Deming, Out of the Crisis, 1st ed. Cambridge, MA,
https://www.lockheedmartin.com/en-us/news USA: MIT Press, 1986.
/features/2020/new-intelligent-factory-network 12. R. D. Snee. “Management holds the key to continued
-comes-to-life.html process verification.” Pharma Manufacturing.
2. “5 things to know about McDonald’s new test restau- Accessed: Jan. 24, 2023. [Online]. Available: https://
rant in Texas.” McDonald Corporation. Accessed: Jan. www.pharmamanufacturing.com/compliance
24, 2023. [Online]. Available: https://corporate /compliance-management/article/11321722
.mcdonalds.com/corpmcd/our-stories/article /management-holds-the-key-to-continued-process
/texas-order-ahead-lane.html -verification
3. A. Brust. “Hitachi Vantara raises data integration, 13. “INFORMS job task analysis.” INFORMS. Accessed: Jan.
catalog, IIoT stakes.” The New Stack. Accessed: Jan. 24, 2023. [Online]. Available: https://info.informs.org
24, 2023. [Online]. Available: https://thenewstack.io /jta
/hitachi-vantara-raises-data-integration-catalog-iiot 14. A. Popa, B. Amaba, and J. Daniels, “A framework of
-stakes/ best practices for delivering successful artificial
4. “KONE works with AWS to boost the power and scale intelligence projects. A case study demonstration,”
of its digital services for smart buildings.” KONE. presented at the SPE Annu. Tech. Conf. Exhib., Dubai,
Accessed: Jan. 24, 2023. [Online]. Available: https:// UAE, Sep. 2021, doi: 10.2118/206014-MS.
www.kone.com/en/news-and-insights/releases/kone 15. M. Hammer and J. Champy, Reengineering the Corpora-
-works-with-aws-to-boost-the-power-and-scale-of-its tion: Manifesto for Business Revolution. New York, NY,
-digital-services-for-smart-buildings-2021-12-09.aspx USA: Harper Collins, 1993.
5. R. Miller. “Google Cloud will shutter its IoT Core service
next year.” Tech Crunch. Accessed: Jan. 24, 2023.
[Online]. Available: https://techcrunch.com/2022 JEFF DANIELS is the director of the Automation Center of
/08/17/google-cloud-will-shutter-its-iot-core-service- Excellence at Lockheed Martin, Fort Worth, TX 76107 USA.
next-year/ Contact him at [email protected].

12 ComputingEdge March 2025

 IEEE CS EXECUTIVE COMMITTEE
President: Hironori Washizaki
www.computer.org President-Elect: Grace A. Lewis
Past President: Jyotika Athavale
Vice President: Nils Aschenbruck
PURPOSE: Engaging professionals from all areas of computing,
Secretary: Yoshiko Yasuda
the IEEE Computer Society sets the standard for education
and engagement that fuels global technological advancement. Treasurer: Darren Galpin
Through conferences, publications, and programs, IEEE CS VP, Member & Geographic Activities: Andrew Seely
empowers, guides, and shapes the future of its members, and VP, Professional & Educational Activities: Cyril Onwubiko
the greater industry, enabling new opportunities to better serve VP, Publications: Charles (Chuck) Hansen
our world. VP, Standards Activities: Edward Au
OMBUDSMAN: Contact [email protected]. VP, Technical & Conference Activities: Terry Benzel
2025–2026 IEEE Division VIII Director: Cecilia Metra
CHAPTERS: Regular and student chapters worldwide provide
2024–2025 IEEE Division V Director: Christina M. Schober
the opportunity to interact with colleagues, hear technical
experts, and serve the local professional community. 2025 IEEE Division V Director-Elect: Leila De Floriani

PUBLICATIONS AND ACTIVITIES IEEE CS BOARD OF GOVERNORS

Computer : The flagship publication of the IEEE Computer Term Expiring 2025:
Society, Computer, publishes peer-reviewed technical content that İlkay Altintaş, Mike Hinchey, Joaquim Jorge, Rick Kazman,
covers all aspects of computer science, computer engineering, Carolyn McGregor, Andrew Seely
technology, and applications. Term Expiring 2026:
Periodicals: The IEEE CS publishes 12 magazines, 18 journals Megha Ben, Terry Benzel, Mrinal Karvir, Andreas Reinhardt,
Deborah Silver, Yoshiko Yasuda
Conference Proceedings & Books: Conference Publishing
Services publishes more than 275 titles every year. Term Expiring 2027:
Standards Working Groups: More than 150 groups produce Sven Dickinson, Alfredo Goldman, Daniel S. Katz, Yuhong Liu,
IEEE standards used throughout the world. Ladan Tahvildari, Damla Turgut

Technical Communities: TCs provide professional interaction

in more than 30 technical areas and directly influence computer IEEE EXECUTIVE STAFF
engineering conferences and publications. Executive Director and COO: Sophia Muirhead
Conferences/Education: The IEEE CS holds more than 215 General Counsel and Chief Compliance Officer:
conferences each year and sponsors many educational activities, Ahsaki Benion
including computing science accreditation. Chief Human Resources Officer: Cheri N. Collins Wideman
Certifications: The IEEE CS offers three software developer Managing Director, IEEE-USA: Russell Harrison
credentials. Chief Marketing Officer: Karen L. Hawkins
Managing Director, Publications: Steven Heffner
AVAILABLE INFORMATION Staff Executive, Corporate Activities: Donna Hourican
To check membership status, report an address change, or obtain Managing Director, Member and Geographic Activities:
information, contact [email protected]. Cecelia Jankowski
Chief of Staff to the Executive Director: Kelly Lorne
IEEE COMPUTER SOCIETY OFFICES Managing Director, Educational Activities: Jamie Moesch
IEEE Standards Association Managing Director: Alpesh Shah
WASHINGTON, D.C.: LOS ALAMITOS: Chief Financial Officer: Thomas Siegert
2001 L St., Ste. 700, 10662 Los Vaqueros Cir.,
Chief Information Digital Officer: Jeff Strohschein
Washington, D.C. 20036-4928 Los Alamitos, CA 90720
Managing Director, Conferences, Events, and Experiences:
Phone: +1 202 371 0101 Phone: +1 714 821 8380 Marie Hunter
Fax: +1 202 728 9614 Email: [email protected] Managing Director, Technical Activities: Mojdeh Bahar
Email: [email protected]
IEEE OFFICERS
IEEE CS EXECUTIVE STAFF
President & CEO: Kathleen A. Kramer
Executive Director: Melissa Russell President-Elect: Mary Ellen Randall
Director, Governance & Associate Executive Director: Past President: Thomas M. Coughlin
Anne Marie Kelly
Director & Secretary: Forrest D. Wright
Director, Conference Operations: Silvia Ceballos
Director & Treasurer: Gerardo Barbosa
Director, Information Technology & Services: Sumit Kacker
Director & VP, Publication Services & Products: W. Clem Karl
Director, Marketing & Sales: Michelle Tubb
Director & VP, Educational Activities: Timothy P. Kurzweg
Director, Membership Development: Eric Berkowitz
Director & VP, Membership and Geographic Activities:
Director, Periodicals & Special Projects: Robin Baldwin Antonio Luque
Director & President, Standards Association:
Gary R. Hoffman
Director & VP, Technical Activities: Dalma Novak
Director & President, IEEE-USA: Timothy T. Lee

revised 14 January 2025

 EDITOR: Christof Ebert, Vector Consulting Services, [email protected]

This article originally

appeared in
DEPARTMENT: INDUSTRY INSIGHTS
vol. 57, no. 8, 2024

Convergence of IT, OT, and

the IoT
Christof Ebert , Vector Consulting

FROM THE EDITOR

The IT-driven convergence of previously separated domains is today a megatrend that even defines new
industries. The merge of enterprise IT, operational technology, and product IT is creating business mod-
els, such as on-demand services for ecologic mobility or artificial intelligence (AI)-based predictive main-
tenance. Risks grow at the same time, such as cybersecurity and the availability of cloud infrastructure.
Companies and competences must keep pace. —Christof Ebert

IT is evolving new business models and ecosystems, such as on-demand ecologic mobility
and artificial intelligence-based diagnostics. Devices are connected by the Internet of
Things toward flexible services. Companies must converge competences to keep pace.

P
reviously isolated systems are increasingly disappeared. Enterprise IT is converging with opera-
connected and allow flexible on-demand ser- tional IT and embedded systems such as the Internet
vices. Performance is optimized by balancing of Things (IoT).1,2 At the same time, embedded indus-
intelligence and access to resources. Operational tries have evolved with cloud solutions and dynamic
technology (OT) and product IT become smart by over-the-air (OTA) SW updates.3
the cloud infrastructure and becoming aware of their The impacts depend on the existing infrastructure
respective environments. Enterprise IT connects and and products but will typically include the following:
integrates with distributed devices, thus creating
ubiquitous computing solutions for even unforeseen › Enterprise IT: This requires the integration of
situations and environments. Convergence increases business processes, workflows, and IT systems
complexity and security risks, and not the least, it also toward efficiency and cost savings in infrastruc-
demands new competences. ture and maintenance with less overhead and
fewer overlaps of traditionally separated and
CONVERGENCE replicated systems and organizations. An example
The convergence of various IT systems has become is the collaboration of IT departments with R&D.
a megatrend across industries. With the seamless › OT: This requires the usage of existing IT
connectivity of devices, edge nodes, sensors, actu- capabilities like real-time data analytics and AI
ators and a wealth of distributed software stacks, to optimize industrial processes, reduce down-
the historic divide of IT versus embedded SW has time, and improve efficient resource utilization
and energy management.
› Embedded SW: The integration of embedded sys-
Digital Object Identifier 10.1109/MC.2024.3407248 tems with IT infrastructure allows fast SW updates,
Date of current version: 26 July 2024 an aligned SW bill of materials, organizational

14 March 2025 Published by the IEEE Computer Society  2469-7087/25 © 2025 IEEE
 FIGURE 1. IT convergence has many impacts. HW: hardware; VUCA: volatile, uncertain, complex, ambiguous; GAI: generative
artificial intelligence; SOA: service-oriented architecture; MBSE: model-based systems engineering; TDD: test-driven develop-
ment; TDRE: test-driven requirements engineering; ALM/PLM: application/product lifecycle management.

cybersecurity strategies, enabling or disabling › Enhanced efficiency: Integration allows the

features, predictive maintenance, advanced real-time monitoring and control of production
diagnostics, and enhanced user interfaces. lines, leading to optimized operations and
reduced downtime.
Figure 1 shows the impacts of convergence on › Ecology: The ecologic footprint of systems is
companies along several dimensions from business to reduced by extending the lifetime of hardware
technology. Effects depend on how such systems are capable of performing new functions by mere
implemented and deployed. A nanodevice implanted SW updates.
in a human has different physical, quality, energy, and › Predictive maintenance: Data analytics can
IT requirements from a server farm used for a pro- predict equipment failures before they occur,
duction plant. Shared SW and information make the reducing maintenance costs and increasing
previous tangible boundaries between the embedded equipment lifespan.
office and back office diminish. › Quality control: Improved data sharing can
Connecting embedded intelligence with cloud- enhance product quality by monitoring and
based backbones, such as manufacturing robots, med- adjusting processes in real time.
ical surgery systems, or autonomous drones, brings › Operational efficiency: Streamlining administra-
many advantages: tive, supply chain, and production workflows

www.computer.org/computingedge 15
INDUSTRY INSIGHTS

reduces errors and enhances focus on value integration problems of systems, and the sudden dis-
instead of frictions of IT systems. rupt of services due to connectivity outages. Take the
› Regulatory compliance: Embedded SWs and their frequent standstills of automated cars, which already
IT systems can be continuously monitored and make the residents of cities such as San Francisco
updated, ensuring compliance and performance. wonder how that would scale in real emergency situ-
› Utilities: Integration supports the development ations when firefighters and ambulances get blocked
of smart grids that manage energy distribution by such cars.4
efficiently, reducing losses and improving In turn, the market for SW engineers is changing.
reliability. Portals such as https://layoffs.fyi/ show an increas-
ing number of traditional developers being laid off,
while new competences are sought. Tech giants
THERE IS A LOT TO LEARN FROM such as Google, Facebook, and Amazon have started
THOSE COMPANIES THAT ALREADY and companies in traditional sectors follow. With
INTEGRATED THEIR PRODUCT IT WITH generative AI (GAI) being deployed for development
OT AND ENTERPRISE IT. tasks, demands are quickly moving to integration and
systems competences.
IT convergence arrives at different speeds across
› Autonomous vehicles: The integration of IT, OT, industries. The telecommunication industry was
and embedded systems is crucial for the develop- among the first to learn it the hard way.3 With the
ment and deployment of autonomous vehicles. advent of converging IT services, the whole industry
was eaten by IT, with prominent examples such as Alca-
We take it for granted that our smartphones are tel, Lucent, and Motorola. Others, like Cisco, Nokia, and
always connected and always on. Similarly, industrial Huawei, reinvented themselves and transformed into
plants, vehicles, consumer goods, and even medical convergence companies that master IT from edge to
implants rely on converging IT services that boost enterprise. The same will happen to transport, mobility,
their performance and allow operations in new envi- and many other industries if they don’t transform. What
ronments with constraints not foreseeable during can we learn? Incumbents must leave old behaviors
initial system design. and become SW companies with converging compe-
tences and a fully new SW-driven culture.
CHALLENGES
The most successful companies are those that no lon- INSIGHTS AND EXAMPLES
ger distinguish between IT and business. But it is a big There is a lot to learn from those companies that
step from demo to reality because convergence needs already integrated their product IT with OT and enter-
new skill sets that we just don’t have. In our current prise IT. Many companies had segregated their IT
industry survey, we found that just one-third of com- departments and product development ever since
panies seamlessly integrate product IT and enterprise they started with SW. Now, they must merge these
IT. Two-thirds are struggling or have not even started. teams and integrate distributed product IT and OT to
Seamless convergence with the high availability of benefit from enterprise cloud services and ubiquitous
cloud services end to end is still a big step to go. Exam- SW updates. This transformation from hardware to
ples include our many mobile devices, which need SW and systems thinking is difficult to implement. Car
frequent reboots, battery replacements, and firmware manufacturers such as BMW, Ford, GM, Hyundai, Mer-
updates. The same goes for IT systems, which often cedes, and Volkswagen started that journey a while
have organically grown interfaces and gaps that slow ago. Thousands of SW engineers and computer scien-
down workflows and need lots of overhead for opera- tists are working on developing operating systems and
tion and maintenance. control units for the customer journey. Challenges of
New challenges appear, as we observe with expo- converging IT/OT/IoT systems are higher due to func-
nentially growing cybersecurity attacks, the increasing tional safety demands.

16 ComputingEdge March 2025

 INDUSTRY INSIGHTS

FIGURE 2. Value provisioning with IT convergence.

Health care and underlying medical devices have For implementation, consider robust reference
started very early with the convergence of medical architectures that allow adaptive behaviors without
device SW and hospital IT systems. Companies such hardware impacts.2,3 Figure 2 shows a typical three-
as Siemens Healthineers underline the need for tier architecture and how convergence is achieved.
convergence in the medical sector. An example is The left side shows the three-tier IoT reference archi-
the combination of the creation of a medical image, tecture, which we see in most converging systems.
for example, MRI scanners providing image analysis
and AI diagnostics. A digital twin of humans even
converges biology with IT, allowing one to test treat- FOR IMPLEMENTATION, CONSIDER
ments and medications and simulate dependencies. ROBUST REFERENCE ARCHITECTURES
Like any other digital twin, it allows one to model THAT ALLOW ADAPTIVE BEHAVIORS
and simulate “in vitro” before going to real life. In WITHOUT HARDWARE IMPACTS.
our current trends survey, only 30% of companies
deploy AI-based solutions and workflows. It is still a
big step to move forward and benefit from technol- These three tiers have quickly become a reference
ogy innovation. as they tie into layered IT structures and allow one to
Many public companies and utilities have under- abstract the cloud, performance, and devices. Like
stood that even without competition, they must invest the seven Open Systems Integration networking tiers,
in IT and digital transformation to better serve cus- they help to efficiently design modular open systems.
tomers—and taxpayers. Public transport shows how
converging vehicle IT and enterprise IT create value. WHERE DO YOU GO FROM HERE?
The goals are efficiency with new customer services IT convergence is far from trivial and means compe-
and mobility performance with improved vehicle oper- tence growth, transformation, and culture change. A
ations. The convergence of distributed IT in vehicles lot can be learned from industry leaders on their trans-
with the backbone and IT infrastructure is critical for formations toward the convergence of enterprise IT,
successful operations. Business continuity is critical OT, and embedded product SW. Here are some take-
for such systems. aways from our industry insights.

www.computer.org/computingedge 17
INDUSTRY INSIGHTS

for robust and mature processes, such as those

FREE IEEE WEBINAR demanded by ISO 33000.

ON CURRENT › Architecture and technology: Invest in robust

INDUSTRY TRENDS architectures of SW-defined systems. Systemat-

ically reduce legacy with overly high complexity
and technical debt. Establish technology road
What are the current industry trends? To get indus-
maps with positive content—that is, what inno-
try insights, we continuously survey practitioners on
vations and new systems to introduce—and
their challenges and resolutions. This webinar of the
negative content, for example, which portfolio
IEEE Computer Society provides industry insights.
elements and systems to terminate.
Venue: Free online interactive webinar on 11 Sep-
tember 2024 at 12p.m. Eastern Daylight Time and
› Cybersecurity: Develop a comprehensive
security strategy that addresses vulner-
6 p.m. Central European Summer Time at https://
abilities across IT, OT, and embedded
event.on24.com/wcc/r/4605968/31B1BD3C05DBF
systems. Ensure continuous monitoring and
2114BAD1EC3C30C9967?partnerref=computer.
incident response strategies across these
often-separated domains to quickly detect
vulnerabilities, communicate attacks, and
resolve problems.
› Business: Change business models to accommo- › Change: Grow consolidated change initiatives
date heterogeneous and fast-changing market with clear business targets and management
demands with shrinking target segments. The sponsorship. Engage stakeholders across
leaders of tomorrow must continuously shape traditional organization silos to connect IT
their markets with digital technologies while competences with business needs.
creating resilient internal workflows.
› Culture: Foster a culture that embraces digital Education programs must change at the same
transformation and continuous improvement. pace (see “Free IEEE Webinar on Current Industry
Stimulate collaboration between IT and OT Trends” for information on current trends). Industry
teams to ensure integration and address is looking for graduates who master both the SW with
technology and process frictions. algorithms and underlying theoretical foundations as
› Competences: Fertilize skills that bridge IT, OT, well as systematic engineering and systems under-
and embedded systems, understanding the standing from domains as diverse as automotive,
unique requirements and technologies of each chemistry, medicine, and transport. Engineers with
domain. Stimulate classic enterprise IT experts the right competence mix, enterprise, and product
to learn about embedded SW, and stimulate and pairing of hard and soft skills are in demand across
engineers to grow their IT competences. Estab- domains, from IT to the IoT, from app development to
lish profound systems thinking and understand- big systems, from automotive to medical.
ing of application domains.
› Regulation: Ensure compliance with
industry-specific regulations and standards,
particularly in highly regulated sectors like
A s uncertainty will further grow, mastering con-
vergence defines future leaders. Albert Einstein,
Physics Nobel prize laureate advised: “We can’t solve
health care and energy. Conduct audits and problems by using the same kind of thinking we used
assessments to ensure ongoing compliance and when we created them.” We will not solve today’s SW
identify areas for improvement. challenges, such as cybersecurity and AI-driven auto-
› Processes: Implement state-of-the-art SW mation, with the domain-centric IT and engineering
development processes supporting automated of yesterday. Future success demands converging
and reliable closed-loop DevSecOps develop- systems that not only deliver features but also trust
ment. With all agility, do not neglect the need and availability.

18 ComputingEdge March 2025

 INDUSTRY INSIGHTS

REFERENCES CHRISTOF EBERT is the managing director of Vector Con-

1. “Converge IT and OT to turbocharge business opera- sulting Services, 70499 Stuttgart, Germany. Contact him at
tions’ scaling power,” McKinsey, New York, NY, USA, https://www.linkedin.com/in/christofebert/ or christof.ebert
Jul. 2022. Accessed: May 24, 2024. [Online]. Available: @vector.com.
https://www.mckinsey.com/capabilities/operations
/our-insights/converge-it-and-ot-to-turbocharge
-business-operations-scaling-power
2. C. Ebert and J. Salecker, “Embedded software tech-
nologies and trends,” IEEE Softw., vol. 26, no. 3, pp.
14–18, May/Jun. 2009, doi: 10.1109/MS.2009.70.
3. G. Doukidis, D. Spinellis, and C. Ebert, “Digital transfor-
mation: A primer for practitioners,” IEEE Softw., vol. 37,
no. 5, pp. 13–21, Sep. 2020, doi: 10.1109/MS.2020
.2999969.
4. C. Contreras, “Self-driving auto companies have made
advancements, but technology is still lacking,” Tech
F O LLOW US
Xplore, Houston, TX, USA, May 2024. Accessed: May 24, @s e curit y p riva c y
2024. [Online]. Available: https://techxplore.com/news
/2024-05-auto-companies-advancements-technology
-lacking.html

IEEE COMPUTER SOCIETY

Ensure your research is

Call for Papers easily discoverable by

being indexed in major
databases and optimized
for search engines.

GET PUBLISHED
www.computer.org/cfp

general-cfp-hHoriz-Sept24.indd 4 9/30/24 6:22 PM

www.computer.org/computingedge 19
 EDITOR: Amir
EDITOR: Amir Dabirian,
Dabirian, [email protected]
[email protected]
This article originally
appeared in
DEPARTMENT: IT
DEPARTMENT: IT TRENDS
vol. 25, no. 6, 2023

Nothing
Nothing Is
Is Harder
Harder toto Resist
Resist Than
Than the
the Temptation
Temptation of AIof AI
AndrewPark
Andrew Park , ,University
Universityof
ofVictoria,
Victoria,Victoria,
Victoria,BC,
BC,V8P
V8P5C2,
5C2,Canada
Canada
Jan Kietzmann , ,Gustavson
JanKietzmann GustavsonSchool
SchoolofofBusiness,
Business,University
University of
ofVictoria,
Victoria, Victoria,
Victoria, BC,
BC, V8P
V8P5C2,
5C2,Canada;
Canada; and
and
UC UC Business
Business School,
School, University
University of Canterbury,
of Canterbury, Christchurch,
Christchurch, 8041,8041,
NewNew Zealand
Zealand
Jayson
JaysonKilloran
Killoran , ,Queen’s
Queen’sUniversity,
University,Kingston,
Kingston,ON,
ON,K7L
K7L3N6,
3N6,Canada
Canada
Yuanyuan
YuanyuanCui,
Cui,Auckland
AucklandUniversity
University of
ofTechnology,
Technology,Auckland,
Auckland,1010,
1010,New
NewZealand
Zealand
Patrick
Patrickvan Esch , Kennesaw
vanEsch , KennesawState
StateUniversity,
University,Kennesaw,
Kennesaw,GA,
GA,30144,
30144,USA
USA
Amir Dabirian , ,California
AmirDabirian CaliforniaState
StateUniversity,
University,Fullerton,
Fullerton,CA,
CA,92831,
92831,USA
USA

The use of generative AI has become increasingly prevalent in the business world. With
The use of generative AI has become increasingly prevalent in the business world.
the ability to create original content and automate certain tasks, businesses have been
With the ability to create original content and automate certain tasks, businesses
quick to adopt this technology. However, as with any emerging technology, there are
have been quick to adopt this technology. However, as with any emerging technology,
potential pitfalls to be aware of. This article aims to review the current state of
there are potential pitfalls to be aware of. This article aims to review the current
generative
state AI in business
of generative and highlight
AI in business some of the
and highlight somepotential risks associated
of the potential with
risks associated
its use. Specifically, we examine issues such as pausing giant AI experiments,
with its use. Specifically, we examine issues such as pausing giant AI experiments,
misinformation, data
misinformation, data accuracy,
accuracy, process
process automation,
automation,shift
shiftof
ofpower,
power,control
controlofof
civilization, organizational
civilization, organizational security,
security,and
andthe
thepotential
potentialfor
forAI-generated
AI-generatedcontent
contenttoto
deceive individuals.
deceive individuals. By bringing
bringing these concerns to the forefront, we hopetotoencourage
these concerns to the forefront, we hope encourage
aamore
morethoughtful
thoughtful and cautious approach
approachto tothe
theuse
useofofgenerative
generativeAIAIininbusiness.
business.

O
scar Wilde famously said that he can resist Despite the excitement about the positive impact
everything except temptation. With signifi- of AI, most of these developments are highly polarizing.
cant advances in AI and machine learning This is not new since technological advances have
(ML), more and more companies find the temptation of always created a rift in society. Melville Krantzberg,
embedding AI within their current offerings too entic- who was a history professor devoted to studying the
ing to resist. Luxury car maker Mercedes-Benz, for history of technology, reminds us that the effects of
instance, integrated ChatGPT in its cars in the United technology are never only positive or negative but
States in the summer of 2023.1 Real estate companies always both—and never neutral.4 Kranzberg further
have just released “AI Realtor Assistants” to allow explains this qualification by arguing that technology’s
buyers to ask questions about properties for sale,2 and interaction with society is such that its consequences
medical chatbots help doctors compose letters to for humans and society often go far beyond the
appeal insurance decisions, draft e-mails to patients, intended purposes of the technology. Importantly, he
provide health insights, and create generate handouts further notes that the effects of a particular technology
for patients.3 With most health practitioners complaining can be very different depending on the contexts into
that they do not have enough patient interactions (and which it is introduced and that the valence of these
vice versa), who would want to resist this development? effects can change over time.
Traditionally, engineers and marketeers have been
1520-9202 © 2023 IEEE
the ones promoting the deployment of new technolo-
Digital Object Identifier 10.1109/MITP.2023.3340529 gies, while sociologists, organizational behavior experts
Date of current version 12 January 2024. and consumer psychologists have favored a slower

20 March 20252023
November/December Published
Published by theby theComputer
IEEE IEEE Computer Society 
Society 13
2469-7087/25 © 2025 IEEE
IT Professional
IT TRENDS IT TRENDS

approach that would take into consideration the longer Organizational security: To generate bespoke,
term organizational and social implications of these organizational content, organizations may inad-
developments. Interestingly, this is now changing, with vertently reveal sensitive information as ML
many Silicon Valley technologists sounding alarming inputs to AI tools.
voices over an approach they believe is too rushed
In this article, we discuss these near-term risks with
when it comes to developing and deploying new
several examples, including 1) the case of AI-based
AI-based products. Until recently, one would not neces-
transcription services obviating manual audio transcribers,
sarily expect AI luminaries like Yoshua Bengio, Elon Musk,
many of whom reside in the developing world, in favor
Steve Wozniak, and politician Andrew Yang to focus on
of transcription results subject to their own interpreta-
identifying and reducing the potential harms of AI. How-
tions and 2) the case of the AI-based code completion
ever, these innovators have just signed an open letter,
tool GitHub Copilot causing concerns that it was
published by the Future of Life Institute,5 calling for a mor-
trained on a large volume of copyrighted code from
atorium on the further development of AI systems more
developers around the world without their permission.
powerful than GPT-4, the foundational technology behind
ChatGPT. The letter, echoing futurologist and innovator
NEAR-TERM ORGANIZATIONAL
Roy Amara’s warning that we tend to overestimate the
RISK VERSUS LONG-TERM
effect of a technology in the short run and underestimate
EXISTENTIAL RISK OF AI
the effect in the long run, argues that such modern AI sys-
In this section, we delineate the actual near-term risks
tems pose three critical existential risks:
for each of the three long-term existential risks pro-
Misinformation: Machines will fill our informa- posed in the open letter. We argue that by ignoring the
tion channels with mis- and disinformation. near-term risks in favor of the futuristic ones listed in
Automation: All jobs will be automated. the letter, organizations can suffer operational and rep-
Control of civilization: Machines will outsmart utational harm.
humans and render us obsolete, leading to a
loss of control of our civilization. Misinformation Versus Data Inaccuracy
We are amidst a fundamental shift in software develop-
We agree that these worries are well founded and ment—a shift from the traditional logic-based approach
concerning from an organizational and social perspec- where software is developed based on a set of rules to
tive. However, most organizational decision makers are example-based software development. Today, develop-
driven by short-term deliverables and are less worried ers rely on large datasets to recognize patterns and
about the long-term impact of their decisions. For this make decisions based on those patterns. For example,
reason, in this article, we focus on the near-term risks in computer vision and image recognition, developers
that correspond to the long-term risks identified. might use examples of images to train the software to
We focus on risks present in AI systems that can and recognize certain patterns. The software can then use
have begun to negatively impact organizational and those patterns to recognize similar images in the
societal outcomes. Some of these near-term risks future.
have been identified more generally by other research- This shift from logic-based to example-based soft-
ers (e.g., Bender et al.,6) and we draw upon Kapoor and ware development has been driven by advances in ML
Narayanan’s related work7 on this topic. However, little and AI. The main advantage is that by using examples
research has been conducted to understand what to train software, developers can create more accurate
these risks mean specifically to organizations and how and effective software that can learn and adapt over
executives can mitigate them. The corresponding near- time. Applications can be found in image and speech
term risks to organizations that are related to these recognition, language tools, autonomous vehicles, behav-
points are as follows: ioral biometrics, and more.8
As the open letter suggests, because AI systems
Data inaccuracy: Improperly or inaccurately are trained on such large swaths of sample data gener-
trained deep learning models can lead to errors ated by people (e.g., text on the Internet), the resultant
that harm organizations. outputs are essentially indistinguishable from human
Power shift: Advanced AI tools shift leverage output. When we move beyond text and large language
away from content creators (e.g., employees models like GPT, generative AI can also create other
and contract workers) and toward the creators types of content. Deepfakes or generative adversarial
of those AI tools. networks (GANs), for instance, make it impossible to

14
www.computer.org/computingedge
IT Professional November/December 2023 21
IT TRENDS IT TRENDS

tell apart genuine content from synthetic material.9,10 its customer Roberto Mata, who claimed personal
Synthesia generates humanlike AI avatars, ElevenLabs injury on a flight; Mata’s lawyer submitted a court filing
is a voice cloning software, and Dalle-2 and Midjourney that referenced six nonexistent precedents to argue in
create images, to name just a few. Because of this, the favor of Mata. The court observed that, in addition to
letter opines that there is the potential for malicious these bogus precedents, the legal filing was littered
actors to spread propaganda and harmful disinformation with made-up quotes and citations. The legal team,
more easily and automatically (e.g., that might impact including head counsel Steven Schwartz, eventually
elections). admitted to using ChatGPT to aid in writing the filing
A big near-term issue, in our view, results from the but insisted the mistakes arose from ignorance of the
black-boxing of AI technologies. In other words, the capabilities of the tool.13 Such are the near-term risks
lack of transparency and resulting intractability in of emerging deep learning tools like large language
explaining deep learning models (e.g., Singh et al.11) models; even without any malicious intent or interven-
makes is virtually impossible to understand how AI tion from bad actors, organizations that blindly rely on
models have created content. If we were to open this AI platforms may suffer detrimental outcomes due to a
black box, we would first see that deep learning models poor understanding of the platform’s capabilities.
are optimized on outcomes based on very large data- Managers who develop products using APIs from
sets. This is not surprising, nor is the insight that this OpenAI or other transformer-based models, such as
optimization occurs through a combination of manual Google’s Bard or Meta’s LLaMa, can benefit by sug-
and automated adjustments of yet another large set gesting specific use cases up front to avoid unrealistic
of parameters.12 These parameters are then tuned expectations of what their products can do. For exam-
through some feedback mechanism, where the output ple, it has been suggested that ChatGPT is garnering
of the model is measured against a desired or ideal out- interest from novice investors to build a high-return
put, and the model weights are continually adjusted equity portfolio, but, when compared against existing
until an acceptable error level is achieved. Together, popular exchange-traded funds and mutual funds,
these elements suggest that the content generated
those suggested by ChatGPT underperformed. How-
by ready-to-use ML products, either used directly or
ever, ChatGPT was effective in providing basic finan-
through application programming interfaces (APIs)
cial literacy to amateur investors, who could then
cannot be audited or reverse engineered so that a
parlay that knowledge into developing their own intu-
third-party inspector or even regulator can reasonably
itions and approaching other experts about ways to
understand the characteristics of the parameters, model
structure their portfolios.14 As such, managers who
weights, and data that comprise the trained model. Con-
work in the financial sector would be prudent to focus
sequently, we do not know the algorithms or statistical
their marketing efforts on encouraging the use of their
models used, how learning has changed parameters
novel AI-based products in the latter case.
throughout time, or what a model looks like for a certain
prediction(s).
This lack of interpretability, or explainability, of the Process Automation Versus
model can easily cause an organization that uses the Power Shift
respective AI tool to develop an overreliance on it with- The open letter poses a broad, rhetorical question
out understanding how its inputs are converted to the about whether we should let AI platforms automate
outputs suggested by the model or how the system away all of our jobs. Of course, this fear is as old as
identified and corrected errors or biases in those technology itself. Otis, still one of the major elevator
decisions. Among the more well-known examples is companies in the world, first started to install commer-
Amazon’s AI recruiting tool. The tool was designed to cial elevators in the 1850s. These elevators were oper-
review resumes and identify the most qualified candi- ated by fully trained operators until a century later. In
dates. However, because Amazon had traditionally 1945, some elevator operators, porters, and mainte-
predominantly hired men, the tool developed a bias nance workers went on strike in New York City commer-
against women and learned to penalize resumes and cial buildings. Otis Elevator consequently developed and
references that contained female names or pronouns, installed a completely automated elevator that every-
and words such as “women” or “female.” Aside from one could operate. This automation, indeed, led to the
the enormously discriminatory outcome from a social demise of the job of the elevator operator.
perspective, the tool also kept Amazon from hiring the While many human tasks become obsolete, they
most qualified candidates. A less well-known example are often replaced by higher skilled (and higher paying,
is the legal proceeding between Avianca Airlines and more fulfilling) jobs. These posts, to borrow from

22 ComputingEdge
November/December 2023 IT Professional 15
March 2025
IT TRENDS IT TRENDS

Zuboff,15 have been “informated.” Elevator operators, could reliably prove that their code specifically informed
once the content of their work was automated, sub- the current version of the GitHub product, likely knowing
sequently shifted to greeting customers, advertising full well that this would be a near-impossible task, given
sales to customers, and making announcements.16 In the black-box problem of deep learning models.20
the context of education, students are empowered Also, related to the previous section, Copilot has
with technology that shifts the power and control to been reported to frequently produce nonworking code
students by giving access to new sources of informa- or code that is nowhere near consistent with what the
tion in the classroom. Instead of visiting a library and developer wants. GitHub’s own FAQ page reports that
working with a librarian, students have ready access to software developers only accept 26% of suggested
materials and can spend their time reading rather than code21; thus, developers and organizations must be vig-
looking for articles. At least until now, the dystopian ilant when reviewing and accepting completed code
vision of machines replacing all human skill sets clearly from Copilot, and, just because the suggested code
has not materialized. With generative AI, new worries compiles (which it often does not), this does not neces-
emerge in all walks of life, including computer pro- sarily mean that it will achieve what the developer had
grammers, technical writers, legal assistants, financial intended at runtime. Much like other AI-based produc-
advisors, market researchers, etc. Even the screen- tivity tools, it does appear to be helpful in writing rou-
writers’ strike in 2023 included the demand that AI, tine, lower order thinking code; thus, organizations
such as ChatGPT, be used only as a tool that can help may benefit by expecting to see some productivity
with research or facilitate script ideas and not as a tool gains for that specific context. As one Redditor (with a
to replace them.17 now deleted) account muses:
While generative AI is the closest we have come to
It’s pretty good for React [sic] and JavaScript at
mimicking human intelligence, and fears about job
speeding up predictable boilerplate type code.
losses because of generative AI are warranted, we
However, it’s like auto text - it gets it wrong too,
again assert that there are much more pressing near-
so you absolutely must know what you’re doing
term concerns for society and organizations. First,
and check it carefully. It’s using the most com-
many AI models today are almost certainly trained on
mon syntax and patterns, not necessarily your
copyrighted work, where the authors have not explic-
preferred or ‘best’ practices. Not good for more
itly provided consent to use their work. This represents
complex or unique use cases. It isn’t equally
an inadvertent (at best—willful at worst) exploitation
good for all languages. . . .22
of an immeasurable number of writers, artists, and
other content creators. ChatGPT learns from the writ- The proliferation of AI products has resulted in a
ing of people; deepfakes are built on the likenesses of new subeconomy of workers and businesses that offer
individuals; and GANs are based on photographs of data labeling and cleansing services, which, in turn, are
human faces, cartoon characters, or product designs— fed into training the underlying ML models of those
all without the consent or explicit permission of the products. Disturbingly, Hara et al.23 found that the
people involved in the original content. These have not thousands of workers who made up their study sample
been—nor likely will ever be—compensated for their trained AI models (mostly autotranscription models
efforts. In the case of GitHub’s industry-redefining that buttress popular services, such as OtterAI) at a
Copilot product, which provides intelligent code com- median wage of only $2/hour. In addition to the low
pletion, debugging, and suggestions in real time with pay, the task of data labeling is low-skilled, boring, and
much greater utility than its analog, non-AI-based pred- highly repetitive, with little career advancement oppor-
ecessors, some software developers reported that tunities or pathways to develop higher value skills. AI
their own previously written code was being suggested data processing workers in Kenya who supply cleansed
verbatim by the tool (e.g., Davis18). data to technology incumbents like Meta and OpenAI
In response, a class action lawsuit was filed by a have reported suffering from mental illnesses due to the
group of software engineers, alleging that GitHub and abhorrent working conditions, prompting labor action
its affiliated parties, which include corporate owner and leading to reputational damage to Meta and
Microsoft and development partner OpenAI, violated OpenAI.24,25 These examples show that the exploitation
copyright, privacy, and business laws by consuming of workers and content creators currently fuels the devel-
copyrighted developer source code without permission opment of the novel AI platforms we use, and designing
to train their models.19 Perhaps in a twist of irony, GitHub policy guidelines that protect these disenfranchised par-
and the defendants asked U.S. District Judge Jon Tigar ties is a much more pressing task than needlessly debat-
in California to dismiss the lawsuit until the plaintiffs ing a speculative future around automation.

16
www.computer.org/computingedge
IT Professional November/December 2023 23
IT TRENDS IT TRENDS

More important is the fact that the tools we use OpenAI competitors like Meta and Google. Models
today and the knowledge they help generate are shaped trained on sensitive information, such as patient records
directly by very powerful tech companies. OpenAI or a company’s trade secrets, could be revealed or com-
decided to filter out certain content for ChatGPT, mean- promised through the continued cat-and-mouse game
ing that these organizational decisions directly impact platform owners must play with prompt engineers and
the data accessible by users. Consequently, conserva- other nefarious actors. In fact, going back to the Copilot
tives in the United States observed that ChatGPT will example, GitHub explicitly states that it will retain
process the same queries asked about Trump differ- “source code that you are editing, related files and
ently than those related to Biden. Whether ChatGPT other files open in the same IDE [integrated develop-
betrays a distinct liberal bias is beyond the scope of ment environment] or editor,”21 meaning that individ-
this article. The fact that the content is curated accord- ual coding patterns and tendencies will be saved to
ing to the beliefs of a tech company, though, should train its models, notwithstanding an opt-out option it
worry us all as ChatGPT quickly replaces more open say it provides.
platforms such as Google and Wikipedia. More impor- This has serious implications for individual privacy,
tantly, as other companies, including the example of as next-generation biometrics technologies are focused
Mercedes shared at the beginning of this article, access on human behaviors, such as typing patterns, rather
the platform through APIs, we grant more and more than static, physical features, such as fingerprints.8 By
power to knowledge-related choices made by technol- identifying the writing and coding patterns of individu-
ogy companies like OpenAI. als, AI tools have the potential to collect and enumerate
While some of these biases are difficult to avoid many behavioral characteristics of individuals with little
and the sunk costs of exploitative labor practices have regard for their privacy or Web safety. Companies that
already been incurred in the trained models that are in release consumer-facing AI products or use the APIs
production today, managers can be cognizant of these of existing platforms may have significant exposure to
biases when evaluating the outputs of such models. the security and privacy risks that are baked into the
trained models that are currently in the wild.
Even in the case of incumbent platform Wikipedia, its
Companies also face the very real possibility that
founder Jimmy Wales has acknowledged that the vol-
their employees will input sensitive trade secrets into
unteer content editors who are integral to its opera-
ChatGPT. A recent report pointed to examples of a
tions are largely younger, technologically competent,
company executive inputting the firm’s strategy into
and college educated, which can sometimes leave a
the tool to generate a PowerPoint slide and of a doctor
left-leaning imprint on some of its politically oriented
inputting sensitive, personally identifiable information
articles.26 Managers should be hyperaware of the out-
of a patient to generate an insurance letter.27 These
puts of their AI-based products or the integration of
sensitive data can readily be consumed as training
AI-based outputs into their products and assess whether
data for future iterations of the product, resulting in
their marketing message will resonate with their cus-
material downside risk for organizations. As a result,
tomer base, given the potentiality of a baked-in political
companies, such as Microsoft, Amazon, Accenture,
leaning in these outputs. Any marketing messages that
and Wal-Mart, have all issued stern warnings and guid-
may inadvertently be influenced by the nature of the
ance to employees who use ChatGPT, and others,
trained language model must be scrutinized to ensure
including JP Morgan and KPMG, have banned its use
that it is consistent with the intent of the marketing
altogether.28 As large language models get cheaper
campaign. and faster to build, they will become ubiquitous, and,
already, Meta’s large language model can run on an
Control of Civilization Versus individual M1 Macbooks.29 As these powerful AI tools
Organizational Security Risks become more available to the public, policymakers
A near-apocalyptic threat is described in the open let- would likely benefit more from focusing on individual
ter, where humans are subservient to machines, which product safety than from debating tenuous, overarch-
evokes imagery from the 1980s and 1990s science fic- ing regulations to broadly impede advances in AI.
tion genre. By directing our public discourse toward Managers should be vigilant in ensuring that their
this existential and nightmarish scenario, however, we AI-based products not only keep their customers’ and
lose and trivialize very clear near-term threats with users’ privacy intact but that their marketing messages
emerging AI tools. For example, prompt engineering highlight this vigilance, particularly for corporate lead-
has already exploited many foundational flaws with ers who work in industries that are predicated on
ChatGPT and other large language models from sensitive, personally identifiable information, such as

24 ComputingEdge
November/December 2023 IT Professional 17
March 2025
IT TRENDS IT TRENDS

health care, security, and insurance. While such per- in the letter but that addressing the near-term risks
sonally identifiable information might be the necessary that we describe in this article and that other research-
input to train deep learning models today, managers ers have also identified is a much more tractable prob-
can still highlight the safeguards they have imposed lem that can be productively worked on so that we do
on the outputs of these models and establish the nec- not have to resist the temptation of AI.
essary rollback mechanisms in their databases in
case they receive any data deletion requests per
relevant General Data Protection Regulation and REFERENCES
Health Insurance Portability and Accountability Act 1. T. Gosh. “Mercedes Benz to add open AI's ChatGPT
requirements. into its cars; all you need to know.” BQ Prime.
Accessed: Jun. 23, 2023. [Online]. Available: https://
WHAT WILL A MORATORIUM www.bqprime.com/technology/mercedes-benz-to-add-
ACHIEVE? open-ais-chatgpt-into-its-cars-all-you-need-to-know-bqc
While there have been arguments for and criticisms 2. T. Aziz, “Changing the real estate game: Vancouver
levied against halting what some view as the inexora- Realtor creates GPT-powered AI tool,” CTV News, Jun.
ble march of technological progress, there are few 2023. Accessed: Jun. 23, 2023. [Online]. Available:
concrete comparison examples that have been iden- https://bc.ctvnews.ca/changing-the-real-estate-game-
tified that can inform what the future may look like if vancouver-realtor-creates-gpt-powered-ai-tool-1.
continued research on AI is, indeed, suspended. 6443415
Among other topics, we investigated the results of a 3. S. Spichak, “How new AI tools for doctors could
similar moratorium that was suggested by scientists worsen racial bias in healthcare,” MSN News, 2023.
in the West on gene-editing technologies (i.e., CRISPR), Accessed: Jun. 23, 2023. [Online]. Available: https://
which has largely been ignored by scientists in several www.msn.com/en-us/health/other/how-new-ai-tools-
other jurisdictions.30 This has resulted in a competitive for-doctors-could-worsen-racial-bias-in-healthcare/
imbalance that could lead to profoundly unexpected ar-AA1crASR
and unwanted organizational, societal, and competitive 4. C. Pitt, J. Paschen, J. Kietzmann, L. F. Pitt, and E. Pala,
outcomes. “Artificial intelligence, marketing, and the history of
As these historical examples show, community- technology: Kranzberg’s laws as a conceptual lens,”
driven moratoriums and oppressive enforcement actions Australas. Marketing J., vol. 31, no. 1, pp. 81–89, 2023,
by regulatory bodies offer no guarantees of curbing the doi: 10.1177/18393349211044175.
development of rapidly emerging novel technologies. 5. “Pause giant AI experiments: An open letter,” Future
In the case of CRISPR, a broad, globally proposed mora- of Life Institute, Cambridge, MA, USA, 2023. [Online].
torium on gene-editing research simply conferred a Available: https://futureoflife.org/open-letter/pause-
disproportionate advantage to scientists who disre- giant-ai-experiments/
garded the moratorium, who proceeded with their own 6. E. M. Bender, T. Gebru, A. McMillan-Major, and S.
research undeterred. In the case of blockchain product Shmitchell, “On the dangers of stochastic parrots:
regulation, heavy-handed enforcement actions by the Can language models be too big?” in Proc. ACM Conf.
Securities and Exchange Commission in the United Fairness, Accountability, Transparency, Mar. 2021,
States simply meant that some cryptocurrency projects pp. 610–623, doi: 10.1145/3442188.3445922.
moved their operations, or opened a subsidiary in El 7. S. Kapoor and A. Narayanan. “A misleading open
Salvador, a much more crypto-friendly country in terms letter about sci-fi AI dangers ignores the real risks.”
of regulation. As such, we argue that not only will a mor- AI Snake Oil. Accessed: Jun. 15, 2023. [Online].
atorium on further research on large AI models not Available: https://aisnakeoil.substack.com/p/a-
have the intended effect, but the doomsday scenarios misleading-open-letter-about-sci
painted in the letter have a more insidious outcome; 8. J. Killoran, Y. G. Cui, A. Park, P. van Esch, A. Dabirian,
they take our attention away from the actual near-term and J. Kietzmann, “Implementing behavioral biometrics
risks that are presented by modern deep learning with TRUST,” IT Prof., vol. 25, no. 1, pp. 13–16, Jan./Feb.
models that we must contend with in the near term: 2023, doi: 10.1109/MITP.2023.3236532.
improperly or inaccurately trained deep learning mod- 9. L. Whittaker, J. Kietzmann, K. Letheren, R. Mulcahy,
els, shifting leverage from workers to AI platform own- and R. Russell-Bennett, “Brace yourself! Why
ers, and organizational security risks. We suggest that managers should adopt a synthetic media incident
we should not dismiss the existential concerns posed response playbook in an age of falsity and synthetic

18
www.computer.org/computingedge
IT Professional November/December 2023 25
IT TRENDS IT TRENDS

media,” Bus. Horizons, vol. 66, no. 2, pp. 277–290, https://www.reddit.com/r/webdev/comments/ylfu70/

2023, doi: 10.1016/j.bushor.2022.07.004. weve_filed_a_lawsuit_challenging_github_copilot/
10. L. Whittaker, R. Mulcahy, K. Letheren, J. Kietzmann, 20. T. Ciaburn. “GitHub, Microsoft, OpenAI fail to wriggle
and R. Russell-Bennett, “Mapping the deepfake out of Copilot copyright lawsuit.” The Register.
landscape for innovation: A multidisciplinary Accessed: Jun. 25, 2023. [Online]. Available: https://
systematic review and future research agenda,” www.theregister.com/2023/05/12/github_microsoft_
Technovation, vol. 125, Art. no. 102784, Jul. 2023, openai_copilot/
doi: 10.1016/j.technovation.2023.102784. 21. “Your AI pair programmer. GitHub Copilot uses the
11. A. Singh, S. Sengupta, and V. Lakshminarayanan, OpenAI Codex to suggest code and entire functions
“Explainable deep learning models in medical image in real-time, right from your editor.” GitHub.
analysis,” J. Imag., vol. 6, no. 6, 2020, Art. no. 52, Accessed: Jun. 27, 2023. [Online]. Available: https://
doi: 10.3390/jimaging6060052. web.archive.org/web/20230323001023/https://github.
12. G. C. Kane, A. G. Young, A. Majchrzak, and com/features/copilot
S. Ransbotham, “Avoiding an oppressive future of 22. “December 2022 - Is GitHub copilot really that good?”
machine learning: A design theory for emancipatory Reddit. Accessed: Jun. 27, 2023. [Online]. Available:
assistants,” MIS Quart., vol. 45, no. 1, pp. 371–396, https://web.archive.org/web/20221207120829/https://
2021, doi: 10.25300/MISQ/2021/1578. www.reddit.com/r/webdev/comments/zc726g/
13. M. Bohannon, “Lawyer used ChatGPT in court—And december_2022_is_github_copilot_really_that_good/
cited fake cases. A judge is considering sanctions,” 23. K. Hara, A. Adams, K. Milland, S. Savage, C. Callison-
Forbes, Jun. 8, 2023. [Online]. Available: https://www. Burch, and J. P. Bigham, “A data-driven analysis of
forbes.com/sites/mollybohannon/2023/06/08/lawyer- workers' earnings on Amazon Mechanical Turk,”
used-chatgpt-in-court-and-cited-fake-cases-a-judge-is- in Proc. CHI Conf. Human Factors Comput. Syst.,
considering-sanctions/ Apr. 2018, pp. 1–14, doi: 10.1145/3173574.3174023.
14. C. DeVon, “ChatGPT probably won’t help you beat the 24. B. Perrigo. “Inside Facebook's African sweatshop.”
stock market—But it can assist investors in other Time. Accessed: Jun. 1, 2023. [Online]. Available:
ways, says CFP,” CNBC, Jun. 2023. Accessed: Jun. 23, https://time.com/6147458/facebook-africa-content-
2023. [Online]. Available: https://www.cnbc.com/2023/ moderation-employee-treatment/
06/23/cfp-chatgpt-wont-help-you-beat-the-stock- 25. B. Perrigo. “Exclusive: OpenAI used Kenyan workers
market-but-has-other-uses.html on less than $2 per hour to make ChatGPT less toxic.”
15. S. Zuboff, In the Age of the Smart Machine: The Time. [Online]. Available: https://time.com/6247678/
Future of Work and Power. New York, NY, USA: Basic openai-chatgpt-kenya-workers/
Books, 1988. 26. L. Fridman. “Podcast #385 with Jimmy Wales.” Lex
16. L. Miller, “Elevator girls moving in and out of the box,” Fridman. Accessed: Jun. 18, 2023. [Online]. Available:
in Modern Girls on the Go: Gender, Mobility, and https://lexfridman.com/jimmy-wales-transcript/
Labor in Japan, A. Freedman, L. Miller, and C. Yano, 27. R. Lemos. “Employees are feeding sensitive biz data
Eds., Stanford, CA, USA: Stanford University Press, to ChatGPT, raising security fears,” DarkReading, Mar.
2013, pp. 41–65. 7, 2023. [Online]. Available: https://www.darkreading.
17. A. Cullins and K. Killkenny, “As writers strike, AI Could com/risk/employees-feeding-sensitive-business-data-
covertly cross the picket line,” The Hollywood chatgpt-raising-security-fears
Reporter, May 3, 2023. Accessed: May 3, 2023. 28. K. Prescott, “Companies block ChatGPT amid privacy
[Online]. Available: https://www.hollywoodreporter. fears,” The Times, Feb. 23, 2023. [Online]. Available:
com/business/business-news/writers-strike-ai- https://www.thetimes.co.uk/article/companies-block-
chatgpt-1235478681/ chatgpt-amid-privacy-fears-5bgb5dlv9
18. T. Davis. “[@DocSparse] @githubcopilot, with ‘public 29. C. Chee [cedrickchee]. “4 Steps in Running LLaMA-7B
code’ blocked … [Tweet].” Twitter. Accessed: Jun. 15, on a M1 MacBook.” GitHub. Accessed: Apr. 10, 2023.
2023. [Online]. Available: https://twitter.com/ [Online]. Available: https://gist.github.com/
DocSparse/status/1581461734665367554?ref_ cedrickchee/e8d4cb0c4b1df6cc47ce8b18457ebde0
src=twsrc%5Etfw 30. A. Park. “Experts are calling for a ban on gene editing
19. M. Butterick. “We’ve filed a lawsuit challenging of human embryos. Here's why they're worried.” Time.
GitHub Copilot, an AI product that relies on Accessed: Apr. 10, 2023. [Online]. Available: https://
unprecedented open-source software piracy.” time.com/5550654/crispr-gene-editing-human-
Reddit. Accessed: Jun. 25, 2023. [Online]. Available: embryos-ban/

26 ComputingEdge
November/December 2023 IT Professional 19
March 2025
IT TRENDS IT TRENDS

31. N. Hochman. “ChatGPT goes woke.” National University, Kingston, ON, K7L 3N6, Canada. Contact him at j.
Review. Accessed: Jun. 23, 2023. [Online]. Available: [email protected].
https://www.nationalreview.com/corner/chatgpt-
goes-woke/ YUANYUAN (GINA) CUI is a marketing academic research-
ing consumer behavior at Auckland University of Technol-
ANDREW PARK is an assistant professor of information sys-
ogy, Auckland, 1010, New Zealand. Contact her at yuanyuan.
tems at the University of Victoria, Victoria, BC, V8P 5C2, Can-
[email protected].
ada. He is the corresponding author of this article. Contact
him at [email protected].
PATRICK VAN ESCH is an assistant professor of marketing
and professional sales at Kennesaw State University, Kenne-
JAN KIETZMANN is a management information systems
saw, Georgia, 30144, USA. Contact him at pvanesch@
and innovation researcher at the University of Victoria, Victo-
kennesaw.edu.
ria, BC, V8P 5C2, Canada, and an Erskine Fellow at UC Busi-
ness School, University of Canterbury, Christchurch, 1010,
New Zealand. Contact him at [email protected]. AMIR DABIRIAN is the provost/vice president for academic
affairs and a professor of marketing at California State Uni-
JAYSON KILLORAN researches management information versity, Fullerton, CA, 92831, USA. Contact him at adabirian@
systems with the Smith School of Business at Queen’s fullerton.edu.

Unlock Your Potential

WORLD-CLASS CONFERENCES —
Over 195 globally recognized conferences.
DIGITAL LIBRARY — Over 900k articles covering
world-class peer-reviewed content.
CALLS FOR PAPERS — Write and present your
ground-breaking accomplishments.
EDUCATION — Strengthen your resume with the
IEEE Computer Society Course Catalog.
ADVANCE YOUR CAREER — Search new positions
in the IEEE Computer Society Jobs Board.
NETWORK — Make connections in local Region,
Section, and Chapter activities.

Explore membership today

at the IEEE Computer Society
www.computer.org

CS-generic-halfhorizontal-Sept24.indd 1 9/30/24 5:

20
www.computer.org/computingedge
IT Professional November/December 2023 27
 EDITOR: Joanna F. DeFranco, The Pennsylvania State University, [email protected]

This article originally

appeared in
DEPARTMENT: INTERNET OF THINGS
vol. 56, no. 12, 2023

Unleashing the Potential:

Integrating ChatGPT and the
Internet of Things for Enhanced User
Experiences and Automation
Mohamad Kassab and Joanna F. DeFranco , The Pennsylvania State University

The integration of the Internet of Things (IoT) and ChatGPT can enhance user experiences and
facilitate process automation. We examine use cases, challenges, and future directions for
ChatGPT’s potential to revolutionize human–machine interactions within IoT ecosystems.

W
orldwide spending on the Internet of state-of-the-art technology to assist in more effec-
Things (IoT) is forecast to be US$805.7 tive treatment and healing strategies, such as sur-
billion in 2023, an increase of 10.6% over gery (robotic assisted), prosthetics, rehabilitation, and
that in 2022, according to a new International Data eldercare.3 In smart city applications, AI-powered ana-
Corporation Worldwide Internet of Things Spend- lytics can process vast amounts of data collected from
ing Guide.1 Investments in the IoT ecosystem are IoT sensors to enhance urban planning, traffic man-
expected to surpass US$1 trillion in 2026 with a com- agement, and energy optimization.4 The IoT has also
pound annual growth rate of 10.4% over the 2023–2027 made an impact on the agriculture industry with inno-
forecast period.1 However, despite the advancements vations to improve efficiency, increase crop yields, and
in the IoT, the user interfaces for interacting with these preserve natural resources.5
devices often rely on traditional methods, such as Conversational AI systems, such as ChatGPT,
mobile apps or physical controls, which can be cum- have witnessed remarkable advancements in natural
bersome and unintuitive for users. language understanding and generation, enabling
human-like interactions. Developed by OpenAI, Chat-
INTRODUCTION GPT is a state-of-the-art conversational AI system
The convergence of artificial intelligence (AI) and the built upon the foundation of deep learning and natural
IoT has emerged as a transformative paradigm, rev- language processing techniques. It is trained on a vast
olutionizing numerous sectors by enabling intelli- corpus of text data, enabling it to generate contextu-
gent automation and enhancing user experiences. In ally coherent responses to user inputs. The model
smart home environments, AI-powered voice assis- has achieved impressive performance in various
tants, such as Amazon’s Alexa and Google Assistant, language-related tasks, including question answer-
have gained popularity as they enable users to control ing, dialog generation, and language translation.6
IoT devices through natural language interactions. In The underlying transformer architecture of ChatGPT
health care, AI and IoT integration has not only paved allows it to capture long-range dependencies (that is,
the way for remote patient monitoring and person- semantic connections between words and sentences)
alized health-care systems2 but also has provided and contextual information, facilitating fluent and
coherent conversation with users.
While existing research and applications have dem-
Digital Object Identifier 10.1109/MC.2023.3297788 onstrated the potential benefits of integrating AI and
Date of current version: 13 November 2023 the IoT, the specific integration of ChatGPT and the IoT

28 March 2025 Published by the IEEE Computer Society  2469-7087/25 © 2025 IEEE
 remains relatively unexplored. This column presents catering responses to individual user preferences and
a few ChatGPT/IoT integration use cases, technical specific contexts. Leveraging ChatGPT’s language
implementation aspects, and challenges that should generation capabilities, personalized responses can
be addressed as this technology gains popularity. be tailored, resulting in heightened user satisfaction
and engagement. Research conducted by Salesforce8
INTEGRATION OF CHATGPT revealed that 66% of customers expect companies
AND THE IOT: USE CASES to understand their needs and provide personalized
AND APPLICATIONS experiences. By integrating ChatGPT with the IoT,
The integration of ChatGPT and the IoT showcases a the potential for personalized interactions becomes
wide range of captivating use cases and applications, achievable, enabling a higher degree of user satis-
profoundly transforming the paradigm of interactions
between humans and devices.
WHILE EXISTING RESEARCH AND
Natural language interface APPLICATIONS HAVE DEMONSTRATED
One of the primary advantages of integrating ChatGPT THE POTENTIAL BENEFITS OF
with the IoT lies in the provision of a natural language INTEGRATING AI AND THE IOT, THE
interface, fostering user interactions with their IoT SPECIFIC INTEGRATION OF CHATGPT
devices akin to human conversation. This integration AND THE IOT REMAINS RELATIVELY
eliminates the complexities associated with naviga- UNEXPLORED.
tion and explicit commands, resulting in a more intui-
tive and accessible interaction paradigm. For instance,
instead of relying on predefined commands to control faction and engagement, leading to improved user
smart home devices, users can engage in dialog with experiences in various domains. For instance, an
their IoT assistants, expressing their intentions and IoT-enabled smartwatch integrated with ChatGPT
preferences using more human-like language. With can monitor heart rate, sleep patterns, and activity
the increasing pervasiveness of voice assistants, 73% levels, and based on these data points, engage in con-
of consumers seek correctness, accuracy, and con- versational interactions with the user, offering per-
sistency in voice assistant interactions, as reported sonalized advice on exercise routines, diet plans, and
by a PwC report.7 The integration of ChatGPT effec- stress-management techniques.
tively addresses this demand. To enable voice inte-
gration with ChatGPT, a combination of speech rec- Real-time monitoring and alerts
ognition technologies that convert spoken language The integration of ChatGPT with IoT devices enables
into text (for example, Google Cloud Speech-to-Text, real-time monitoring and alerts, empowering devices
Amazon Transcribe, or Microsoft Azure Speech Ser- to actively detect critical events or anomalies. By con-
vice) and the ChatGPT API can be employed to enable tinuously analyzing data streams from connected sen-
voice-based interactions with IoT devices. sors, ChatGPT models possess the capability to iden-
tify patterns that deviate from normal behavior and
Context-aware assistance promptly notify users about potential issues, facilitat-
The integration of ChatGPT with IoT devices facili- ing proactive actions or automated responses to mit-
tates personalized and context-aware interactions, igate risks across various domains. For instance, in a

www.computer.org/computingedge 29
INTERNET OF THINGS

high-security facility equipped with an IoT-enabled IMPLEMENTATION

surveillance system integrated with ChatGPT, dis- CONSIDERATIONS AND
tributed video cameras connected to a central mon- CHALLENGES
itoring system can be utilized. The ChatGPT model While the integration of the IoT with ChatGPT brings
is trained to recognize suspicious activities, such as forth promising opportunities, it also presents several
unauthorized access attempts, unusual movements, challenges that need to be addressed to ensure suc-
or unattended objects, by analyzing live video feeds cessful implementation and deployment.
in real time. Through continuous analysis of the video
data, the system can effectively identify and classify Data security and privacy concerns
potential security threats. Once an anomaly or suspi- The integration of AI chatbot systems like Chat-
cious activity is detected, ChatGPT triggers an imme- GPT with IoT devices introduces potential security
diate alert, notifying security personnel or designated risks and privacy concerns as both ChatGPT and IoT
authorities through dedicated applications or com- devices may handle sensitive user information, neces-
munication channels, thereby enabling prompt action sitating robust security measures to protect data
and response to potential security breaches, effec- confidentiality, integrity, and availability. According
tively mitigating risks and ensuring the safety of the to the “2022 State of AI in the Enterprise” report by
facility and its occupants. Furthermore, by leverag- Deloitte, 50% of respondents cited management of
ing ChatGPT’s language generation capabilities, the AI-related security risks as one of the top inhibitors
surveillance system can generate detailed incident to starting and scaling AI projects. 9 The chatbot and
reports that encompass vital information, such as virtual assistant industry, which shares similarities
time stamps, descriptions of observed activities, and with ChatGPT in terms of natural language interac-
additional contextual details. tions, has witnessed security incidents. For instance,
in 2020, a chatbot vulnerability allowed unauthorized
Predictive analytics access to customer data of Verizon’s telecommuni-
Integrating ChatGPT with the IoT can facilitate pre- cations company, affecting millions of users. Hackers
dictive analytics by leveraging historical data and are also using ChatGPT to create smart malware as
machine learning capabilities. For example, the inte- it can autonomously sift through more data than a
gration can be leveraged for the development of a human can.10
predictive maintenance model within the context OpenAI follows security practices to protect appli-
of industrial systems. The initial phase involves cation programming interface (API) data. For example,
training the model using historical data, enabling it retains API data only for 30 days for abuse and misuse
it to discern patterns and establish relationships monitoring purposes (https://openai.com/policies/api
between various features and maintenance events. -data-usage-policies). Moreover, the OpenAI API is
Subsequently, the performance of the GPT-based only accessible over Transport Layer Security, ensur-
predictive maintenance model can be evaluated ing that customer-to-OpenAI requests and responses
using established metrics, such as accuracy, pre- are encrypted. To further enhance security, essential
cision, recall, or F1 score (a combined measure of measures to safeguard data confidentiality can be
precision and recall). Validation techniques, includ- adopted, for example,
ing cross-validation, can be employed to ensure the
model’s capacity for generalization. Once the GPT › strict input validation mechanisms on the
model is trained, it can be seamlessly integrated sensors to prevent malicious commands or data
into the operational system, establishing connec- from being processed by ChatGPT
tions with real-time data sources, like sensor feeds. › granular user authorization and access control
This integration facilitates continuous monitoring mechanisms to restrict access to ChatGPT
of machinery or equipment, ensuring prompt iden- functionalities
tification of potential maintenance requirements or › compliance by cybersecurity standards, such
anomalies in the system. as the National Institute of Standards and

30 ComputingEdge March 2025

 INTERNET OF THINGS

Technology Cybersecurity Framework,11 and of ChatGPT models using domain-specific data can
regulations and laws, such as Europe’s General improve the contextual understanding and accuracy
Data Protection Regulation or the Health Insur- of responses.
ance Portability and Accountability Act. Building trust in IoT systems integrated with Chat-
GPT is crucial for user adoption and acceptance. Users
Training data and adaptation need to have confidence in the system’s reliability,
Training data availability, relevance, and privacy are transparency, and the accuracy of its responses. The
important considerations when integrating ChatGPT black-box nature of deep learning models like ChatGPT
with IoT devices. Collecting domain-specific training can hinder user trust and understanding. Providing
data related to the IoT context can improve the mod- explanations for system decisions, disclosing limita-
el’s understanding of user queries and enhance the tions, and offering user control over the conversa-
accuracy of generated responses. Fine-tuning tech- tional experience can help address these concerns.
niques can be employed to adapt ChatGPT to specific Additionally, adhering to ethical guidelines, ensuring
IoT use cases, incorporating domain-specific language unbiased training data, and conducting rigorous test-
patterns and context. However, acquiring and labeling ing and validation can foster user trust and confidence
training data for specific IoT domains may pose chal- in the integrated system.
lenges (for example, limited availability, heterogeneity,
and diversity), necessitating novel approaches, such FUTURE DIRECTIONS
as data augmentation or transfer learning. Privacy is The integration of ChatGPT with IoT systems holds
also a concern as data collection and Internet scrap- great promise for enhancing user experiences and
ing must remain lawful, and consent is granted when enabling intelligent interactions. However, there are
private information is used to train ChatGPT. several future directions and challenges that need
to be addressed to fully exploit this integration. One
Data processing important area of focus is improving the contextual
The integration of the IoT with ChatGPT requires han- understanding of ChatGPT in the IoT domain, enabling
dling a massive influx of data generated by IoT devices. it to interpret IoT-specific terminologies, device func-
ChatGPT API responses even for short API calls with tionalities, and user preferences. Incorporating mul-
200–400 tokens take 20–30 s. Efficient data prepro- timodal interactions, such as voice and gestures, can
cessing techniques, such as filtering, aggregation, enrich the user experience and will require research in
model parallelism, and distributed inferences, can be fusion techniques and multimodal dialog systems.
explored to optimize resource utilization and accom- Another area of focus should be improving explain-
modate performance requirements. ability and trust. It is essential to improve methods
that provide explanations for ChatGPT’s responses
Contextual understanding that can enhance user confidence.
and ambiguity Edge computing and federated learning tech-
IoT devices generate vast amounts of sensor data, and niques can address concerns related to latency, pri-
interpreting these data accurately within the context vacy, and bandwidth limitations. Other considerations
of a conversation is crucial for meaningful interac- are ethics, which includes bias mitigation, fairness,
tions with ChatGPT. However, contextual understand- and user consent.
ing and disambiguation can be challenging because And finally, interoperability and standardization
of the inherent variability and noise present in IoT efforts are crucial to enable seamless integration
data streams. For instance, sensor readings might be between ChatGPT and diverse IoT devices.
incomplete, inconsistent, or prone to environmental
factors. Advanced data preprocessing techniques,
data fusion algorithms, and context-aware modeling
approaches can be employed to address these chal-
B y tackling these challenges and exploring these
future directions, the integration of ChatGPT and
the IoT can revolutionize human–machine interactions
lenges. Additionally, ongoing training and fine-tuning in IoT ecosystems.

www.computer.org/computingedge 31
INTERNET OF THINGS

REFERENCES /consumer-intelligence-series/voice-assistants.pdf
1. “Worldwide internet of things spending guide,” IDC 8. “What are customer expectations, and how have they
Corporate, Needham, MA, USA, 2023. [Online]. Avail- changed? Customer expectations hit all-time highs,”
able: https://bit.ly/3qZM5jv Salesforce, 2021. [Online]. Available: https://bit
2. F. Alshehri and G. Muhammad, “A comprehensive .ly/44hAkDu
survey of the Internet of Things (IoT) and AI-based 9. “‘State of AI in the enterprise’ fifth edition uncovers
smart healthcare,” IEEE Access, vol. 9, pp. 3660–3678, four key actions to maximize AI value,” Deloitte, Oct.
2021, doi: 10.1109/ACCESS.2020.3047960. 2022. [Online]. Available: https://bit.ly/3Pw7zyT
3. J. DeFranco and M. J. Metro, “Internet of telemedicine,” 10. R. Reiter, “3 ways hackers use chatGPT to cause
Computer, vol. 55, no. 4, pp. 56–59, Apr. 2022, doi: security headaches,” Dark Reading, May 2023. [Online].
10.1109/MC.2022.3143625. Available: https://bit.ly/3pRWgqi
4. V. V. Graciano Neto and M. Kassab, What Every 11. “Cybersecurity framework,” Nat. Inst. Standards Tech-
Engineer Should Know about Smart Cities. Boca Raton, nol., Gaithersburg, MD, USA, 2013. [Online]. Available:
FL, USA: CRC Press, 2023. https://www.nist.gov/cyberframework
5. J. DeFranco, N. Kshetri, and J. Voas, “‘Sensoring’ the
farm,” Computer, vol. 56, no. 10, pp. XX–XX, Oct. 2023,
doi: 10.1109/MC.2023.3296762. MOHAMAD KASSAB is an associate research professor of
6. ChatGPT: A Model for Generating Conversational software engineering at The Pennsylvania State University,
Responses. (2023). OpenAI. [Online]. Available: https:// Malvern, PA 19355 USA. Contact him at [email protected].
openai.com
7. “Prepare for the voice revolution: An in-depth look at JOANNA F. DEFRANCO is an associate professor of software
consumer adoption and usage of voice assistants, and engineering at The Pennsylvania State University, Malvern,
how companies can earn their trust—And their busi- PA 19355 USA as well as the associate director of the Doctor
ness,” PWC, Mar. 2018. [Online]. Available: https://www. of Engineering degree program. She is also an associate edi-
pwc.com/us/en/advisory-services/publications tor in chief of Computer. Contact her at [email protected].

CG& A IEEE Computer Graphics and Applications bridges the theory

and practice of computer graphics. Subscribe to CG&A and
www.computer.org/cga

• stay current on the latest tools and applications and gain

invaluable practical and research knowledge,
• discover cutting-edge applications and learn more about
the latest techniques, and
• benefit from CG&A’s active and connected editorial board.

32 ComputingEdge March 2025

Get Published in the
IEEE Transactions on Privacy
This fully open access journal is
soliciting papers for review.
IEEE Transactions on Privacy serves
as a rapid publication forum for
groundbreaking articles in the realm
of privacy and data protection.
Submit a paper and benefit from
publishing with the IEEE Computer
Society! With over 5 million unique
monthly visitors to the IEEE Xplore®
and Computer Society digital
libraries, your research can benefit
from broad distribution to readers in
your field.

Submit a Paper Today!

Visit computer.org/tp to learn more.
 EDITOR: Domenico Talia, University of Calabria, [email protected]

This article originally

appeared in
DEPARTMENT: HUMANITY AND COMPUTING
vol. 57, no. 7, 2024

Algorithms That Can Deny Care,

and a Call for AI Explainability
Domenico Talia , University of Calabria

Recent examples of negative use of big data and machine learning come from their use
in health care decisions involving a large number of patients. These cases highlight the
need for algorithm explainability to help us better understand how artificial intelligence
works in solving problems so we can then evaluate its accuracy and effectiveness.

I
n the first article published in this column,1 we
discussed several success stories about artifi- THE SOFTWARE SYSTEM USED BY
cial intelligence (AI) and machine intelligence UNITEDHEALTH GROUP AND HUMANA
solutions effectively used in health care for detect- IS THE NH PREDICT AI TOOL THAT
ing diseases, supporting doctors, helping patients, USES A DATABASE OF A FEW MILLION
improving health-care processes, and saving lives. PATIENTS COMPILED OVER YEARS.
Together with many cases where big data, machine
learning algorithms, and AI systems are improving
people’s diagnoses and treatments, recently raised According to the filed lawsuits, the United States’s
examples of negative use of these technologies in real largest health insurance companies pressured their
scenarios involving a large number of patients. The use medical staff to reduce payments for very ill patients
of faulty data analysis and machine learning is becom- at large scale, exploiting an algorithm that calculates
ing a problem for patients and the health-care industry the needed rehabilitation days and limits rehabilita-
because of problematic uses due to profit goals, algo- tion care for older and disabled Americans while prof-
rithmic racial bias in patient care,2 and ethical issues. its increased.3 On the other hand, UnitedHealth and
Humana state that the algorithm they used, which
A CLASS ACTION LAWSUIT THAT predicts how long patients will need to stay in rehabili-
INVOLVES AN ALGORITHM tation, is just a tool used to estimate recoveries. They
Recently, a class action lawsuit was filed against Unit- say the decision is made by (human) managers and
edHealth and a subsidiary based on the hypothesis employees. Some newspapers in the United States
that they are illegally using an algorithm to deny or wrote that managers must follow the algorithm pre-
limit rehabilitation care to seriously ill patients, even dictions precisely so payment could be discontinued
though the companies know the algorithm has a high by the date it predicted.
error rate. After one month, a similar class action law- The software system used by United Health Group
suit was filed against Humana for the same algorithm. and Humana is the nH Predict AI tool that uses a data-
We must consider that UnitedHealth and Humana base of a few million patients compiled over years.
are among the top providers of the popular plans for The algorithm of nH Predict uses this large database
seniors and account for nearly half of all medical assis- to analyze a patient’s diagnosis, living situation, age,
tance enrollees in the United States. gender, physical function, admission date, and other
information for predicting how much postacute care
a patient “should” need. According to the lawsuits, the
Digital Object Identifier 10.1109/MC.2024.3387012 algorithm settles on the day when UnitedHealthcare or
Date of current version: 26 June 2024 Humana will cut payment for patient care. nH Predict

34 March 2025 Published by the IEEE Computer Society  2469-7087/25 © 2025 IEEE
 also considers the patient’s usual living setting, that is,
IT IS EVIDENT THAT THE ROLE OF
if they are at home alone or in an assisted living facility.
THE PREDICTION ALGORITHM IN
The outcome report provided by nH Predict pro-
THIS SCENARIO IS VERY IMPORTANT
vides a sort of profile of each patient that includes a AND SIGNIFICANTLY CONTRIBUTES
score for a few of the patient’s functions also based on TO THE FINAL DECISION ABOUT THE
the data of similar patients analyzed in the past. The REHABILITATION DURATION.
profile includes scores on the patient’s basic mobility,
such as wheelchair skills or ability to take the stairs,
cognitive abilities, such as memory and communica- In other words, insurers use opaque (for the public)
tion, and daily activity (for example, dressing and predictive algorithms that are becoming progressively
bathing). The profile report produced by nH Predict influential in decisions about patient care and cover-
includes a total average score for the patient that is age. These cases call for algorithm explainability, an
based on a combination of single scores. issue that in the area of AI is becoming urgent.
When the nH Predict patient report is ready, physicians
and company managers use it for making evidence-based AN APPEAL TO EXPLAINABILITY
decisions on a rehabilitation plan for the patient based on As discussed by de Franco et al.,4 a problem may occur
personal medical condition and functional necessities. It when algorithms make important decisions that affect
is evident that the role of the prediction algorithm in this people or are coded to influence humans provid-
scenario is very important and significantly contributes ing suggestions and information to the public. This is
to the final decision about the rehabilitation duration. not only the case of the nH Predict system, but it may
Unfortunately, the machine learning algorithm occur with many algorithms, such as social media pro-
used in the nH Predict system is proprietary and not filing systems, financial intelligence products, customer
disclosed in detail by the owner NaviHealth. Thus it is profiling, and other algorithms that are designed to
unclear how the nH Predict system works exactly, that increase profits for companies rather than do the best
is how the patient data are used together with the infor- for the common good. In all these cases, individuals and/
mation of other similar patients to estimate medical or society need to know how the algorithms work and
needs, length of stay, and discharge date. In a few words, question about their real goals. This calls for algorithm
the nH Predict algorithm suggests a decision to the explainability that describes what happens in the code
health insurance company manager to process patient from input to output. AI algorithm explainability makes
data in a way that is not known to patients and their learning models transparent and solves the “black box”
families. In fact, when patients or their physicians have problem. To reach this goal, auditing of algorithms can
requested to know how the nH Predict’s reports are be effective to guarantee that they are legal, ethical, and
generated, as expected, UnitedHealth has denied their safe.5,6 Algorithm auditors can look at the inputs and
requests, telling them the algorithm is proprietary. When outputs of a decision system from the outside.
prescribing physicians disagree with UnitedHealth’s Explainable AI (XAI) defines a set of methods
algorithm-based determination of how much postacute and techniques that refer to learning algorithms that
care their patients need, their judgments are ignored. humans can comprehend and trust together with the
Doctors and patients in those cases are neither results/decisions they generate. XAI is the opposite of
aware of the algorithm procedures, nor able to question black box systems where owners and users cannot (or
their data processing strategies and final decisions. don’t want to) explain why an AI algorithm reached a

www.computer.org/computingedge 35
HUMANITY AND COMPUTING

INSURERS USE OPAQUE (FOR THE should be subject to suitable safeguards, which
PUBLIC) PREDICTIVE ALGORITHMS should include specific information to the data
THAT ARE BECOMING PROGRESSIVELY subject and the right to obtain human intervention,
INFLUENTIAL IN DECISIONS ABOUT to express his or her point of view, to obtain an
PATIENT CARE AND COVERAGE. explanation of the decision reached after such
assessment, and to challenge the decision.”

certain decision or a specific result. Black box AI sys- In addition, Article 22 grants an individual a “right of
tems designed for decision-making map data features human intervention.” Under this right, an individual may
into a class by predicting events or the behavioral traits ask for a human to review the AI algorithm’s decision to
of citizens without providing the reasons why they did determine whether or not the system made a mistake.
it. This opaque approach is elusive not only for the lack An approach similar to that one used in EU GDPR
of transparency but also for possible biases the black should be embodied in laws and regulations in other
box algorithms inherit from human discrimination and/ countries allowing people to have clear explanation of
or errors originating (often hidden) in the training data, how algorithms work, how they use their data, and the
which may lead to unjust or erroneous decisions.7 rules they follow to make decisions.
AI models used in the United States by physicians
to detect diseases such as cancer or mental diseases, NEW SKILLS AND ABILITIES TO
or suggest the most effective treatment, are assessed FACE AI-BASED DECISIONS
by the Food and Drug Administration. But tools used Together with the increasing role of algorithms in our
by insurers for determining whether hospitalization or daily lives and with the increasingly widespread use of
treatments should be reimbursed are not subjected to AI in decision making, we can witness that the scarcity
the same inspection; however, they are equally impor- of digital culture and computational thinking becomes
tant for patients. In Europe, the impact of algorithm an ever-greater handicap for individuals and for the
decisions is regulated under the EU’s General Data entire society. The digital illiteracy widely spread in the
Protection Regulation (GDPR). A professional using population risks of being unaware users of “intelligent”
personal data for automated processing must explain systems that make decisions about us. The digital tsu-
to the people concerned how the system makes deci- nami is advancing, and citizens must get skills and abil-
sions. The individual data subject (for instance, the ities to handle it, otherwise they will be overwhelmed.
person who was rejected for a payment or a patient who People skill and knowledge on AI and the most
was denied a certain medical service) has the right to promising areas of digital technology show a delay
ask the company why it made the decision it did, and that, if it will not be filled, will be another key element
the company must then explain how the system came for the failure to progress. We must not work to train
to its decision. If the company can’t explain the decision everyone to become a software designer, but we must
in response to an individual’s request, it would not be understand that it is essential to provide basic knowl-
complaint with the GDPR and can be prosecuted by law. edge to deal with apps, software tools, and platforms
For instance, Recital 71 of GDPR8 states: that use AI applications. We need concrete and wide-
spread actions to increase expertise of young people in
“The data subject should have the right not to be information technology and put the citizens in condi-
subject to a decision, which may include a measure, tion of living the present time with the necessary skills.
evaluating personal aspects relating to him or Otherwise, we risk the rising of a community of people
her which is based solely on automated process- who will be passive entities instead of being active
ing and which produces legal effects concerning subjects in a world where AI is becoming pervasive.
him or her or similarly significantly affects him or To address these issues, it is necessary to build a
her, such as automatic refusal of an online credit widespread basic training that makes people understand,
application or e-recruiting practices without any for example, what AI is, how computers work in solving
human intervention. … In any case, such processing problems, being able to evaluate information online,

36 ComputingEdge March 2025

 HUMANITY AND COMPUTING

its truthfulness, and its usefulness. It serves to provide repositories. To fill this gap is essential to work on new
young and old people the opportunity to understand the cognitive paradigms, to change education programs
computational processes, the logic that governs them, and practices, and to offer the opportunity to people to
and their potential that can be exploited in work and access data and learn how to use them successfully and
in everyday life. While everyone agrees on the obvious how them are used in AI systems to profile them or to
need to teach mathematics from elementary schools, make decisions on their lives. This integrated strategy
it should be equally normal to teach the principles of could result in new opportunities for those who wish
computer science in all schools. Kids need to know how to have an active role in the new digital society, offer a
to understand and compose simple algorithms, compu- way to defend people rights, and can improve the global
tational thinking, and thinking methodically to strate- knowledge of people on intelligent technologies.
gies useful for defining solutions. They need to know
how AI works and how to cooperate with it as intelligent REFERENCES
actors and not to be subjected to it. On the other hand, 1. D. Talia, “Machine intelligence for human health: A few
adults must be able to discuss and understand how AI noteworthy cases,” Computer, vol. 55, no. 10, pp. 82–86,
algorithms are used when they concern decisions about Oct. 2022, doi: 10.1109/MC.2022.3190786.
them, their rights, their health, and their work. 2. D. M. Christensen, J. Manley, and J. Resendez, “Medical
The widespread use of AI solutions in daily life is cre- algorithms are failing communities of color,” Health
ating a large social and cultural divide between the few Affairs, Washington, DC, USA, Sep. 2021. [Online]. Avail-
experts and the many oblivious users. This separation, able: https://www.healthaffairs.org/content/forefront
other than creating new inequalities, will cause much /medical-algorithms-failing-communities-color
damage to our society. According to a recent study of 3. C. Ross and B. Herman. “Denied by AI: How Medicare
the World Economic Forum, employers estimate that Advantage plans use algorithms to cut off care for
44% of workers’ skills will be disrupted in the next four seniors in need.” STAT. Accessed: Apr. 15, 2024. [Online].
years. It is necessary to take steps to manage the future Available: https://www.statnews.com/2023/03/13
that arrived too quickly. Politics, schools, and universi- /medicare-advantage-plans-denial-artificial-intelligence/
ties are called to face this main challenge. If this will not 4. J. F. DeFranco, J. Voas, and N. Kshetri, “Algorithms:
be done now, enormous problems will arise in the com- Society’s invisible puppeteers,” Computer, vol. 55, no. 4,
ing decades, when the intensive use of new intelligent pp. 12–14, Apr. 2022, doi: 10.1109/MC.2021.3128675.
technologies will create many new jobs that do not exist 5. D. Talia, From Algorithms to Thinking Machines: The
today and will abolish many of those being done today. New Digital Power. New York, NY, USA: ACM, 2023.
At the same time, private companies using AI in their 6. M. Jovanović and M. Schmitz, “Explainability as a user
decision processes about clients, should realize that requirement for artificial intelligence systems,” Com-
they must provide clear explanation of automated deci- puter, vol. 55, no. 2, pp. 90–94, Feb. 2022, doi: 10.1109/MC
sions. We are now also experiencing the new big data .2021.3127753.
divide that refers to the split between those who have 7. N. Prentzas, A. Kakas, and C. S. Pattichis, “Explainable
access and ownership to large and distributed data AI applications in the medical domain: A systematic
sources and those that do not. Andrejevic9 worked on review,” 2023, arXiv:2308.05411.
the concepts of access, ownership, and use of digital 8. “General data protection regulation.” Intersoft Consult-
data, arguing that a big data divide is occurring today ing. Accessed: Feb. 1, 2024. [Online]. Available: https://
that deals with the asymmetric relationship between gdpr-info.eu/
those who collect, store, and mine large quantities of 9. M. Andrejevic, “Big data, big questions: The big data
data, and those whom data collection targets. divide,” Int. J. Commun., vol. 8, no. 1, pp. 1673–1689, 2014.

T he issues involved in this asymmetric relationship

relate to a basic divide that occurs for people and
organizations without the technical means to access
DOMENICO TALIA is a full professor of computer engineer-
ing at the University of Calabria, 87036 Rende, Italy. Contact
and analyze data stored on the Internet or in large data him at [email protected].

www.computer.org/computingedge 37
 COLUMN: VISION
COLUMN: VISION AND
AND VIEWS
VIEWS
This article originally
appeared in

Lifeblood of Health is Data vol. 29, no. 1, 2022

Lifeblood of
Ramesh Jain , University of Health
California, isUSAData
Irvine, CA, 92697,

Ramesh Jain , University of California, Irvine, CA, 92697, USA

Health is the most important aspect of life. However, health is the least developed
aspect of modern society that increasingly uses data as a major resource for
understanding complexities in a system and relies on predictive control for the
Health is the most important aspect of life. However, health is the least developed aspect
overall improvement of systems. Progress in biology, multimodal sensors, mobile
of modern society that increasingly uses data as a major resource for understanding
devices, computing,
complexities and and
in a system related areas
relies is rapidly establishing
on predictive that
control for the health
overall is a
improvement of
perpetualProgress
systems. personalindata system
biology, rather than
multimodal current
sensors, population-based
mobile episodic
devices, computing, and related
disease-oriented
areas healthcare that
is rapidly establishing approach.
healthLifestyle is more
is a perpetual important
personal dataforsystem
controlling
rather than
health than
current medicine. Technology
population-based episodicnow allows us to perpetually
disease-oriented healthcare measure
approach. lifestyle
Lifestyle is more
and analyze
important foritcontrolling
to build personal
health models and estimate
than medicine. healthnow
Technology state for developing
allows us to perpetually
measure
predictive,lifestyle and analyze
preventive, it to build
personalized personal
approach for models and estimate
maximizing healthiness health state for
for most
developing
people in the predictive, preventive,
world. These emerging personalized
approaches approach for maximizing
have demonstrated theirhealthiness
early
for most and
success people areinnow
theready
world.for
These emerging
helping peopleapproaches
achieve theirhave demonstrated
health goals. Datatheir
is theearly
success
lifebloodand are now
of health; ready for helping
technology people
has a great achieve their
opportunity health
to bring goals.
this Data is the lifeblood
imminent
of health; technology
transformation now. has a great opportunity to bring this imminent transformation now.

H
ealth is the most important aspect of society, person based on their perpetual lifestyle, environment,
yet healthcare is in shambles in every country. and biological data, and estimates her health state to
The COVID-19 pandemic has shown that health provide needed continuous guidance. This approach
is the most important aspect of each individual and the empowers people to better manage their health and
whole human society. It also exposed strong weak- quality of life with data-enabled, personalized, and timely
nesses in the current reactive healthcare systems glob- advice. This requires collecting real-time lifestyle and
ally. As discussed by a well-respected health scientist Dr. environment data in real-life situations to understand
Eric Topol in his books,13 time is right for the creative lifestyle and environmental factors as they affect an indi-
destruction of medicine. Progress in systems biology, vidual and use them to build a personal model that will
multimodal sensing technologies, mobile devices, artifi- steer his health state to meet his goals.
cial intelligence including machine learning, and cloud Although everyone is a product of genetics, envi-
computing are exposing limitations of current episodic ronment, and lifestyle, yet the popular evidence-based
healthcare approaches. Future health will use multi- medicine relies on population data assuming that
modal perpetual measurements combined with sophisti- each person is just a representative sample of a popu-
cated multimodal diagnostics to build high fidelity model lation. As shown in Figure 1, there is a big mismatch in
of everyone for perpetual nonobtrusive guidance based factors that make us happy and the amount that we
on the personal model and current health state of the spend on these factors. Environment and lifestyle
person to help her maximize quality of life. related behavior are responsible for 70% or more of
Time is right to develop technology toward a more our health,15 but current healthcare mostly gives lip
personalized health model that considers health rather service to these and focuses exclusively on medicine.
than sickness as the driving theme for each person. By Increasingly this realization is resulting in more atten-
recognizing that each person is a unique dynamic sys- tion to lifestyle and environment. Future health will
tem resulting from genetics, but strongly influenced by focus more on lifestyle and will use medicine only
lifestyle and environment, it builds models of each when essential rather than using medicine and provid-
ing lip service to lifestyle.
It is interesting to point out that all ancient health
systems believed that we could control our health
1070-986X  2022 IEEE through lifestyle. Hippocrates has popularly said “Let
Digital Object Identifier 10.1109/MMUL.2022.3151996 food be thy medicine, and let medicine be thy food,”
Date of current version 4 May 2022. more than 2400 years ago.

38
128 March
IEEE2025
MultiMedia Published
Publishedby
bythe
theIEEE
IEEEComputer
ComputerSociety
Society  2469-7087/25 © 2025
January-March 2022 IEEE
 VISION
VISIONAND
ANDVIEWS
VIEWS

FIGURE 2. Main component of cybernetic systems is feed-

back for controlling the real-time behavior of complex real-
world systems. In these systems, the destination (or desired
state) is perpetually compared with the current state, and
the discrepancy or error is used to guide the system.

FIGURE 1. On the left are the health factors that make us cure a disease are efforts to augment body’s homeo-
happy. On the right, the common expenses by people are stasis to bring back the body to a normal state.
shown on the same factors. Chronic diseases such as diabetes, cardiac dis-
eases, asthma, acidity, and cancer result due to failing
homeostasis. When a chronic disease is detected,
This approach to caring for human health is a navi- medication is given to compensate for the deficient
gational approach that builds on the strengths of element in the homeostasis loop. In many cases,
cybernetics8 and is imminent. It uses more effective the deficiencies may be compensated externally by
and efficient personal lifestyle management and facili- adjusting lifestyle parameters. This could be done
tates the application of health knowledge to guide even in the prodromal state of the disease. A very
and maintain health and maximize quality of life. good example of this approach is now available for
Diabetes Type 1 where a sensor to measure blood glu-
cose level continuously is connected to an insulin
Cybernetics is the Key pump using a computing device to implement the
Cybernetic principles14 proposed by Norbert Wiener as loop like Figure 2 to help a patient cope with this seri-
a mechanism for control and communication in ous health disease.2
machines as well as living systems have transformed
the design of complex systems. Continuous measure-
ments are a key component in closed-loop feedback Cybernetization Results in
control, which is essential for implementing systems Transformations
that work in real-world noisy environments. The application of cybernetization opens new opportu-
Cybernetics is about setting goals and devising nities to develop humanitarian systems that could
action sequences to accomplish and maintain those improve people’s quality of life. The cybernetization is
goals in the presence of noise and disturbances (see enabled by the availability of appropriate sensors to
Figure 2). In a recent book,7 John Markoff traces many estimate the system state and perpetually feed this
breakthroughs in AI, such as machine learning, adap- information back to the system, to generate new con-
tive systems, and robotics to cybernetics. trol signals as required to move toward the desired goal.
One of the latest transformative examples is a
commonly used map-based navigation system. Navi-
Homeostasis and Cybernetics gation systems have eventually replaced maps, which
The metabolic processes in our body can only function date all the way back to about 2300 B.C. Over the
in very specific physical and chemical environments. years, maps evolved with technology.12 At the start of
Homeostasis is a self-regulating process by which the 21st century, online maps became very popular.
body maintains stability while adjusting to changing Powerful algorithms let people plan their routes using
external conditions. A disruption of homeostatic digital maps, so users could print turn-by-turn direc-
mechanisms leads to diseases. Most approaches to tions for optimal routes based on different requests,

www.computer.org/computingedge
January-March 2022 IEEE MultiMedia 39
129
 VISIONAND
VISION ANDVIEWS
VIEWS

FIGURE 3. Cybernetization of navigation systems required automatically knowing the current location and using it perpetually to
determine routes, taking traffic and other factors into consideration.

such as “minimize the distance traveled” or “avoid abnormal uncomfortable state. You reluctantly
highways and toll-roads.” decide to see a doctor.
The real transformation, however, came with the 2) At the doctor’s office, an assistant takes routine
availability of new sensing technology. When GPS measurements of biomarkers, including weight,
became less expensive, more readily available, and temperature, heart rate, and blood pressure. You
sufficiently accurate, it started appearing in not only also fill forms or answer a series of questions
cars but also mobile phones and other devices, offer- related to your family’s health, and your health
ing current location data.11 Once crowd sourcing and history. When the doctor comes, she listens to
other approaches grew powerful enough to determine signals inside your body using a stethoscope
real-time traffic, the navigation space was “cyberne- and asks questions related to your health status
tized,” as shown in Figure 3. Navigation systems then and how long you have been facing the current
became so effective and easy to use that most people problem. She then uses all this multimodal data
stopped using actual maps and started expecting to estimate your current health state and build
their mobile phones or car dashboards to present and update your personal model. If she feels that
step-by-step audio guidance, with automatic rerouting she has insufficient data and information to esti-
based on real-time information. These sophisticated mate the problem that you have then she
cyber-physical-human systems apply geo-spatial and decides to get more data.
temporal “knowledge captures” in maps to continu- 3) This requires diagnostic imaging and pathology
ously update vehicle state using GPS and other sen- tests to obtain more information about your inter-
sors, and present the guidance data using user- nals. These new detailed observations maybe
friendly computer–human interfaces. used in the context of your personal model to esti-
mate your health state.
4) Combining all these multimodal observations
Beyond Symptom-Driven Reactive from different sources and using all medical
Healthcare knowledge, the doctor estimates your current
Currently, most people think of their health only when health state, characterizes it as one of the
they feel very sick and visit a doctor. A successful doc- known disease states and determines its level of
tor’s visit results in the diagnosis of the person’s severity.
health state (possibly a disease) and a prescription for 5) The doctor uses your personal model and current
medicine and any other regimen that will help cure health or disease state, her own medical knowl-
the disease or ease the discomfort. edge about diseases, and general environmental
Let us consider this process: knowledge to recommend corrective actions in
the form of prescriptions and a regimen. These
1) You are not feeling good. Some of your health may involve medication, lifestyle or environmental
state parameters are resulting in an uneasy and changes, or some other treatment.

40
130 ComputingEdge
IEEE MultiMedia March
January-March 20222025
 VISION
VISIONAND
ANDVIEWS
VIEWS

FIGURE 4. Measure, Estimate, Guide, Influence (MEGI) cycle is the norm in current healthcare practice.

6) You try to follow the regimen. Taking medica- 2) Diagnosis is usually Estimation of a disease and
tions is relatively easy; making lifestyle changes its level of severity.
(such as controlling your diet) is often more diffi- 3) The Guide step is to personalize these prescrip-
cult. There is generally no mechanism to verify tions and regimens based on the specific
compliance. The assumption is that you are patient.
interested and able to comply, but there are not 4) Doctors Influence patients to follow the pre-
many good approaches to remind people about scriptions and regimens and request a follow-up
compliance and adherence. visit.
7) Depending on the severity of the disease, the
doctor might want to see you periodically to
repeat steps 2–6 as needed. Depending on the Refactoring Health
severity, the doctor may ask you to visit her In MEGI approach, much of the treatment is to cure
clinic in three months, or a week, or get admitted symptoms, not causes. Important and serious limita-
to hospital immediately. tions of the current healthcare, as noted above are as
follows:

What Does This Process Reveal? 1) Reactive and Episodic: The process starts with
The lifeblood of health is multimodal data. the user when he feels that he has a serious
Data comes in a variety of forms using different problem.
“sensors,” such as personal feelings, vital-sign meas- 2) Post-event Observations: All data collection is
urements, imaging and pathology reports, the doctor’s reactive and takes place in a doctor’s office or in
observations based on the data collected via his or a hospital after the problem is detected.
her audio-visual-tactile senses, and many more. Data 3) Open Loop: Much of the corrective actions are
is converted into actionable information. Doctor’s left to the user and are unmonitored.
knowledge from different fields is used for estimation 4) Symptom-based: Finally, and most importantly, a
and recommendations. user’s health is a continuous process based on
In this process, the first three steps involve data homeostasis and requires corrective actions as
acquisition. Step 4 is state estimation, step 5 is recom- and when needed, much sooner than reaching
mendations from a professional using knowledge of the emergency situation caused by human
the field, step 6 is compliance by the patient and detectable symptoms.
step 7 is follow-up or repetition at regular intervals to
manage further complications. These steps form a These and related detailed observations by health
MEGI cycle: Measure, Estimate, Guide, and Influence, scientists lead to a new emerging paradigm called Pre-
as shown in Figure 4. This cycle has evolved over a dictive, Preventive, Personalized, and Participatory
long period and has stabilized to its current form. (P4) medicine.4 P4 considers that instead of treatment
of an illness, focus should be on prediction and pre-
1) During the Measure step, data is collected from vention of an illness for an individual. P4 can only be
many relevant sources. implemented by collecting all possible relevant health

www.computer.org/computingedge
January-March 2022 IEEE MultiMedia 13141
 VISIONAND
VISION ANDVIEWS
VIEWS

data and analyzing this to detect and prevent diseases guiding the person to take corrective actions based
before they start. This approach is strongly data on all available medical and environmental knowledge.
dependent and is a perpetual process rather than the The cybernetization of the MEGI cycle is natural
current episodic process used in healthcare world- next step in empowering personal health.
wide. More importantly, P4 approach is similar to
cybernetics as presented by Wiener as an underlying
control and communication mechanism in complex Building a Personal Health Navigator
systems applied to arguably the most complex system Let us now map the seven steps of MEGI in a cyber-
known to humans—the human body. P4 approach netics version of the MEGI cycle with the following
may result in augmentation of homeostasis using life- components.
style and other parameters. Measurement: Using a smartphone and augment-
ing wearable sensors with emerging sensors that can
measure many physiological functions, we could iden-
Cybernetizing the MEGI Cycle tify normal bodily parameters to determine the health
Can we “cybernetize” the MEGI cycle—that is, apply state. A smartphone would continuously collect all
the principles of cybernetics to the MEGI cycle? these measurements. This data would measure most
Given how progress in sensors, computing and AI of the information related to the person’s health state.
has disrupted so many fields, it is natural to ask Only under very specific situations would special
whether MEGI can be cybernetized. Personal health- measurements be required. (Steps 1–3 above.)
care is one of the most important areas ripe for disrup- Estimation: All those measurements could be used
tion because of the implication for human quality of to estimate the person’s health using mathematical as
life. well as medical knowledge-based techniques. This
Common sensors (accelerometers, gyroscopes, estimation could indicate proximity to a disease. The
GPS, cameras, and microphones) are now found in goal of estimation would be to measure health state
almost every smartphone, and other measurement without assigning the state to a specific disease or
technologies are appearing, including those that mea- other semantic labels. (Step 4 above.)
sure temperature, perspiration, heart rate, sleep, gal- Guidance: If the user wanted to change his or her
vanic skin resistivity, blood oxygen, blood sugar, and health state, the user would issue a request and the
blood pressure, making cybernetic health possible. system would use medical, environmental, and other
The human body and other biological systems relevant knowledge sources and the personal health
have an intricate play of real-time sensors and actua- state to provide the right guidance in terms of lifestyle
tor functions in homeostasis. In current medical prac- or environmental changes or medications for getting
tice, the process of fixing a disease state and to the desired state. This guidance would be perpetual
restoring the normal health state is based on alerts until the user achieved the desired state and then will
from the person. However, usually these alerts are too function to maintain the state. Guidance at each
late, resulting in serious consequences. Often the moment is very similar to contextual recommendation
alerts remain below the level of consciousness for the systems, which are currently of great interest to AI
person, but they are usually manifested in some bio- researchers.3 (Step 5 above.)
logical markers that can be measured using sensors. Influence and Adherence: Providing guidance
These measurements can be used to estimate the per- alone is not enough. Guidance must be situationally
son’s health state, and, if required, corrective actions actionable and easy to follow. In most cases, mecha-
can be taken to return the person to the desired state. nisms for nudging, incentivizing, and inspiring might
Such actions might include making lifestyle or environ- be required, along with subtle approaches for measur-
mental changes or in some cases taking medication. ing compliance. Using perpetual measurements, it is
These actions would be determined based on biologi- possible to estimate whether the medication and regi-
cal and medical knowledge and a model of the person. ments advised are being followed and how to make
Progress in technology can extend the cybernetic them effective. (Steps 6–7 above.)
loop outside a person’s body using emerging sensors What is equally important is to implement this
and following cybernetic principles of continuous cycle on continuous basis by performing these steps
monitoring for proper communication and control. AI frequently to make sure that the person’s health state
is very useful in enabling this loop, and machine-learn- remains in a safe zone. The above steps are for dealing
ing-based tools facilitate diagnosis and help estimate with nonemergency health situations. The system
health states. Contextual reasoning will help by could function autonomously most of the time but

42
132 ComputingEdge
IEEE MultiMedia March
January-March 20222025
 VISION
VISIONAND
ANDVIEWS
VIEWS

FIGURE 5. Health navigator is a closed-loop system based on continuous measurement. Guidance is based on a knowledge-
based recommendation engine that uses the state of the system. Healthcare professionals may provide a supervisory role in an
otherwise autonomous system.

could guide the user to a nearby resource in an predict the behavior of many cyber physical systems
emergency. by collecting longitudinal perpetual data. Life logs5
Of course, numerous challenges must be addressed were introduced in the computing community to
before we can reap significant benefits from cybernetic understand individual behavior. By collecting longitu-
health. Here, I briefly discuss components that must be dinal perpetual data using mobile phones, wearable
developed to implement a personal health navigation devices, and all other social and environmental sour-
system (see Figure 5). ces, including healthcare organizations, it is possible
Personal Models: Everyone is a unique system that to create a Personal Chronicle (called Personicle) for a
must be modeled to effectively estimate her health person to model her physiological, social, and mental
state and provide precise guidance. The personal behavior in different situations.9 A Personicle is a
model captures how a person reacts to different collection of all person-centered data related to their
stimuli under specific conditions. To model a person, physiological states, environment, lifestyle, food, sleep,
longer-term information comes from her genome mental and emotional state. In a Personicle all this
modulated by the proteome, transcriptome and epige- data is analyzed to detect events and organize it into
nome and reflected in the metabolome. Lifestyle, envi- related events across all aspects of the person. The
ronment, and socio-economic factors also play an idea is to collect all possible data and convert it into
important role in building a model of the person. This events to understand complex causal relationships
model is not static; it changes with age and other life that exist among different aspects of a person based
conditions, so model building is a dynamic process. on the multimodal disparate observations from diverse
Such a model plays a key role in recommending life- sources representing the person.
style, such as food,10 because everyone has a different By applying event mining techniques to a Per-
lifestyle and the system should use her model for sonicle, a model representing the person can be built.6
selecting the right suggestions. This model can then be used to predict health state of
Models are built by collecting historical data and the person in near future based on the estimates of
then analyzing it to build predictive models. These the current state and likely inputs.
models require causality analysis of longitudinal data Health States Estimation: Based on measurements of
about a person. Digital twins have been used to different biomarkers, the health state of a person must

www.computer.org/computingedge
January-March 2022 IEEE MultiMedia 43
133
 VISIONAND
VISION ANDVIEWS
VIEWS

be estimated. In current medical practice, this step is the cases, the information and prescriptions and the com-
same as diagnosing a disease condition. To better under- pliance may also be presented to health professionals
stand health, more quantitative approaches require and family and friends of the person. In general, tech-
defining more health state objectives.8 The health state niques must be developed to influence people and
can be classified as a disease state in the same way as a help them with compliance.
combination of basic color components may be called
pink or purple, but to manage colors more precisely, one
must consider the primary components. A person’s Going Forward
health state characterizes health objectively and can be Disruptive transformations in health have become a
assigned different symbolic labels, like diabetes. To possibility because of the meta-nexus of biology,
implement predictive, preventive, and precise medicine, genetics, multimodal sensors, computing, and mobile
it is important to objectively characterize health. Dis- devices. Progress in these areas opens the possibility
ease-centric estimation looks at the health state through of building personal models and using them in a cyber-
colored glasses and is likely to result in biased decisions. netic framework to develop personal health naviga-
Estimation techniques1 for health states will require tors. In addition to the obvious advantages of a
deep biological knowledge. Formal state-space models closed-loop feedback system, the ability to measure
may emerge over time with associated observability biomarkers and take frequent actions allow this per-
and controllability conditions. In the interim, however, sonal navigation system to function satisfactorily
we might need to build rule-based modeling techniques even when the model and measurements may not be
to implement other aspects of the complete system. perfect.
Situationally Actionable Recommendation: Pre- This major opportunity is created by the smart-
scriptions and specifications of regimens to deal with phone, which has become a surrogate of the person in
diseases are common techniques currently used in all parts of the world. Smartphones could provide
medicine. These techniques are based on available measurements and alerts and suggest certain actions.
medical knowledge and other relevant knowledge The last few years are showing that a major progress
sources, including environmental knowledge. Effec- is being made in different multimodal sensors to mea-
tively, all knowledge sources are organized to find and sure different aspects of the physiological and even
recommend appropriate actions in each situation for mental states of a person. All these sensors are being
each user. This means that very precise personal mod- put in different wearable forms with smart phones and
els, such as a personal food model for taste and health cloud as collection mechanisms.
of a person10 must be built by the system. There is a clear transformation happening in
The knowledge sources may need to be organized health. Health used to be considered mostly related to
at a finer granularity to deal with changing the health biology and associated chemical and physical pro-
state rather than merely getting the person out of the cesses. Now data is the driver of health using cyber-
disease state. Moreover, depending on the perspec- netic processes. Biology, physics, and chemistry are
tive of the designer, different knowledge sources may mechanisms to implement life as we know it. By
be used. Advances in contextual recommendation understanding the role of data in different intertwined
engines will work with this step and will be developed processes, it is possible to augment homeostasis in
to provide health navigation system for everyone time to provide the best quality of life. We can now
based on their personal health model including what design and build devices that will be a sentient mentor
behavior changes may work effectively for the person. to anyone who desires to have one. These devices will
Techniques to Encourage Compliance: Once a rec- be 24/7 personalized healthcare available to even
ommendation about lifestyle and medications is poor people in remote parts of the world.
made, the person is responsible for following up and Furthermore, all the data collected for individuals
complying with the specifications. As is well known, could be shared and aggregated to build powerful
influencing people to follow the recommendations is population models related to diseases to advance
challenging. In most cases, such changes require health sciences. Effectively, we have an opportunity to
behavior change for the person. It is well known that help people better manage their health while advanc-
except in extreme cases, behavior change is one of ing our understanding of diseases and expediting
the most difficult steps in adopting healthy habits. research of potential cures. The availability of such
However, since we are working in closed-loop situa- massive amounts of individual and population-level
tion, the system can use learning techniques to iden- data is unprecedented. We need to build on it, navigat-
tify correct nudging approaches. Moreover, in some ing society toward a healthier future.

44
134 ComputingEdge
IEEE MultiMedia March
January-March 20222025
 VISIONAND
VISION ANDVIEWS
VIEWS

REFERENCES 8. N. Nag and R. Jain, “A navigational approach to health:

Actionable guidance for improved quality of life,” IEEE
1. Y. Bar-Shalom, Estimation With Applications to
Comput., vol. 52, no. 4, pp. 12–20, Apr. 2019.
Tracking and Navigation. Hoboken, NJ, USA: Wiley,
9. H. Oh and R. Jain, “From multimedia logs to personal
2001.
chronicles,” in Proc. 25th ACM Int. Conf. Multimedia,
2. H. Blauw, A. C. van Bon, R. Koops, and J. H. Devries,
2017, pp. 881–889.
“Performance and safety of an integrated bihormonal
10. A. Rostami, V. Pande, N. Nag, V. Wang, and R. Jain,
artificial pancreas for fully automated glucose control
“Personal food model,” in Proc. ACM Multimedia, 2020,
at home, in diabetes,” Obesity, Metab., vol. 18,
pp. 4416–4424.
pp. 671–677, 2016.
11. I. Skog and P. Haendel, “In-car positioning and
3. S. Dehuri, M. R. Patra, B. B. Mishra, and A. K. Jagadev,
navigation technologies – A survey,” IEEE Trans. Intell.
Intelligent Techniques in Recommendation Systems:
Transp. Syst., vol. 10, no. 1, pp. 4–21, Mar. 2009.
Contextual Advancements and New Methods. Hershey,
12. C. Thompson, Smithsonian Magazine, (extracted on
PA, USA: IGI Global, 2012.
Feb. 2022). [Online]. Available: https://www.
4. M. Flores, G. Glusman, K. Broggard, N. D. Price, and
smithsonianmag.com/innovation/brief-history-maps-
L. Hood, “P4 medicine: How systems medicine will
180963685/
transform the healthcare sector and society,”
13. E. Topol, The Creative Destruction of Medicine. New
Personalized Med., vol. 10, no. 6, pp. 565–576, 2013.
York, NY, USA: Basic Books, 2012.
5. C. Gurrin, A. F. Smeaton, and A. R. Doherty,
14. N. Wiener, Cybernetics, Or Control and Communication
“LifeLogging: Personal big data,” in Foundations and
in the Animal and the Machine. Cambridge, MA, USA:
Trends in Information Retrieval. Boston, MA, USA: Now
MIT Press, 1948.
Publishers, vol. 8, no. 1, 2014.
15. Extracted on Feb. 2022. [Online]. Available: https://
6. L. Jalali and R. Jain, Event Mining for Explanatory
bipartisanpolicy.org/report/what-makes-us-healthy-vs-
Modeling. New York, NY, USA: ACM Press, 2021.
what-we-spend-on-being-healthy/
7. J. Markoff, Machines of Loving Grace: The Quest For
Common Grounds Between Humans and Robots. New
York, NY, USA: Harper Collins, 2015. RAMESH JAIN is with the University of California, Irvine, CA,
USA. Contact him at [email protected].

IEEE TRANSACTIONS ON
SUBMIT
TODAY SUSTAINABLE COMPUTING
SUBSCRIBE AND SUBMIT
For more information on paper submission, featured articles, calls for
papers, and subscription links visit: www.computer.org/tsusc

www.computer.org/computingedge
January-March 2022 IEEE MultiMedia 45
135
 EDITOR:Nir
EDITOR: NirKshetri,
Kshetri, [email protected]
[email protected]
This article originally
appeared in
DEPARTMENT: IT
DEPARTMENT: IT ECONOMICS
ECONOMICS
vol. 25, no. 6, 2023

The
The Future
Future of
of Education:
Education: Generative Artificial
Generative Artificial
Intelligence’s Intelligence’s
Collaborative Role
Collaborative
With TeachersRole With Teachers
Nir
NirKshetri
Kshetri , University
, UniversityofofNorth
NorthCarolina
Carolinaat
atGreensboro,
Greensboro,Greensboro,
Greensboro,NC,
NC,27412,
27412,USA
USA

The
Thegrowing
growing consensus
consensus isis that
that generative
generativeartificial
artificialintelligence
intelligence (GAI)
(GAI) is is capable
capable of
of enhancing and enriching teaching and learning experiences. This article
enhancing and enriching teaching and learning experiences. This article examines how
examines how teaching
GAI enhances GAI enhances teaching
and learning byand learning
evaluating itsby evaluating
impact its impact
on primary, on
secondary, and
primary, secondary, and administrative tasks
administrative tasks performed by teachers. performed by teachers.

I
t is increasingly clear that generative artificial intelli- PREPARING A COURSE PLAN
gence (GAI) tools hold the potential to dramatically
GAI can be utilized in the key steps of a course plan-
transform the global education market, which was
ning process, which involves identifying the key fea-
valued at $6 trillion in 2022 and is projected to increase
tures and benefits of the course and selecting subject
to $8 trillion by 2030 (https://tinyurl.com/57x9hmhp).
matter and content as well as methods of teaching
These tools can help enhance and enrich teaching and
and other activities that are important for achieving
learning activities. They could have an even greater
learning outcomes.2 One of the key tasks in this pro-
impact by tailoring assessments and, perhaps, even the
cess is to create a course outline or syllabus. GAI tools
entire learning experience. Nevertheless, it is important
can be of great help in this process.
to note that GAI will not entirely take over the role of
GAI tools such as ChatGPT can help to spotlight
human teachers. Indeed, in all industries, AI augmentation
and highlight a course’s primary features and benefits
(or augmented intelligence), a collaborative approach
and use them as the foundation for creating a compre-
where humans work alongside AI to boost cognitive
hensive course outline or syllabus. Based on these fea-
abilities and performance, has proven highly effective,
tures and benefits, ChatGPT can develop a course
delivering superior results.1
checklist that guides learners in achieving their goals.
GAI may thus give us new tools to augment teach-
Using the checklists, ChatGPT can compile a list of the
ing and learning. This means that GAI supplements con-
necessary course materials and formulate a lesson out-
ventional teaching methods instead of substituting for
line for each module.
them (https://www.binance.com/en/feed/post/1269502?
However, teachers should not leave the preparation
ref=525628006). Teachers play a crucial role in ensuring
of the course plan entirely to GAI tools. It is important
accuracy, relevance and usefulness in instruction,
to enhance a syllabus by incorporating the instructor’s
and learning and assessment. Especially as large lan-
personal touch, emphasizing his or her unique voice
guage models (LLMs) sometimes provide inappropriate
and values and adding specialized sections if needed.
responses or misinterpret, it is crucial to fact-check and
It is also a good idea to share the syllabus with a
verify the information generated by such tools.
colleague to seek their input. In addition, teachers
This article focuses on exploring how GAI can con-
should ask for suggestions and request feedback
tribute to improving the productivity and effectiveness
from students (https://online.ucla.edu/using-chatgpt-
of teaching and learning. We assess the influence of
for-syllabus-refresh/).
GAI on both primary and secondary tasks, along with
the administrative tasks that teachers are obligated to
DELIVERING A COURSE
carry out, as indicated in Table 1.
Course delivery encompasses providing the course to
1520-9202 © 2023 IEEE
its intended audience or recipients as per its design.
Digital Object Identifier 10.1109/MITP.2023.3333070 One aspect of gauging “success” in course delivery is
Date of current version 12 January 2024. the ability to realize the anticipated course outcomes

468 March 2025

IT Professional Published
Published by
by the
the IEEE
IEEEComputer
ComputerSociety
Society 2469-7087/25 © 2025
November/December 2023 IEEE
 ITITECONOMICS
ECONOMICS

TABLE 1. Balancing GAI’s roles with teacher responsibilities.

Task GAI’s Roles Teacher Responsibilities

Preparing a course plan Highlighting the principal attributes and Incorporating personal touch.
advantages of a course. Seeking input and suggestions
Crafting a comprehensive course outline or from a colleague.
syllabus. Receiving feedback from the
Formulating a course checklist to help learners students on the syllabus.
accomplish their goals.
Compiling a list of the required course materials.
Devising a lesson structure for each module.
Delivering a course. Pinpointing the activities and resources essential Combining individual style, unique
for enhancing student engagement and achieving flair, personal innovation, and
learning outcomes. teaching background.
Improving the efficiency of Crafting quizzes and tests based on standardized Examining GAI-generated
time-consuming secondary content becomes more efficient and less time information for correctness and
activities. consuming with GAI tools such as ChatGPT. utility before making it available
to students.
Automating repetitive Utilizing automation to efficiently manage Reviewing GAI-generated content
tasks such as grading recurring responsibilities such as evaluating and making necessary
students’ work. students’ work. adjustments, edits, and revisions
to uphold the integrity of the
grading and feedback process.
Improving personalization Augmenting the personalization of learning and Monitoring student advancements
of learning and assessment; generating custom lessons and and pinpointing areas where
assessment. assignments for individual students. supplementary support may be
essential or where students
demonstrate outstanding
competence.
Performing administrative Increasing efficiency in administrative tasks such Editing GAI-generated e-mails and
tasks efficiently. as writing letters of recommendation and recommendations before sending
communicating with students and parents. them to ensure accuracy as such
tools may not produce flawless
results.

through the implementation process. That is, students AI should be used as a means for augmenting, but
derive the complete intended advantages from the not replacing, the role of teachers in delivering a
course.3 course. Although teachers can utilize ChatGPT’s con-
Commonly cited causes of course delivery failure cepts as a springboard, it is important to incorporate
often revolve around subpar course components and their personal style and creativity, distinctive flair, and
activities as well as insufficient resources.3 GAI tools teaching experience (https://www.weareteachers.com/
can help identify the activities and resources needed chatgpt-for-teachers/).
to promote student engagement. For instance, ChatGPT
serves as an excellent resource for brainstorming inno- IMPROVING THE EFFICIENCY OF
vative class activities that can effectively maintain stu- TIME-CONSUMING SECONDARY
dent engagement. For an engineering course focusing ACTIVITIES SUCH AS PREPARING
on sustainable energy, ChatGPT might recommend an STUDY GUIDES AND CREATING
activity involving the design of a sustainable home QUIZZES
model using CAD software. Similarly, for a history pro- Teaching a course involves a number of subtasks, such
fessor instructing on World War II, ChatGPT could sug- as preparing study guides and designing a sequence of
gest a thought-provoking role-playing exercise in which assignments and tests. GAI can support teachers
students engage in debates about the pivotal events in the preparation of students for assessments and
leading up to the war, fostering a deeper connection has the capability to generate assessments directly
with historical facts (https://tinyurl.com/4uya5mtd). (https://tinyurl.com/3dp4f46j).

www.computer.org/computingedge
November/December 2023 IT Professional 947
 ITECONOMICS
IT ECONOMICS

When it comes to crafting quizzes and tests based chat with ChatGPT. The next step is to copy and paste
on standardized content, ChatGPT stands out as an the student’s submission to receive specific feedback.
invaluable tool, effectively reducing the time burden The educators who used ChatGPT for grading have
on teachers (https://tinyurl.com/2pnpjpp3). To ensure reported that the bot provides numerous insightful and
that ChatGPT formulates questions suitable for stu- valuable comments within the feedback (https://blog.
dents’ skill level and educational background, it is tcea.org/chatgpt-grading).
important to provide context within the prompts. Use- It is crucial to maintain a transparent relationship
ful information to include in the prompts consists of with the students. Sharing the process and the prompts
the students’ grade level, age, the name of the course is a key aspect. Students should also be asked to do
being instructed, and whether the course adheres to a the same and request feedback before they turn in their
standardized curriculum, such as Algebra II or advanced work. The teacher’s role includes providing instruction
placement English literature (https://tinyurl.com/2pnpjpp3). and assessments and delivering precise, meaningful,
ChatGPT’s subscription plan, called ChatGPT Plus, and constructive feedback. The responsibility lies squarely
also makes it possible to upload documents on which with the teacher, necessitating a thorough review of
the quizzes are based (https://beebom.com/how-upload- GAI’s output, and making appropriate adjustments, edits,
document-chatgpt/). Additionally, it’s important to high- and revisions to ensure the reliability of the grading and
light that Generative Pre-trained Transformer (GPT)-3.5 feedback (https://blog.tcea.org/chatgpt-grading).
is limited to a maximum context length, the number of
tokens it can process, of 4096 (roughly 3000 words), IMPROVING PERSONALIZATION
whereas GPT-4 can accommodate contexts that OF LEARNING AND ASSESSMENT
extend up to 32,768 tokens (approximately 25,000 Personalized teaching materials can be beneficial for
words) (https://beebom.com/what-is-chatgpt/). all students, given their individualized learning prefer-
Teachers, however, should not completely rely on ences. GAI tools can help augment the personalization
study guides and quizzes created by GAI tools. They of learning and assessment.6 For instance, ChatGPT
must perform a comprehensive review of questions enables teachers to generate custom lessons and
generated by tools such as ChatGPT. These materials assignments for individual students, taking into account
must be revised and edited as needed to ensure that their learning style and preferences, thereby alleviating
they are clear, concise, and meet the goals of the learn- the necessity for teachers to develop separate lesson
ing process.4 plans.7 As a result, educators may choose to use GAI
tools to develop multiple iterations of their teaching
AUTOMATING REPETITIVE materials to address a wide range of student needs.8
TASKS SUCH AS GRADING ChatGPT is capable of rephrasing responses to suit
STUDENTS’ WORK diverse reading levels. A teacher shared her experience
According to the First Annual Merrimack College of sending 10th grade assignments to ChatGPT and
Teacher Survey,5 the median number of hours teachers instructing the chatbot to restructure the content for a
reported spending on grading and providing feedback sixth grade audience. This adjustment allowed her stu-
of student work during their work weeks was 5 h. GAI dents to comprehend the material and engage effec-
has great potential to save much of this time. tively in classroom discussions.9
The first step in grading student writing is to To facilitate GAI’s use in personalizing learning and
develop a rubric. However, it is also worth noting that assessment and enhancing engagement, teachers can
one can seek ChatGPT’s support in creating the rubric also play the role of a supportive “guide on the side”
as well. To effectively implement GAI in grading stu- instead of being “sage on the stage.” In such a role, the
dent work, it is important to make clear what the teacher can design class activities that can focus
teacher intends to evaluate with as much specificity as on students’ use of GAI (https://spencerauthor.com/
possible. Teachers define the number of quality tiers, aipersonalized/). Delivery of real-time personalized inter-
their designated names, and the corresponding point action is among the many benefits of GAI tools. For
allocations. Additionally, seek ChatGPT’s input on any instance, in a course related to English language compe-
potential criteria you may have overlooked. Assign tencies, students can engage in dynamic conversations
ChatGPT the role of an expert educator and ask the with GAI tools and can explore various dimensions of
bot to make sure that it understands the role. After the English language, including vocabulary, sentence for-
ChatGPT indicates that it is prepared, a teacher can mation, and idiomatic expressions. This interactive prac-
proceed by copying and pasting the rubric into the tice with AI fosters a safe and noncritical setting for

4810 ComputingEdge
IT Professional March
November/December 20232025
 ITIT
ECONOMICS
ECONOMICS

students to refine their English language proficiency Despite the aforementioned benefits of GAI tools
(https://alicekeeler.com/2023/06/10/leveraging-chatgpt- such as ChatGPT, dependence on such tools alone
to-support-students/). may not guarantee student engagement and perfect
Some educators also envision a future scenario learning outcomes. Teachers should thus carefully review
where upcoming textbooks are packaged alongside ChatGPT-generated e-mails before sending them.12
GAI tools that are trained on the book’s content. This
would allow students to engage in a discussion with CONCLUSION
the bot regarding the book’s material, potentially replac-
GAI tools can enhance the efficiency and efficacy of
ing traditional reading. Furthermore, the chatbot could
learning- and teaching-related tasks as well as adminis-
create tailored quizzes to assist students with mastering
trative tasks. GAI’s advantages for educators include
areas where they face difficulties.8
time savings, workload reduction, and the provision
It is important to note, however, that GAI tools
of intuitive and user-friendly technology. These tools
have limited capability to customize teaching for each
can help teachers recognize the specific needs and
student and meet individual needs of a particular stu-
strengths of individual students and engage in activi-
dent.10 The success of students is directly linked to
ties to help them succeed. Students benefit from GAI
teachers’ abilities to understand their unique learning
through an interactive learning experience that caters
needs and preferences. GAI tools such as ChatGPT
to their unique requirements. GAI is thus valuable for
can assist teachers in the identification of such needs
students seeking to increase their likelihood of pros-
and individual student strengths by examining their
pering in their studies. One of the major benefits of
performance data. In this way, educators can utilize
ChatGPT is that it offers an interactive and individually
these tools to identify consistent patterns and trends
tailored learning experience, which can enhance learner
and monitor the progress of students (https://tinyurl.
engagement.
com/3da9th5p). Although these tools can be helpful in
identifying areas that may necessitate additional sup-
port or areas in which students excel, in many cases, REFERENCES
teachers are responsible for taking appropriate actions 1. “Top trends on the Gartner hype cycle for artificial
to help students succeed. intelligence, 2019,” Gartner.com. Accessed: Oct. 3,
2023. [Online]. Available: https://www.gartner.com/
PERFORMING ADMINISTRATIVE smarterwithgartner/top-trends-on-the-gartner-hype-
TASKS EFFICIENTLY cycle-for-artificial-intelligence-2019
GAI can also help effectively manage day-to-day 2. T. A. Wikle and T. D. Fagin, “GIS course planning: A
administrative tasks, such as communicating with stu- comparison of syllabi at US college and universities,”
dents and parents7 and writing letters of recommenda- Trans. GIS, vol. 18, no. 4, pp. 574–585, 2014, doi: 10.
tion. Educators can utilize ChatGPT and other LLMs 1111/tgis.12048.
to create messages to students and parents concern- 3. J. Mouton and L. Wildschut, “Service-learning in
ing upcoming assignments, exams, and other crucial South Africa: Lessons learnt through systematic
details. These tools are especially useful for creating evaluation,” Acta Academia, vol. 3, no. sup-3, pp.
e-mails that address sensitive and emotionally chal- 116–150, 2005, doi: 10.10520/EJC15193.
lenging subjects. For instance, by offering a simple cue, 4. “Advice from ChatGPT on how to best use ChatGPT
ChatGPT can produce an e-mail directed to the parent to create quiz questions,” UCLA, Los Angeles, CA,
of a fifth grade student struggling in his or her math USA, 2023. Accessed: Oct. 3, 2023. [Online]. Available:
class (https://tinyurl.com/3dp4f46j). It also offers assis- https://online.ucla.edu/chatgpt-generated-quiz-
tance with organizing communication timelines and instructions/
analyzing responses from various sources, including 5. H. Kurtz, “A profession in crisis: Findings from a
parent–teacher meetings.7 national teacher survey,” EdWeek Res. Center, Apr. 14,
The results of a recent social media poll showed 2022. Accessed: Oct. 3, 2023. [Online]. https://www.
that more than 33% of high school teachers write more edweek.org/research-center/reports/teaching-
than 10 letters of recommendation for students, on profession-in-crisis-national-teacher-survey
average, each year.11 ChatGPT is also capable of com- 6. I. Pittawala, “Is ChatGPT a threat to education?” UC
posing student recommendation letters, utilizing just a Riverside News, Jan. 24, 2023. Accessed: Oct. 3, 2023.
few descriptors provided by a teacher (https://tinyurl. [Online]. Available: https://news.ucr.edu/articles/2023/
com/tdxwamyu). 01/24/chatgpt-threat-education

www.computer.org/computingedge
November/December 2023 IT Professional 1149
 IT ECONOMICS
IT ECONOMICS

7. K. Novak, “AI Is an Ally: Saving teachers time with Understanding the potential benefits of ChatGPT in
ChatGPT,” The Journal, May 11, 2023. Accessed: Oct. promoting teaching and learning,” SSRN, Jan. 25,
3, 2023. [Online]. Available: https://thejournal.com/ 2023. Accessed: Oct. 3, 2023, doi: 10.2139/ssrn.
articles/2023/05/11/ai-is-an-ally-saving-teachers-time- 4337484. [Online]. Available: https://ssrn.com/
with-chatgpt.aspx#::text=ChatGPT%20can%20help% abstract=4337484
20teachers%20save,rubrics%20or%20other% 11. H. Hardison, “Knock out those letters of
20grading%20schema recommendation: pro tips from teachers,” Educ.
8. W. D. Heaven, “ChatGPT is going to change Week, Nov. 14, 2022. Accessed: Oct. 3, 2023. [Online].
education, not destroy it,” MIT Technol. Rev., Apr. 6, Available: https://www.edweek.org/teaching-learning/
2023. Accessed: Oct. 3, 2023. [Online]. Available: knock-out-those-letters-of-recommendation-pro-tips-
https://www.technologyreview.com/2023/04/06/ from-teachers/2022/11
1071059/chatgpt-change-not-destroy-education- 12. K. Paterson. “6 ChatGPT prompts that will generate
openai/ great sales emails.” Zapier. Accessed: Oct. 3, 2023.
9. A. Blose, “As ChatGPT enters the classroom, teachers [Online]. Available: https://zapier.com/blog/chatgpt-
weigh pros and cons,” Nea Today, Apr. 12, 2023. sales-emails/
Accessed: Oct. 3, 2023. [Online]. Available: https://
www.nea.org/nea-today/all-news-articles/chatgpt- NIR KSHETRI is a professor in the Bryan School of Business
enters-classroom-teachers-weigh-pros-and-cons and Economics, the University of North Carolina at Greens-
10. D. Baidoo-Anu and L. Owusu Ansah, “Education in boro, Greensboro, NC, 27412, USA, and IT Professional’s IT
the era of generative artificial intelligence (AI): Economics editor. Contact him at [email protected].

Call for Papers

Publish your work in the IEEE Computer
Society’s flagship journal, IEEE Transactions
on Computers. The journal seeks papers on
everything from computer architecture and
software systems to machine learning and
quantum computing.

Learn about calls for papers

and submission details at
www.computer.org/tc

tc-generic-cfp-half-Sept2024.indd 1 9/30/24 6:38 P

5012 ComputingEdge
IT Professional March
November/December 20232025
Get Published in the IEEE Open
Journal of the Computer Society

Get more citations by

publishing with the IEEE
Open Journal of the
Computer Society

Your research on computing and

informational technology will
benefit from 5 million unique
monthly users of the IEEE Xplore®
Digital Library. Plus, this
journal is fully open and
compliant with funder
mandates, including Plan S.

Submit your paper today!

Visit www.computer.org/oj to learn more.
 EDITORS:Ryan
EDITORS: Ryan Cooper,
Cooper, [email protected]
[email protected]
Kari
Kari Jordan, Jordan, [email protected]
[email protected]
This article originally
appeared in

DEPARTMENT:
DEPARTMENT: EDUCATION
EDUCATION
vol. 25, no. 2, 2023

Leverage
Leverage Biology Biology to to LearnLearn Rapidly From
Rapidly
MistakesFrom Without Mistakes Feeling WithoutLike a Failure
Feeling Like a Failure
Lauren E. Margulieux , Georgia State University, Atlanta, GA, 30302, USA
Lauren
James E. Margulieux
Prather , Georgia
, Abilene State
Christian University,
University, Atlanta,
Abilene, TX,GA, 30302,
79699, USAUSA
James Prather
Masoumeh , Abilene
Rahimi andChristian University,
Gozde Cetin Uzun Abilene,
, GeorgiaTX, 79699,
State USA Atlanta, GA, 30302, USA
University,
Masoumeh Rahimi and Gozde Cetin Uzun , Georgia State University, Atlanta, GA, 30302, USA

Our biology affects how we interact with the world, including how we learn
new knowledge and respond to challenges. This article explores the impact of
Our biology affects
neurochemicals how
in our we interact
brain withand
on learning theexplains
world, including how weour
how to leverage learn new to
biology
knowledge and respond to challenges. This article explores the impact of neurochemicals
improve education and problem solving, focusing on computing education. Within this
in our brain on learning and explains how to leverage our biology to improve education
context, the article particularly examines the role of failure while learning. Learning,
and problem solving, focusing on computing education. Within this context, the article
especially in technical fields, includes making errors on the path to success. Although
particularly examines the role of failure while learning. Learning, especially in technical
these includes
fields, errors trigger the errors
making necessary neurochemical
on the conditions
path to success. for rapid
Although theselearning, these
errors trigger
failures can also be demotivating. To gain the benefits of failure while mitigating
the necessary neurochemical conditions for rapid learning, these failures can also its be
negative consequences,
demotivating. To gain thethis articleof
benefits recommends
failure whileevidence-based behavioral
mitigating its negative strategies
consequences,
for making
this the best out ofevidence-based
article recommends failing while learning and designing
behavioral strategiesforfor
failure in learning
making the best
environments.
out of failing while learning and designing for failure in learning environments.

T
his article explores the profound impact of neu- particularly demotivating when a learner is not inher-
rochemicals on learning and how to leverage ently interested in computing. To support students,
biology to support education and training. Hor- this article discusses evidence-based behavioral tools
mones like cortisol, dopamine, and serotonin play a piv- that learners can use to benefit from failure while mini-
otal role in shaping behavior and cognitive processes, mizing negative consequences. The causal connec-
making them crucial factors in optimizing learning. In tions between neurochemicals and human behaviors
particular, this article examines the role of making are well studied in neuroscience, and this article cites
errors and failing as a catalyst for learning. Failure trig- the most robust findings from this research related to
gers the neurochemical conditions necessary for neu- learning (additional references can be found in the sup-
roplasticity, enabling the brain to adapt and grow after plemental material, available at http://10.1109/MCSE.
making mistakes.1 Despite this benefit for learning, fail- 2023.3297750). Although these connections are causal,
ure is also an unpleasant experience that people typi- they are not deterministic. Thus, this article does not sup-
cally avoid. Understanding the neurochemicals at play port biological determinism and instead advocates for
allows learners and educators to use strategies that integrating biological tools with other tools. By embracing
mitigate the demotivational consequences of failure. this holistic approach, we pave the way for advancements
The primary audience for this article is those learn- in education that harness the power of our biology.
ing or teaching computing, especially for students
who do not wish to become computer scientists. Com-
HOW FAILURE IMPROVES
puting is a design-focused field that requires a lot of
NEUROPLASTICITY
failure on the path to success. These failures can be
Neuroplasticity, the brain’s ability to learn and adapt,
1521-9615 © 2023 IEEE
plays a crucial role in our ability to acquire new knowl-
Digital Object Identifier 10.1109/MCSE.2023.3297750 edge and skills. Although children effortlessly engage
Date of current version 23 August 2023. in passive neuroplasticity, this natural learning process

52
44 March 2025 in Science & Engineering
Computing Published by the IEEE by
Published Computer
the IEEESociety 
Computer Society 2469-7087/25 © 2025
March/April 2023 IEEE
 EDUCATION
EDUCATION

diminishes over time and ceases completely around agitation. However, our capacity to produce these mole-
age 25. However, adults can still achieve comparable cules diminishes after 90 min of sustained attention,
learning outcomes to adolescents through active neu- leading to a decline in learning efficiency. Moreover,
roplasticity, which requires concentrated effort and high levels of epinephrine trigger a quitting mechanism
incremental learning.2 Activating neuroplasticity is pos- that compels us to stop, impairing attentional control.
sible by increasing the presence of epinephrine, a neu- Learners’ experience of different epinephrine levels
rochemical in the brain that is released in response to and the likelihood of triggering the quitting mechanism
facing challenges or stress, like failure. Research shows depend on individual tolerance. Tolerance for epineph-
that epinephrine enhances memory by triggering a cas- rine can be increased through training, where individu-
cade of events. Epinephrine increases amygdala activa- als expose themselves to safe but challenging stressors
tion, which in turn triggers activation of cholinergic and dissociate stress from feelings of danger, such as
neurons that improve memory formation.3 Therefore, tackling a difficult programming problem.8 Developing
experiencing failure during the learning process can a high tolerance for epinephrine will benefit learners
effectively trigger this cascade of events and enhance who frequently face situations involving uncertainty or
neuroplasticity.3 frequent errors, such as solving computing problems.
Repetitions and failures are often part of the learn- To maintain optimal levels of epinephrine, learners
ing process, including in computing education. Various should limit learning bouts to 90 min with at least
instructional paradigms, such as productive failure, 20 min of rest between bouts. Because epinephrine is
investigate the role of failure in learning.4 Productive a short-lived molecule with a half-life of only a couple
failure involves asking students to attempt a task of minutes, learners can also use behavioral tools to
before providing instruction on how to achieve it. After change their current level of epinephrine. Although
spending at least 30 min failing to accomplish the task, there are many possible behavioral tools, some of the
students are then given instruction on the correct most effective are breathing techniques. To decrease
solution.4 Meta-analyses of instructional design studies epinephrine, learners can use exhale-dominant breath-
have shown that productive failure before instruction ing, where individuals take deep breaths through the
has a significant, strong effect on learning outcomes.5 nose and exhale slowly over several seconds. Exhale-
In fact, the larger quantity of incorrect solutions learn- dominant breathing artificially lowers the heart rate,
ers create, the better they perform on later tasks.4 Addi- producing a state of relaxation and decreasing alert-
tionally, personal failure is found to be more impactful ness. Conversely, to increase alertness, such as for an
than vicarious failure,4 likely because it elicits the affec- early class or during a late-night work session, learners
tive component that triggers neuroplasticity-promoting can use the opposite technique. Inhale-dominant breath-
neurochemicals. ing involves deep inhalation through the nose followed
Productive failure and other instructional para- by quick exhalation through the mouth, stimulating
digms based on failure have improved learning in vari- the production of epinephrine and enhancing alertness.
ous domains, including creative and applied fields such These breathing techniques and other behavioral tools
as design, business, and computing. Although failure is can be employed to modulate alertness levels by chang-
often embraced in these professions, formal education ing epinephrine levels in different learning contexts.
rarely teaches the skills of dealing with uncertainty
while problem solving and learning from and coping Improve Focus With Acetylcholine
with failure.6 However, intentional applications of failure Neuroplasticity is triggered by epinephrine and ends
in computing education have shown positive results, with the activation of cholinergic neurons, which release
particularly in tasks like program debugging, where novi- acetylcholine. Acetylcholine is crucial for learning, mem-
ces frequently make errors.7 ory formation, and strengthening memories, and block-
ing its receptors inhibits memory formation. Alzheimer’s
Balance Alertness and Calmness patients, who have cholinergic neuron loss, experience
With Epinephrine memory difficulties, while cholinergic stimulation improves
The neurochemical epinephrine is produced in response memory in older adults. Attention (i.e., epinephrine)
to high-stress situations that trigger a spike in epineph- and focus (i.e., acetylcholine) are distinct biological
rine, and to situations that require sustained attention processes, with attention being generated first before
or cognitive strain, which produce lower, consistent lev- focus can be directed.9 Initially, there can be a brief
els. Lower levels of epinephrine are better for learning period at the start of a task when focusing attention is
as they create a feeling of calm alertness rather than challenging due to the influence of epinephrine, which

www.computer.org/computingedge
March/April 2023 Computing in Science & Engineering 53
45
EDUCATION
EDUCATION

promotes attention but also induces movement and difficult tasks, but it takes time to establish these con-
undifferentiated awareness.10 When transitioning through nections, which is why habits take at least 18 days to
failure or uncertainty, conscious effort may be required form. The dopamine system is also influenced by
to maintain focus during the increase in epinephrine reward prediction error, where the release of dopamine
levels before acetylcholine sharpens focus. Although is based on the disparity between expected and actual
there are limited behavioral recommendations for opti- rewards. This system helps calibrate our behavior and
mizing acetylcholine levels, maintaining visual focus encourages us to pursue achievable goals while dis-
on a single point, such as the tip of a pen, for 30–60 s couraging pursuits that are less likely to succeed. To
can help generate sufficient acetylcholine to direct maintain motivation in the face of failure, reframing
attention toward the learning task.10 large projects as achievable short-term goals and
focusing on actions rather than outcomes can help
MAINTAIN MOTIVATION mitigate the negative impact of reward prediction error
THROUGH FAILURE and sustain long-term motivation.11,12 For example, set-
Although failure can enhance learning by activating ting a goal to study or practice problem solving each
neuroplasticity, it is important to avoid excessive day for an hour will result in better motivation than set-
failure, which can be unproductive. The same neuro- ting a goal to attain a certification by a particular date
chemicals that promote neuroplasticity also cause or to complete weekly assignments.
frustration and demotivation when making errors. If In addition to framing and focusing on achievable
learners fail so much that these emotions lead to giving short-term goals, other features of the dopamine sys-
tem can help maximize our benefit from it, such as the
up, learners experience the negative aspects of failure
following:
without reaping its benefits.
The effect of failure on motivation is influenced by
Dopamine tends to be higher in the morning. In
how we frame it psychologically. When we view failures
addition, dopamine is the precursor to epineph-
as being on the right track, we are more likely to per-
rine, which improves neuroplasticity. Thus, many
sist, whereas perceiving them as setbacks often leads
people find doing concentrated work easier in
to quitting. This framing process occurs in the prefron-
the morning.
tal cortex, which is larger in humans compared to other
Avoid other activities that require little effort to
animals. Framing allows humans to pursue long-term activate the second part of the dopamine sys-
goals by leveraging a hormone that normally rewards tem, especially in the morning. These activities,
short-term goals: dopamine.11 such as using social media or eating high-fat
Although pop culture treats dopamine as a mole- and high-sugar treats, can offset the dopamine
cule of reward for achievement, this is only half the system and decrease motivation all day.
story. The dopamine system consists of two parts: one Caffeine increases dopamine, producing feel-
related to the expectation of a reward, driving us to ings of motivation. In addition, it can increase
overcome challenges and pursue goals we value, and the number of dopamine receptors and might
the other associated with the reward for achieving have a protective effect on dopaminergic neu-
those goals.11 The first part is what we should harness rons for long-term health.
to drive motivation because it rewards the pursuit and Exposure to bright light, such as screens, in the
anticipation of valuable things.11 These things include middle of the night (a few hours after sunset to
food when we are hungry, water when we are thirsty, or a couple of hours before sunrise) blunts dopa-
other things we enjoy, like coffee, chocolate, or novel mine release long term, reducing motivation
information. Although the dopamine system naturally during the day.
motivates us subconsciously, humans can exert con-
scious control over it through the prefrontal cortex,
allowing us to frame and redirect it toward abstract IMPROVE TOLERANCE
pursuits,12 such as learning computing. FOR FAILURE
Learning computing, or applying computing for that
Manage Dopamine to Manage matter, involves enduring failure on the path to suc-
Motivation cess, making a high tolerance for failure crucial for
The prefrontal cortex allows us to perceive challenging learners of computing. Of course, tolerance for failure is
pursuits as pleasurable by framing them as valuable.12 also valuable in other technical fields, like engineering
This framing enables us to receive dopamine from and science. Tolerance for failure varies among

54
46 ComputingEdge
Computing in Science & Engineering March
March/April 20232025
 EDUCATION
EDUCATION

individuals and can be increased through training or another effective way to boost serotonin production
decreased through negative experiences. Understand- by increasing the amount of tryptophan that reaches
ing tolerance for failure can shed light on how students the brain, and thus is used for mood regulation.14
experience failure, why some students thrive or wither Finally, although gratitude practices are commonly
in different environments, and how to support learners’ associated with serotonin increase, it is now believed
success while learning computing. that they primarily influence our perception of events
This tolerance is associated with the activity level rather than directly impact serotonin levels. Framing
in the amygdala, the same brain region responsible for events through appreciation rather than evaluation or
activating cholinergic neurons to enable neuroplastic- comparison can lead to more positive experiences, and
ity.3 The amygdala’s level of activation affects the bal- expressing gratitude can also influence how others
ance between positive learning benefits and negative interact with us.15 Any of these four tools can be used
emotions, with experiences that cause high amygdala to increase serotonin, which is likely to benefit more
activation leading to increased risk aversion and fear of than just learning of computing.
failure.13 Hormones like serotonin buffer these effects
by inhibiting amygdala activity, thereby increasing risk Maintain Optimal Levels of Cortisol
tolerance and reducing fear of failure, while other hor- Although we commonly think of stress as unilaterally
mones like cortisol increase activation of the amygdala harmful, stress also helps us maintain alertness and
and decrease tolerance for failure. achieve goals. For example, our body naturally produ-
ces a cyclical level of cortisol throughout the day to
Increase Tolerance for Failure help us wake up in the morning, be alert throughout
With Serotonin the day, and then relax at night. However, many people
Serotonin, often known for its dysregulation in depres- face constant exposure to stress, which disrupts this
sion, is associated with feelings of reward and satisfac- cycle and has long-term negative consequences. Part
tion, fostering gratitude and contentment.14 Serotonin of this dysregulation is that cortisol increases activity
levels naturally rise throughout the day, peaking in the in the amygdala, increasing our perception of how
afternoon and declining after sunset when they are stressful events are. As a result, chronic stress can cre-
converted into melatonin to help us sleep. The produc- ate a self-sustaining feedback loop in which the amyg-
tion of serotonin is influenced by sunlight, which explains dala is always active at a high level.
the link between seasonal depression and reduced sun- Reducing unhealthy cortisol levels can be achieved
light exposure in winter.14 through various behavioral approaches, with exercise
In terms of tolerance for failure, higher serotonin and meditation being particularly effective. Although
levels are linked to persistence in challenging situa- exercise may temporarily increase cortisol during activ-
tions and a willingness to embrace new experiences. ity, low- and moderate-intensity exercise has a long-term
Behavioral interventions that boost serotonin, such as cortisol-reducing effect.16 Bursts of exercise immediately
exercise, promote less defensive behaviors and a pref- after a stressful event can also help clear cortisol from
erence for larger delayed rewards. Serotonin’s effects the bloodstream.16 Further, exercise offers additional
are mediated through the prefrontal cortex and amyg- long-term benefits, including enhanced prefrontal activ-
dala, reducing the perception of threats and decreas- ity, which is relevant to framing, motivation, and toler-
ing the negative emotions associated with novel or ance for failure. Research supports the positive impact
challenging experiences. of exercise interventions on education, including studies
To maintain a healthy level of serotonin and sup- involving computing students. Fit breaks, short periods
port tolerance for failure, learners can use several free of physical activity inserted into lecture times, have
behavioral tools. Sunlight exposure is a significant fac- been found to improve student well-being, stress levels,
tor, and modern individuals who spend most of their retention, and academic performance.17
time indoors typically experience lower levels of sun- Meditation, on the other hand, provides a different
light than are optimal.14 Increasing exposure to bright approach to cortisol management. By altering how
light, especially in the morning, can help raise serotonin stressful events are perceived and helping individuals
levels.14 Early exposure to bright light is beneficial even separate stressors from their reactions, meditation inter-
for people who prefer to work later in the day. Diet also rupts the fight-or-flight response and regulates the
plays a role, as serotonin is produced from tryptophan. intensity of stress responses. Several reviews highlight
Consuming tryptophan-rich foods, particularly earlier the benefits of meditation in education, demonstrating
in the day, supports serotonin synthesis. Exercise is improvements in student well-being and academic

www.computer.org/computingedge
March/April 2023 Computing in Science & Engineering 55
47
EDUCATION
EDUCATION

achievement. In one randomized experiment involv- REFERENCES

ing college students, just 13 min of daily meditation 1. A. Darabi, T. L. Arrington, and E. Sayilir, “Learning from
resulted in decreased cortisol levels and feelings of failure: A meta-analysis of the empirical studies,”
stress, alongside increased attention and memory.18 Educational Technol. Res. Develop., vol. 66, no. 5,
Both exercise and meditation present valuable strate- pp. 1101–1118, Oct. 2018, doi: 10.1007/s11423-018-9579-9.
gies for cortisol management and have demonstrated 2. B. A. Linkenhoker and E. I. Knudsen, “Incremental
positive effects on various aspects of education. training increases the plasticity of the auditory space
map in adult barn owls,” Nature, vol. 419, no. 6904,
CONCLUSION pp. 293–296, Sep. 2002, doi: 10.1038/nature01002.
Learners’ experience of failure involves an inter- 3. L. Cahill and J. L. McGaugh, “Mechanisms of
connected set of neurochemical systems that affect emotional arousal and lasting declarative memory,”
neuroplasticity, motivation, and tolerance for failure. Trends Neurosciences, vol. 21, no. 7, pp. 294–299,
Although failure activates neuroplasticity to facilitate Jul. 1998, doi: 10.1016/S0166-2236(97)01214-9.
learning, it can also lead to task avoidance if used inap- 4. M. Kapur, “Productive failure in learning math,”
propriately. The authors aimed to provide a new per- Cogn. Sci., vol. 38, no. 5, pp. 1008–1022, Jun. 2014,
spective for understanding and designing factors in doi: 10.1111/cogs.12107.
computing education related to failure, emphasizing 5. T. Sinha and M. Kapur, “When problem solving
the importance of framing of failure and fear of failure followed by instruction works: Evidence for
and their biological consequences. productive failure,” Rev. Educational Res., vol. 91,
Educators can use these behavioral tools to sup- no. 5, pp. 761–798, Oct. 2021, doi: 10.3102/
port their students through failure. Much like with 00346543211019105.
meta-cognitive strategies, learners often benefit from 6. S. Dobson and B. Walmsley, “Fail fast, fail often …
being reminded of strategies they can use to improve But don’t fail this course! Business and enterprise
their learning. For example, educators can guide stu- education through the lens of theatre and the
dents to set small, achievable goals throughout the creative arts,” Industry Higher Educ., vol. 35, no. 4,
semester (e.g., study class notes for 1 h after class) pp. 336–346, Aug. 2021, doi: 10.1177/0950422220955071.
or toward large projects to improve their motivation 7. J. Prather et al., “On novices’ interaction with
through dopamine. Further, offering fit breaks during compiler error messages: A human factors approach,”
class or encouraging students to get some sunlight or in Proc. ACM Conf. Int. Comput. Educ. Res. (ICER),

exercise for the sake of their studies might improve New York, NY, USA: Association for Computing
Machinery, 2017, pp. 74–82, doi: 10.1145/3105726.
their tolerance for failure. The authors suggest using
3106169.
these factors alongside other instructional tools in
8. D. Spiegel, “Dissociation and hypnosis in post-
computing education, such as peer interactions.
traumatic stress disorders,” J. Traumatic Stress, vol. 1,
The authors acknowledge that learners’ relationship
no. 1, pp. 17–33, Jan. 1988, doi: 10.1002/jts.2490010104.
with failure cannot be expected to change instantly as
9. A. M. Himmelheber, M. Sarter, and J. P. Bruno, “The
these biological systems have served life-saving func-
effects of manipulations of attentional demand on
tions throughout our evolution. However, through train-
cortical acetylcholine release,” Cogn. Brain Res.,
ing and reframing, individuals can become better at
vol. 12, no. 3, pp. 353–370, Dec. 2001, doi: 10.1016/
failing over time, reaping the rewards of neuroplasticity
S0926-6410(01)00064-7.
while experiencing fewer negative consequences. The
10. A. Huberman, “Teach and learn better with a
hope is that these tools will be utilized in conjunction
‘neuroplasticity super protocol,’” Neural Netw.,
with other approaches to enhance computing educa-
Oct. 2021. [Online]. Available: https://hubermanlab.
tion and support students in their learning journey. com/teach-and-learn-better-with-a-neuroplasticity-
super-protocol/
ACKNOWLEDGMENTS 11. D. Z. Lieberman and M. E. Long, The Molecule of
This material is based upon work supported by the More: How a Single Chemical in Your Brain Drives
National Science Foundation (NSF) under Grant 1941642. Love, Sex, and Creativity–And Will Determine the
Any opinions, findings, and conclusions or recommen- Fate of the Human Race. Dallas, TX, USA: BenBella
dations expressed in this material are those of the Books, 2018.
author(s) and do not necessarily reflect the views of 12. A. Looby, L. Zimmerman, and N. R. Livingston,
the NSF. This article has supplementary downloadable “Expectation for stimulant type modifies caffeine’s
material available at http://10.1109/MCSE.2023.3297750. effects on mood and cognition among college

56
48 ComputingEdge
Computing in Science & Engineering March
March/April 20232025
 EDUCATION
EDUCATION

students,” Exp. Clin. Psychopharmacol., vol. 30, no. 5, memory, mood, and emotional regulation in non-
pp. 525–535, Mar. 2021, doi: 10.1037/pha0000448. experienced meditators,” Behavioural Brain Res., vol. 356,
13. P. Xu et al., “Amygdala–Prefrontal connectivity pp. 208–220, Jan. 2019, doi: 10.1016/j.bbr.2018.08.023.
modulates loss aversion bias in anxious individuals,”
Neuroimage, vol. 218, Sep. 2020, Art. no. 116957, LAUREN E. MARGULIEUX is an associate professor in the
doi: 10.1016/j.neuroimage.2020.116957. Department of Learning Sciences, Georgia State University,
14. S. N. Young, “How to increase serotonin in the human Atlanta, GA, 30302, USA, and founding director of the
brain without drugs,” J. Psychiatry Neuroscience, Snap Inc. Center for Computing in Teacher Education.
vol. 32, no. 6, Nov. 2007, Art. no. 394.
Contact her at [email protected].
15. S. Cox, “The power of gratitude,” Nursing Manage.,
vol. 49, no. 4, Apr. 2018, Art. no. 56, doi: 10.1097/01. JAMES PRATHER is an associate professor of computer sci-
NUMA.0000531176.85470.42.
ence at Abilene Christian University, Abilene, TX, 79699,
16. W. A. Suzuki, “How body affects brain,” Cell
USA. Contact him at [email protected].
Metabolism, vol. 24, no. 2, pp. 192–193, 2016,
doi: 10.1016/j.cmet.2016.07.022.
MASOUMEH RAHIMI is a Ph.D. student in the Instructional
17. A. Koulanova, A. Maharaj, B. Harrington, and J. Dere,
Technology program at Georgia State University, Atlanta,
“Fit-breaks: Incorporating physical activity breaks in
GA, 30302, USA. Contact her at [email protected].
introductory CS lectures,” in Proc. 23rd Annu. ACM
Conf. Innov. Technol. Comput. Sci. Educ., 2018,
pp. 260–265, doi: 10.1145/3197091.3197115. GOZDE CETIN UZUN is a Ph.D. student in the Instructional
18. J. C. Basso, A. McHale, V. Ende, D. J. Oberlin, and Technology program at Georgia State University, Atlanta,
W. A. Suzuki, “Brief, daily meditation enhances attention, GA, 30302, USA. Contact her at [email protected].

IEEE Annals of the History of Computing publishes work

covering the broad history of computer technology, including
Volume 46 Number 2
technical, economic, political, social, cultural, institutional,
and material aspects of computing. Featuring scholarly articles
by historians, computer scientists, and interdisciplinary
scholars in fields such as media studies and science and
technology studies, as well as firsthand accounts, Annals is the
Logistical Histories
primary scholarly publication for recording, analyzing, and
of Computing
debating the history of computing.

www.computer.org/annals

AN.general_hHalf_Sept2024.indd 1 9/30/24 5:50 PM

www.computer.org/computingedge
March/April 2023 Computing in Science & Engineering 57
49
 05.23 08.24 07.24 05.24 04.24
Computer

Computer

Computer

Computer
Trustworthy AI—Part II
MAY 2023

AUGUST 2024

MAY 2024

APRIL 2024
Venture Scientists and
Technology Entrepreneurship

VENTURE SCIENTISTS AND TECHNOLOGY ENTREPRENEURSHIP

TECHNOLOGY PREDICTIONS
TRUSTWORTHY AI—PART II

DEPENDABLE COMPUTING
CS Election Candidates, p. 136
IEEE Division V Delegate/
Director Candidates, p. 151
Volume 56 Number 5

Volume 57 Number 8

Volume 57 Number 5

Volume 57 Number 4
vol. 57 no.8 www.computer.org/computer vol. 57 no.7 www.computer.org/computer
vol. 56 no.5 www.computer.org/computer vol. 57 no.5 www.computer.org/computer vol. 57 no.4 www.computer.org/computer

02.24 12.23 11.23 10.23 09.23

Computer
Computer

Computer

Computer

Computer
Ethics and Computing COMPUTER ENGINEERING STATE OF TELEMEDICINE—PART II STATE OF TELEMEDICINE—PART I

SEPTEMBER 2023
FEBRUARY 2024

DECEMBER 2023

NOVEMBER 2023

OCTOBER 2023
EDUCATION
Emerging
Disruptive
Technologies
EMERGING DISRUPTIVE TECHNOLOGIES

COMPUTER ENGINEERING EDUCATION

STATE OF TELEMEDICINE—PART I
STATE OF TELEMEDICINE—PART II
ETHICS AND COMPUTING

CS PRESIDENT’S MESSAGE 14
IEEE CS AWARDS 18
Volume 56 Number 12

Volume 56 Number 9
Volume 57 Number 2

Volume 56 Number 10
Volume 56 Number 11

vol. 57 no.2 www.computer.org/computer vol. 56 no.12 www.computer.org/computer vol. 56 no.11 www.computer.org/computer vol. 56 no.10 www.computer.org/computer vol. 56 no.9 www.computer.org/computer

CALL FOR SPECIAL ISSUE

PROPOSALS
Computer solicits special issue proposals from leaders and experts within a
broad range of computing communities. Proposed themes/issues should
address important and timely topics that will be of broad interest to Computer’s
readership. Special issues are an essential feature of Computer, as they
deliver compelling research insights and perspectives on new and established
technologies and computing strategies.
Please send us your high-quality proposals for the 2025–2026 editorial calendar.
Of particular interest are proposals centered on:
• 3D printing • Dis/Misinformation
• Robotics • Legacy software
• LLMs • Microelectronics
• AI safety

Proposal guidelines are available at:

www.computer.org/csdl/magazine/co/write-for-us/15911
IEEE COMPUTER SOCIETY D&I FUND

Drive Diversity
& Inclusion in
Computing

Supporting projects
and programs that
positively impact
diversity, equity, and
inclusion throughout
the computing DONATE TODAY!
community.
 Conference Calendar

I EEE Computer Society conferences are valuable forums for learning on broad and dynamically shifting top-
ics from within the computing profession. With over 200 conferences featuring leading experts and thought
leaders, we have an event that is right for you. Questions? Contact [email protected].

APRIL Custom Computing Machines), • ICFEC (IEEE Int’l Conf. on Fog

9 April Fayetteville, USA and Edge Computing) Tromsø,
• SaTML (IEEE Conf. on Secure • MOST (IEEE Int’l Conf. on Mobil- Norway
and Trust wor thy Machine ity, Operations, Services and 26 May
Learning), Copenhagen, Technologies), Newark, USA • FG (IEEE Int’l Conf. on Auto-
Denmark 5 May matic Face and Gesture Recog-
15 April • CAI (IEEE Conf. on Artificial nition), Tampa/Clearwater, USA
• CSASE (Int’l Conf. on Com- Intelligence), Santa Clara, USA 27 May
puter Science and Software • HOST (IEEE Int’l Symposium on • WoWMoM (IEEE Int’l Sympo-
Eng.), Duhok, Iraq Hardware Oriented Security sium on a World of Wireless,
16 April and Trust), San Jose, USA Mobile and Multimedia Net-
• COOL CHIPS (IEEE Symposium 6 May works), Fort Worth, USA
on Low-Power and High-Speed • RTAS (IEEE Real-Time and
Chips and Systems), Tokyo, Embedded Technolog y and JUNE
Japan Applications S ymposium), 2 June
22 April Irvine, USA • MDM (IEEE Int’l Conf. on Mobile
• PacificVis (IEEE Pacific Visu- 11 May Data Management), Irvine, USA
alization Conf.), Taipei City, • ASYNC (IEEE Int’l Symposium 3 June
Taiwan on Asynchronous Circuits and • IPDPS (IEEE Int’l Parallel and
26 April Systems), Portland, USA Distributed Processing Sym-
• ICSE (IEEE/ACM Int’l Conf. • ISPASS (IEEE Int’l Symposium posium), Milano, Italy
on Software Eng.), Ottawa, on Performance Analysis of 5 June
Canada Systems and Software), Ghent, • ISMVL (IEEE Int’l Symposium
28 April Belgium on Multiple-Valued Logic),
• VTS (IEEE VLSI Test Sympo- 12 May Montreal, Canada
sium), Tempe, USA • SP (IEEE Symposium on Secu- 9 June
rity and Privacy), San Francisco, • DCOSS-IoT (Int’l Conf. on Dis-
MAY USA tributed Computing in Smart
4 May 19 May Systems and the Internet of
• ARITH (IEEE Symposium on • CCGrid (IEEE Int’l Symposium Things), Lucca, Italy
Computer Arithmetic), El Paso, on Cluster, Cloud and Internet 11 June
USA Computing), Tromsø, Norway • CVPR (IEEE/CVF Conf. on Com-
• FCCM (IEEE Annual Int’l Sym- • ICDE (IEEE Int’l Conf. on Data puter Vision and Pattern Rec-
posium on Field-Programmable Eng.), Hong Kong ognition), Nashville, USA

60 March 2025 Published by the IEEE Computer Society  2469-7087/25 © 2025 IEEE
16 June 6 July 4 August
• CSF (IEEE Computer Secu- • ISVLSI (IEEE Computer Society • COINS (IEEE Int’l Conf. on
rity Foundations Symposium), Annual Symposium on VLSI), Omni-layer Intelligent Sys-
Santa Cruz, USA Kalamata, Greece tems), Madison, USA
18 June 7 July 6 August
• CBMS (IEEE Int’l Symposium • IOLTS (IEEE Int’l Symposium • IRI (IEEE Int’l Conf. on Informa-
on Computer-Based Medical on On-Line Testing and Robust tion Reuse and Integration and
Systems), Madrid, Spain System Design), Ischia, Italy Data Science), San Jose, USA
• ICHI (IEEE Int’l Conf. on Health- • SERVICES (IEEE World Con- 20 August
care Informatics), Rende, Italy gress on Services), Helsinki, • RTCSA (IEEE Int’l Conf. on
21 June Finland Embedded and Real -T ime
• ISCA (ACM/IEEE Annual Int’l 8 July Computing Systems and Appli-
S y mp osium on C omputer • C OMP SAC (IEEE Annual cations), Singapore
Architecture), Tokyo, Japan C o mputers, Sof tware, and 24 August
23 June Applications Conf.), Toronto, • HCS (IEEE Hot Chips 37 Sym-
• DSN (Annual IEEE/IFIP Int’l Canada posium), Stanford, USA,
Conf. on Dependable Systems 14 July 26 August
and Networks), Naples, Italy • ICALT (IEEE Int’ l Conf. on • PST (Annual Int’l Conf. on Pri-
• SVCC (Silicon Valley Cybersecu- Advanced Learning Technolo- vacy, Security, and Trust), Fred-
rity Conf.), San Francisco, USA gies), Changhua, Taiwan ericton, Canada
26 June 20 July
• IEEE Cloud Summit, Washing- • ICDCS (IEEE Int’l Conf. on Dis-
ton, DC, USA tributed Computing Systems),
• MARIS (Symposium on Mari- Glasgow, United Kingdom
time Informatics and Robot- 21 July
ics), Syros, Greece • CISOSE (IEEE Int’l Congress
30 June on Intelligent and Service-Ori-
• EuroS&P (IEEE European Sym- ented Systems Eng.), Tucson,
posium on Security and Pri- USA
vacy), Venice, Italy • ICCP (IEEE Int’l Conf. on Com-
• ICME (IEEE Int’l Conf. on Mul- p u t at i o na l Ph o t o g r a p hy), Learn more
timedia and Expo), Nantes, Toronto, Canada about IEEE
France 28 July
Computer
• SCC (IEEE Space Computing
JULY Conf.), Los Angeles, USA
Society
2 July • SMC-IT (IEEE Int’l Conf. on conferences
• ISCC (IEEE Symposium on Space Mission Challenges for
Computers and Communica- Information Technology), Pas- computer.org/conferences
tions), Bologna, Italy adena, USA
 IEEE COMPUTER SOCIETY CAREER CENTER

Career Accelerating
Opportunities
Explore new options—upload your resume today

careers.computer.org

Changes in the marketplace shift demands for vital skills and talent.
The IEEE Computer Society Career Center is a valuable resource tool
to keep job seekers up to date on the dynamic career opportunities
offered by employers.
Take advantage of these special resources for job seekers:

JOB ALERTS TEMPLATES WEBINARS

CAREER RESUMES VIEWED

ADVICE BY TOP EMPLOYERS

No matter what your career level, the IEEE Computer

Society Career Center keeps you connected to
workplace trends and exciting career prospects.