CIA-2.

Digital Business for Managers

UNDER THE GUIDANCE OF

PROF. TARUN SINGHAL

SEPTEMBER 2022

SUBMITTED BY:

RASHMI RATHI

1MBA

REGISTRATION NO.

22221065
 AI AS TECHNOLOGY:

Artificial intelligence refers to a computer system's capacity to imitate human cognitive proce
sses like learning & problem-solving. Expert systems, natural language processing, speech
recognition, and machine vision are some examples of specific AI applications. For the creation
and training of machine learning, it requires a foundation of specialized hardware & software.
There is no one programming language that is exclusively associated with AI, but a handful
are, including Python, R, and Java.

In contrast to the natural intelligence exhibited by animals, including humans, AI is intelligence

demonstrated by robots. Any system that senses its environment and takes actions to maximize
its chances of attaining its goals is considered an intelligent agent, and the study of intelligent
agents is what AI research is referred to as . Previously, robots that mimic and exhibit "human"
cognitive abilities associated with the human mind were referred to as "Artificial Intelligence."
Major AI researchers are now describing AI in terms of rationality that does not constrain how
intelligence can be expressed.

ABOUT THE COMPANY:

Since its foundation in 1962, the entire Walmart business has adhered to the founder Sam
Walton's motto, "saving people money, so they can live better." It is an American multinational
retail corporation that operates a chain of supermarkets. With more than 11,500 shops operating
under 56 banners in 27 countries and e-commerce websites in 10 countries, Walmart has
become the largest grocery and retail chain in the United States with a largely brick-and-
mortar-based business strategy. They service close to 270 million clients and have a global
workforce of about 3 million colleagues. With a total capitalization of more than $500 billion,
Walmart is also the biggest business in the world by revenue.

Walmart works under a diversified business strategy with three main divisions: Walmart U.S.,
Walmart International, and Sam's club, with over $555 billion in Net sales in 2021. Along with
Walmart+, a subscription service that offers savings, retail deliveries, and unlimited free
shipping is also available. As of July 31, 2022, there will be 10,573 total units, comprising
4,723 Walmart U.S. stores, 5,250 Walmart international stores, and 600 Sam's Clubs.

1. Walmart U.S.

Walmart's business model includes financial services as such as money orders, prepaid cards,
wire transfers, and bill payment options. They market private-label products and goods with
licences. Supercentres, discount shops, and various small store formats are the three main store
forms they use to give clients physical access.

2. Walmart International

The company's foreign division encompasses the 27 nations where there are Walmart stores.
They also have stores in other formats, including wholesale, retail, and others. Joint ventures
and totally owned subsidiaries are used to run the stores.

3. Sam’s Club

It runs in 44 U.S. states as well as Puerto Rico. The topic of membership-only warehouse clubs
is covered in this section. They provide both individual clients and corporate owners a
subscription service. With the use of a tool called "Club Pickup," the digital access element in
this section supports the physical clubs and members.
 MARKET NEED FOR CHOOSING & IMPLEMENTING THE TECHNOLOGY:

According to Srini Venkatesan, EVP, U.S. Omni Tech at Walmart Global Tech, Walmart
has increased its number of internal data scientists over the previous 6 years. These data
scientists work on teams that focus on emerging technologies, search and customization,
supply chain forecasting, optimization, and labour/demand planning.

It is "evolving from being an Automator of retail to becoming an enabler of retail - that is

where AI and ML are very relevant for us," according to Venkatesan, who oversees all
technology teams that support the company's global marketplace and stores.

1. SUPPLY CHAIN: According to Venkatesan, Walmart's supply chain AI efforts have

advanced over the past few years, moving from merely forecasting sales demand—how
much will sell—to forecasting consumer demand—what the customer will actually want
to buy—by analysing data from multiple sources, such as Google searches.
The use of AI and ML for replacement logic has been funded. To choose the best next-
available item, deep learning of AI analysis- hundreds of factors are considered in real-
time, including size, type, brand, price, individual preference, and current inventory.
Unexpected substitutions result in a poor customer experience and could cost a business
money or market share.
A timely invention in light of the COVID-19 Pandemic's surge in demand for online
food shopping, Walmart aims to make it simpler for its staff to select the appropriate
replacement when the requested item isn't on the store.

2. REMOTE FITTING ROOMS: It can be challenging to predict how well internet

clothing purchases will fit. Beyond that, customers must negotiate the nuances of
shipping costs and return procedures. The ease of buying clothes in person has endured
in a world where online interactions have recently spread to unexpected areas of our
lives, if only because fitting rooms provide answers to many of the questions before one
buys any new item. For merchants, returns of not well-fitted clothing can significantly
reduce otherwise strong sales. As a result, Walmart's P&L will lose actual money due to
decreased net sales as well as additional processing and handling costs that it will have
to pay.
 One of Walmart's most recent AI-driven products, the dynamic virtual fitting room
technology from Zeekit, which it acquired last year, is likewise powered by computer
vision. Customers can browse for clothing online and get a realistic idea of how an item
will look on them. However, remote fitting rooms could significantly lower returns and
restocking costs for stores if they can increase customer confidence before they add an
item to their cart.

3. SEARCH AS EXPANDING FRONTIER: If we approach a query from a standard

approach, we won't really get good results. We may have a query that is frequently
repeated and people may give a very strong signal about what it means. Venkatesan cited
"avocados from Mexico" as an illustration, adding, "The reason that's noteworthy is that
most avocados don't tell you that they're from Mexico." In contrast, he noted, the inquiry
itself is quite straightforward; it's clear what the user wants. "So, we put that in the
category of semantic inquiries where you really have to be on top of that, realise that the
avocado component is significant or infer in general from other things that you know
about an item that is likely to be important."
Pedersen also talked on Walmart's efforts to support bilingual searches, which help
Spanish-speaking users locate specific products on the website and in the app. Because
the search field is empty, he noted, "one of the interesting things about search
experiences in general is that individuals may type in whatever they like." Walmart
employs AI-based language identification to serve customers who are looking in
Spanish. "You determine that this search is probably going to be in Spanish, so you
translate it using machine translation."

4. TEXT TO SHOP: Text to Shop, a conversational AI platform from Walmart, was

recently released. It allows customers to text or voice their needs, and the service will
add it to their cart. It will offer product recommendations if a customer needs something
they've never bought before. Gosby added, "...make it simpler for the customer to convey
what it is that they want. Overall, these technologies are about giving customers
confidence to make purchases. It's basically a digital assistance platform that leverages
voice and text chat..." But, she admits, "making this simple is actually really hard - being
able to understand if you say things like add chocolate milk and pizza to my cart that
you really mean chocolate milk as opposed to chocolate versus milk."
5. SHELF SCANNING ROBOTS: Walmart intends to expand the tests of its robots that scan
shelves over 50 more locations, some of which are in its home state of Arkansas. The machines,
which are expected to take over shelf scanning in the near future, will rove across the aisles
inspecting prices, lost goods, and stock levels to gauge the store's level of inventory. In addition
to saving human personnel the effort of manually inspecting, this would free them up to
concentrate on other crucial matters. The robots are currently totally autonomous to perform
their responsibilities on their own, employing the principles of 3D imaging. However, specialists
will need to be on site to handle the problem in the event of a technological impairment.

6. DRIVERLESS SAME-DAY DELIEVERY: Walmart announced the launch of its

autonomous vehicle delivery service in Miami, Austin, and Washington, D.C. in September
2021. With initial test runs beginning in 2022, the objective is to enable same-day delivery from
neighbourhood stores to neighbouring houses. A customer's online order will be processed by
Argo's cloud-based technology, which will then plan and route deliveries. The goods will
subsequently be delivered in Ford cars with self-driving technology from Argo AI.

Social & Ethical issues:

While discussing on AI, there are talks regarding immoral things AI holds. These morally
troubling outcomes are the centre of AI debates. In contrast to being a goal in and of itself,
human flourishing is a promising means of advancing innovation and progress while also
boosting individual and societal well-being. Few issues related to AI could be: Cost to
innovation, Harming the integrity, Lack of trust and accuracy, Security issues, &
Unemployment.

• The first group of problems brought by ML are Artificial neural networks provide the
foundation for many of ML approaches that have contributed to the current success of AI.
These methods' opacity, unpredictability, and need for big datasets to train the technology
are the characteristics that raise ethical questions. In most cases, no one can predict in
advance how the system will respond to a certain set of inputs—not the developer, not the
deployer, and not the user. Similar is the case with Walmart. Additionally, because the
system is flexible and dynamic and learns, past behaviours are not a perfect indicator of
future behaviour in similar circumstances.
• Data protection and privacy are two fundamental ethical concerns that are commonly
raised. Data protection can be seen as a way to defend Informational Privacy. ML-based
AI has various hazards for data security. On the one hand, it requires sizable data sets for
training, & access to those sets may raise security concerns. Beyond AI, cybersecurity is a
recurring issue. AI systems can be vulnerable to brand-new security flaws like model
poisoning attacks. Walmart keeps heavy check of privacy and security. It might lead to a
bigger problem still. Walmart Intelligent Retail Lab (IRL) Breaches Privacy Nightmares
While Promising A Better Tomorrow.
• AI has the potential to enable the re-identification of anonymized personal data. Although
most nations have well-established data protection laws, AI has the potential to introduce
new data protection issues that are not currently covered by the law, raising ethical
questions.
• Although dependability is an issue, typical deterministic testing regimes may not be
appropriate to ML because of their opaqueness and unpredictable nature. The outcomes are
reliant on the training data's quality, which might be challenging to judge.
• If the output has an impact on ethical value, it may be necessary to evaluate the
trustworthiness of ML in several ways, which can be problematic ethically. For instance,
an AI system that is used to identify disease in pathology may function effectively in
research settings with properly annotated training data , but it does not guarantee that the
same system employing the same model would function similarly under clinical conditions.
• Transparency may be further constrained by the business confidentiality of algorithms &
models. Although the term "transparency" is debatable in itself, a lack of openness calls
into questioning the accountability & makes it challenging to identify & address issues of
bias & discrimination.
• One significant issue is that biases can be reproduced by ML, whether on purpose or
accidentally. Not only is it unethical, but discrimination based on particular qualities also
violates people's human rights. AI poses a risk to this human right. Another important
ethical concern with ML is safety, particularly for systems that communicate with the real
world directly.
• (Un)employment is the main cause for concern. Long recognised is the potential for AI-
related technologies to usher in a new era of automation and consequently eliminate jobs.
In fact, Norbert Wiener predicted that the employment situation will be gravely affected by
computers vs. humans fighting for jobs.
• Freedom is yet another fundamental ethical dilemma. It is simple to understand how an
individual's freedom might be impacted if AI played a role in the parole decision. However,
AI has a broader and more nuanced impact on freedom. The technology that are all around
us influence the range of potential actions by granting or denying access to information.

Common Practices to Mitigate AI Risk

The evolving best practices and potential mitigators are listed below to help organizations in
their internal talks about potential AI concerns. The training data, the learning process, and the
output predictions are the three potential intervention sites and there are three kinds of
mitigation methods to match: pre-processing, in-processing, and post-processing. The benefits
of post-processing methods include their suitability for run-time situations and the possibility
that they don't need access to the training process. Additionally, because post-processing
algorithms employ a black-box methodology, they can be used with any ML model because
they don't necessarily require access to the internal workings of models or their derivatives.

A. Oversight-Processes
AI systems, which learn from data and provide output based on that learning, are
significantly impacted by the quality of the data they use. Both the data quality and any
potential biases in the training set might be evaluated. When appropriate and practical,
comparing AI systems against alternative models and using well-known methodologies to
enable model interpretability may be included in the evaluation process. Increased
confidence in AI systems may result from better understanding of the variables influencing
their suggestions.
B. Monitoring for Drift
Drift in AI can result in various mistakes & hazards. For instance, poor model accuracy
could occasionally be attributable to the Target and independent variables connection
changing over time. As a result, drift detection becomes a key tool in reducing various AI-
related risks, such as those linked to a model's security, privacy, & fairness. One technique
to shed light on the "accuracy drift" of the model is through monitoring, which can take
into account the data model receives during production and assess accuracy of the model.
Additionally, it might determine whether the input data significantly differs from the
model's training data, which might aid in the detection of "data drift."
 Enterprise applications may benefit from accuracy drift detection because it can spot a
decline in model accuracy before the change has a substantial impact on the business. Data
drift aids businesses in understanding how data properties vary when data is used. While
Accuracy drift can make your model worse.
C. Addressing Discrimination in AI
The majority of lending institutions have compliance, fair lending, and system governance
teams that look for signs of prejudice in input variables and processes. Some of the job may
eventually be streamlined and automated. A human-centred strategy may be necessary for
fair AI. A well-trained diversified group of people with generalized knowledge &
experience are unlikely to be completely replaced by an automated procedure.
In order to reduce inequalities, mitigation algorithms identify the "best" method for a
specific degree of quality & discrimination measure. The algorithms look for substitute
systems in cases where no higher-quality systems can be found for any given level of
discrimination. On the other hand, no solution can be developed that lessens discriminating
for any given degree of quality. Before using it in a production, more testing & research
must be done in a production environment.

By typically excluding one or two variables that cause disparate impact from the system
and testing a few substitutes, feature selection may be used to reduce dissimilar impact. In
complicated AI/ML systems, it has been demonstrated that these strategies have only
modest success. More modern methods reduce prejudice by concentrating on data pre-
processing, decision-making within algorithms, and output post-processing. The legal
context in which the system is used and the system's use itself determine whether these
procedures are appropriate for use in a particular case.

D. Ensuring Quality Explanations

For enterprises, ensuring that AI/ML explanations are both trustworthy and useful could
be difficult. For instance, explanations might be imprecise estimates, inconsistent
statements, or flawed AI/ML systems. In order to potentially lower the risks connected
with explainability, companies may test explanatory procedures in human assessment
studies or for accuracy and stability.

E. Potential Risk-Mitigation
Organizations should only provide the information needed by the appropriate consumers
or as permitted by law in order to reduce any risks. Depending on the implementation and
control environment, existing security mechanisms like real-time anomaly detection, user
authentication, and API throttling could potentially be used to safeguard AI/ML systems
trained on sensitive data with predictions visible to end users.
Effective risk mitigants for AI deployments could be traditional, robust technological and
cyber controls. As machine learning matures, the developing field of adversarial learning
may aid in developing secure machine learning systems. Some theoretical mitigation
approaches are being further investigated in the technology sector, despite the fact that this
is still an area of research that is still developing. For instance, differential privacy is a
suggested technique for preserving the confidentiality of the training data. Having robust
technology and cyber controls may be an effective mitigation, depending on
implementations and the surrounding environment. Strong information security procedures
may be able to prevent model extraction attacks, but the ability to identify extracted models
via the watermarking technique still exists. The AI/ML system used for watermarking is
trained to create distinctive outputs for particular inputs. It may be a sign of intellectual
property theft if another system generates the same distinctive result from the same inputs.
REFERENCES:

• Pereira, D. (2022, July 15). Walmart Business Model. The Business Model Analyst.
https://businessmodelanalyst.com/walmart-business-model/
• Owen, R. (2021, August 7). AI at Walmart – Comparison to Amazon, and Two Unique
Use-Cases. Emerj's editorial. https://emerj.com/ai-sector-overviews/ai-at-walmart/
• Burns, E. (2022, February). What is artificial intelligence (AI)?. Tech Target.
https://www.techtarget.com/searchenterpriseai/definition/AI-Artificial-Intelligence
• Goldman, S. (2022, June 9). AI is embedded everywhere at Walmart. Venture Beat.
https://venturebeat.com/business/ai-is-embedded-everywhere-at-walmart/
• Cuofano, G. (n.d.). The Walmart Business Model. Four Week MBA.
https://fourweekmba.com/walmart-business-model/
• Stahl, BC. (2021, March 18). Ethical issues of AI. Artificial Intelligence for a Better
Future. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7968615/
• Wharton University of Pennsylvania. (n.d.). Artificial Intelligence Risk & Governance.
Retrieved from https://ai.wharton.upenn.edu/artificial-intelligence-risk-governance/