Edge AI and Robotics Teaching Kit

Lecture 3.2
Ethics and Security
 The Edge AI and Robotics Teaching Kit is licensed by NVIDIA and UMBC under the
Creative Commons Attribution-NonCommercial 4.0 International License.

2
Topics

Singularity and Superintelligence

Privacy and Surveillance

Manipulation of Behavior

Transparency of AI

Data Bias (covered last lecture – module 11)

Human Robot Interaction

Automation and Employment

Policies – Trustworthy and Responsible AI

3
Learning Objectives

Discuss common ways that online data can be used to manipulate behavior

List concerns of AI in the workforce

Review policies addressing concerns AI

List the seven requirements of Trustworthy AI as defined by the EU

List the four principles of Explainable AI

4
Recommended Reading

Ethics of Artificial Intelligence and Robotics (Stanford Encyclopedia of Philosophy)

Draft NISTIR 8312 - Four Principles of Explainable Artificial Intelligence

The Social Dilemma

5
 Will AI take over the world?

6
Future and Ethics of AI

What will super-intelligent AI bring?

• Automation and unemployment? What if people are not longer needed to make
economy work?
• Smart weapons? Automated terrorists?

What will a super-intelligent AI be able to do better?

• predict the future?
• optimize (constrained optimization)?
• Build other robots?

©D.L. Poole and A.K. Mackworth 2010-

7
2020
Four Principles of Explainable AI
• Explanation: Systems deliver accompanying evidence or reason(s) for all outputs.
• Meaningful: Systems provide explanations that are understandable to individual users.
• Explanation Accuracy: The explanation correctly reflects the system’s process for generating the
output.
• Knowledge Limits: The system only operates under conditions for which it was designed or when
the system reaches a sufficient confidence in its output.

Retrieved from
https://www.nist.gov/system/files/documents/2020/08/17/NIST%20Explainable%20AI%20Draft%
20NISTIR8312%20%281%29.pdf

8
Future and Ethics of AI

Whose values/goals will they use? (Why?) Will

we need a new ethics of AI?

©D.L. Poole and A.K. Mackworth

2010-2020
9
Future and Ethics of AI

Is super-human AI inevitable (wait till computers get faster)?

(Singularity)
Is there fundamental research to be done?
Is it easy because humans are not as intelligent as we like to
think?

©D.L. Poole and A.K. Mackworth 2010-

2020
10
Root of the problem

Is AI the danger or humans?

How do you address the root of the problem?

11
 Privacy and Surveillance

12
Privacy and Surveillance

Federal Agencies have strict privacy laws to protect data they collect

Private industry uses that data as a commodity. Internet websites and vendors are required by FTC to
provide privacy statement Privacy policy - Wikipedia.

You can research anyone’s address, phone number and sometimes network.

Phone apps can be used to track location. When is this good and when is this bad?

Security and doorbell cameras are everywhere

What are your responsibilities to safeguard privacy from camera recordings?

13
How to protect privacy

Ensure applications comply with NIST Cybersecurity, Data, and Privacy Framework Guidelines:
https://nvlpubs.nist.gov/nistpubs/CSWP/NIST.CSWP.01162020.pdf

Secure physical, local and physical access to native version of data

Encrypt transmission of native version of data

Auto redaction of sensitive text and data (pdf-redactor · PyPI)

Redact faces from images and videos containing people:

• https://github.com/NVIDIA-AI-IOT/redaction_with_deepstream

• https://docs.microsoft.com/en-us/azure/media-services/previous/media-services-redactor-walkthrough

• https://github.com/NVIDIA-AI-IOT/deepstream_python_apps/blob/master/notebooks/deepstream_test_1.ipynb

14
OpenCV Redacting an Image (CPU)

fpath='./cascades/haarcascade_frontalface_alt.xml'
import numpy as np
epath='./cascades/haarcascade_eye.xml'
import cv2
import matplotlib.pyplot as plt
face_detect = cv2.CascadeClassifier(fpath)
eye_detect = cv2.CascadeClassifier(epath)
# A function for plotting the images
def plotImages(img):
faces = face_detect.detectMultiScale(gray, 1.3, 5)
plt.imshow(img, cmap="gray")
for (x,y,w,h) in faces:
plt.axis('off')
cv2.rectangle(image,(x,y),(x+w,y+h),(255,0,0),2)
plt.style.use('seaborn')
roi_gray = gray[y:y+h, x:x+w]
plt.show()
roi_color = image[y:y+h, x:x+w]
roi = cv2.GaussianBlur(roi_color, (23, 23), 30)
image = cv2.imread('lena_color.tif')
# impose this blurred image on original image to get final image
image[y:y+roi.shape[0], x:x+roi.shape[1]] = roi
# Converting BGR image into a RGB image
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Display the output
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
plotImages(image)

15
DeepStream Lab - GPU
Video Analytics for image analytics and redaction

Nvidia Online Lab: Getting Started with DeepStream for Video Analytics on Jetson Nano
https://courses.nvidia.com/courses/course-v1:DLI+C-IV-02+V1/info

This lab will walk you through installing and configuring DeepStream on your Jetson Nano.

You will need to use either your mini-USB connection or directly attach your Nano to a Monitor, keyboard and mouse to
view the videos.

Sample Redaction Application using Deepstream: https://github.com/NVIDIA-AI-IOT/redaction_with_deepstream

16
Manipulation
of Behavior

‘If the product is free, you’re the

product,’ Tristan Harris, former
design ethicist at Google

17
Manipulation of behavior

Sites are aimed at ‘attention extraction’ to get you click as much as possible and then harvest and sell
your data

With accumulation of data from repeated interaction with systems, users can be targeted and nudged to
perform actions

Advertisers and online sellers and exploit users with bias, addiction and other behavioral issues to
maximize profits like casinos

18
Human Futures (Data Markets about you)

Data is used to advertise to your targeted insecurities and empathy

Weight loss

Age fighting products

Safety

Attractiveness

Relationship Advice

Animal rescue

19
Who Wins?

You spend more time viewing posts

You spend more time sharing your opinions

Data is harvested from online usage and sold to advertisers

You purchase in response to emotions rather than logic:

You purchase items you might not necessarily purchase

You spend much more on a particular product than its worth

20
 Transparency in AI

21
Transparency in AI
AI should be traceable and explainable

AI systems for automated decision support and “predictive analytics” raise “significant concerns
about lack of due process, accountability, community engagement, and auditing” (Whittaker et al.
2018: 18ff).*

A person affected by an automated decision may not understand how that decision was made.,
the system is “opaque” to that person. Experts who developed machine learning algorithms
behind the system may not even know how a particular pattern was identified. Bias in decision
systems and data sets is exacerbated by this opacity. To remove bias, the analysis of opacity and
bias go hand in hand, and political response must tackle both issues together.*

*Adapted from from Ethics of Artificial Intelligence and Robotics (Stanford Encyclopedia of Philosophy)

22
Explainable AI

Four Principles of Explainable Artificial Intelligence (published NIST.GOV

Draft NISTIR 8312 - Four Principles of Explainable Artificial Intelligence:

Explanation – system provides an explanation

Meaningful – explanations are understandable

Explanation Accuracy – explanations are inline with the system’s process for generating output

Knowledge Limits – only operates under the system it was designed for or the output is not deemed reliable

23
 Human Robot Interaction

24
Human Robot Interaction

Can Robots be substituted for humans as caregivers?

Psychologists?

Nurses

Household help

Parents

Interesting movie about Robot raising an orphan: I Am Mother - Wikipedia

25
Human Robot Interaction

Robots useful for augmented support:

Lifting immobile patients

Monitoring vital signs and sending alerts

Robotic limbs

Ethics of Artificial Intelligence and Robotics (Stanford Encyclopedia of Philosophy) makes a case about
longer lifespans will have increased need for healthcare workers but not necessarily in the availability of
healthcare workers. Could robots shore up health care workers or enable fewer workers?

26
 Automation and Employment

27
Automation and Employment

Longstanding fear that automation will replace human workers

Automation can lead to productivity gains that results in fewer workers

What currently seems to happen in the labor market as a result of AI and robotics automation is “job
polarization” or the “dumbbell” shape (Goos, Manning, and Salomons 2009): The highly skilled technical
jobs are in demand and highly paid, the low skilled service jobs are in demand and badly paid, but the
mid-qualification jobs in factories and offices, i.e., the majority of jobs, are under pressure and reduced
because they are relatively predictable, and most likely to be automated (Baldwin 2019).

28
Autonomous AI

There is some discussion of “trolley problems” in this context. In the classic “trolley problems” (Thomson 1976;
Woollard and Howard-Snyder 2016: section 2) various dilemmas are presented. The simplest version is that of a
trolley train on a track that is heading towards five people and will kill them, unless the train is diverted onto a side
track, but on that track there is one person, who will be killed if the train takes that side track.

This Robot would let 5 People die | AI on Moral Questions | Sophia answers the Trolley Problem - YouT
ube

29
Autonomous Accuracy

How accurate can we make automation models?

What are the risks for failures even if rare?

Loss of life if event leads to a vehicle accident?

https://finance.yahoo.com/news/tesla-autopilot-investigated-again-u-120223737.html

Automated checkouts – benefit of cost savings in staff productivity might outweigh occasional loss in revenue due
to inaccurate checkouts.

30
 AI Policy

31
Policies

Lot of effort to create policies and may conflict with other polices

Trustworthy AI in the government:

Executive Order on Promoting the Use of Trustworthy Artificial Intelligence in the Federal Government –
The White House (archives.gov)

EU policy document suggests “trustworthy AI” should be lawful, ethical, and technically robust, and then spells
this out as seven requirements: human oversight, technical robustness, privacy and data governance, transparency,
fairness, well-being, and accountability (AI HLEG 2019 [OIR]).

32
Other Resources

National Institute of Standards and Technology (NIST) – https://www.nist.gov/artificial-intelligence ,

recommends hardware and risk framework, provides input into policy recommendations

National Security Commission on Artificial Intelligence (NSCAI)-

https://www.nscai.gov/2021-final-report/

33
34
35
36
Edge AI and Robotics Teaching Kit

Thank You