Ethical Considerations in AI: Representation, Bias, and Fairness

 
AI: Representation and Problem Solving
 
AI and Ethics
 
Instructors: Nihar Shah and Tuomas Sandholm
Slide credits: CMU AI with some drawings from ai.berkeley.edu
 
AI is in a fairly unique position with respect to ethics
 
Not only must AI developers behave ethically
 
The AI systems, themselves, must behave ethically
 
(Approximately) True Story
Causality
Classical: Correlation is not causation
https://statisticseasily.com/correlation-vs-causality/
 
Prediction is not causation
 
 
As in our story…
 
Past data may not reflect the future
 
Any intervention changes the system
Any intervention can also change people’s behavior
Explanability/interpretability
What is the AI doing? How does it really work?
Why is it giving the output it is giving?
 
Are “explanations” faithful to the AI’s actual working?
Do explanations make humans complacent?
Bias/(Un)fairness
 
https://www.reuters.com/article/idUSKCN1MK0AG/
 
https://www.nber.org/system/files/working_papers/w9873/w9873.pdf
 
One cause (although not the only one): 
Data used to create the AI was biased
 
AI mimics the same patterns
 
https://www.washingtonpost.com/techn
ology/2019/12/19/federal-study-
confirms-racial-bias-many-facial-
recognition-systems-casts-doubt-their-
expanding-use/
Privacy
 
 
https://cybernews.com/security/chatgpt-samsung-leak-explained-lessons/
 
Privacy
 
What are some tradeoffs we make when giving AI our private information?
Do you have a limit for the type or amount of data you are willing to provide?
How can we make the data usage of AI systems more explicit?
Are there ways that we can still enjoy the benefits of AI systems without
giving up our private information?
 
Right to be forgotten?
 
LinkedIn or Facebook can delete a post
Google can delete a search result
 
But how to make the AI un-learn?
 
https://en.wikipedia.org/wiki/Right_to_be_forgotten
 
Surveillance
 
16
 
Surveillance
 
17
 
Liability
 
How do we determine what went wrong (or right) in deployed AI
systems?
Which techniques that we learned about have explainable results?
Who is responsible for “mistakes” that AI systems make?
What if someone or another AI misuses AI technology? Who is
responsible?
 
Think about:
 
Self-driving cars
 
Healthcare decision-making
           Air traffic control
 
https://cms.law/en/gbr/publication/artificial-intelligence-who-is-liable-when-ai-fails-to-perform
 
Employment
 
What types of jobs should AI do instead of
people?
 
What will people do if their jobs are
taken?
 
How can AI systems improve or change
people’s jobs even if it doesn’t take the
jobs?
 
https://www.nytimes.com/2023/08/24/upshot/artificial-intelligence-jobs.html
 
https://www.bloomberg.com/news/articles/2024-02-08/ai-is-
driving-more-layoffs-than-companies-want-to-admit
 
Misinformation
 
Should we worry about future A.I.?
Weak AI
Strong AI
 
Singularity
 
Narrow AI
Limited number
of applications
 
 
 
 
Artificial General
Intelligence (AGI)
Recursive self-
improvement
Beyond human
control
 
 
 
AI Weapons/Safety
 
https://www.youtube.com/watch?v=9CO6M2HsoIA
Value (Mis)alignment
 
”If we use, to achieve our purposes, a mechanical agency with whose operation we
cannot interfere effectively… we had better be quite sure that the purpose put into
the machine is the purpose which we really desire.” – Norbert Weiner
 
A
r
e
 
t
h
e
 
A
I
s
 
g
o
a
l
s
 
s
a
m
e
 
a
s
h
u
m
a
n
 
g
o
a
l
s
/
p
r
e
f
e
r
e
n
c
e
s
?
 
https://www.ted.com/talks/stuart_russell_how_ai_might_make_us_better_people
 
25
 
AI and Ethics: Careful Thought in AI Research
Slide Note
Embed
Share

Exploring the ethical dimensions of artificial intelligence (AI) reveals the unique position AI systems hold in behaving ethically. From causality to interpretability, bias, and privacy concerns, this discussion sheds light on the importance of ethical considerations in AI development and implementation.


Uploaded on Mar 09, 2024 | 1 Views


Download Presentation

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

E N D

Presentation Transcript


  1. AI: Representation and Problem Solving AI and Ethics Instructors: Nihar Shah and Tuomas Sandholm Slide credits: CMU AI with some drawings from ai.berkeley.edu

  2. AI is in a fairly unique position with respect to ethics Not only must AI developers behave ethically The AI systems, themselves, must behave ethically

  3. (Approximately) True Story

  4. Causality Classical: Correlation is not causation Prediction is not causation https://statisticseasily.com/correlation-vs-causality/

  5. Any intervention changes the system As in our story Past data may not reflect the future

  6. Any intervention can also change peoples behavior

  7. Explanability/interpretability What is the AI doing? How does it really work? Why is it giving the output it is giving? Are explanations faithful to the AI s actual working? Do explanations make humans complacent?

  8. Bias/(Un)fairness https://www.reuters.com/article/idUSKCN1MK0AG/

  9. One cause (although not the only one): Data used to create the AI was biased https://www.nber.org/system/files/working_papers/w9873/w9873.pdf AI mimics the same patterns

  10. https://www.washingtonpost.com/techn ology/2019/12/19/federal-study- confirms-racial-bias-many-facial- recognition-systems-casts-doubt-their- expanding-use/

  11. Privacy https://cybernews.com/security/chatgpt-samsung-leak-explained-lessons/

  12. Privacy What are some tradeoffs we make when giving AI our private information? Do you have a limit for the type or amount of data you are willing to provide? How can we make the data usage of AI systems more explicit? Are there ways that we can still enjoy the benefits of AI systems without giving up our private information?

  13. Right to be forgotten? LinkedIn or Facebook can delete a post Google can delete a search result But how to make the AI un-learn? https://en.wikipedia.org/wiki/Right_to_be_forgotten

  14. Surveillance 16

  15. Surveillance 17

  16. Liability How do we determine what went wrong (or right) in deployed AI systems? Which techniques that we learned about have explainable results? Who is responsible for mistakes that AI systems make? What if someone or another AI misuses AI technology? Who is responsible? Think about: Air traffic control Self-driving cars Healthcare decision-making https://cms.law/en/gbr/publication/artificial-intelligence-who-is-liable-when-ai-fails-to-perform

  17. Employment What types of jobs should AI do instead of people? What will people do if their jobs are taken? How can AI systems improve or change people s jobs even if it doesn t take the jobs? https://www.nytimes.com/2023/08/24/upshot/artificial-intelligence-jobs.html https://www.bloomberg.com/news/articles/2024-02-08/ai-is- driving-more-layoffs-than-companies-want-to-admit

  18. Misinformation

  19. Should we worry about future A.I.? Singularity Weak AI Strong AI Artificial General Intelligence (AGI) Recursive self- improvement Beyond human control Narrow AI Limited number of applications

  20. AI Weapons/Safety https://www.youtube.com/watch?v=9CO6M2HsoIA

  21. Value (Mis)alignment Are the AI s goals same as human goals/preferences? If we use, to achieve our purposes, a mechanical agency with whose operation we cannot interfere effectively we had better be quite sure that the purpose put into the machine is the purpose which we really desire. Norbert Weiner

  22. https://www.ted.com/talks/stuart_russell_how_ai_might_make_us_better_peoplehttps://www.ted.com/talks/stuart_russell_how_ai_might_make_us_better_people

  23. AI and Ethics: Careful Thought in AI Research 25

Related


More Related Content

giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#