Intro to AI Ethics + Black Mirror

Fall 2024

Announcements

1. Don’t forget to order your books (Sandel + Benjamin)! We’ll be discussing the first few chapters of the Sandel book next Thursday
2. Today at 5:00PM: Welcome Back Gathering (Sponsored by the CS+Math Departments)
3. Readings for Tuesday are posted: “The Science of Morality”

Outline

• A High-Level Overview of AI Ethics (1:20-1:45PM)
• Discussion of the Black Mirror Episodes
• Small-group discussions (1:50PM - 2:10PM)
• Break (2:10-2:15PM)
• Whole-group discussion (2:15-3:00PM)

Outline

• A High-Level Overview of AI Ethics
• Discussion of the Black Mirror Episodes
• Small-group discussions (1:50PM - 2:10PM)
• Break (2:10-2:15PM)
• Whole-group discussion (2:15-3:00PM)

What is Artificial Intelligence?

• A digital agent that uses an algorithm to perform a task that usually requires human intelligence
• Types of AI algorithms:
• Static algorithms – traditional programs that perform a fixed sequence of actions, usually classified as knowledge-based systems
• Dynamic algorithms—that learn and evolve by interacting by using data from the environment. Usually classified as “Machine Learning” algorithms (supervised, unsupervised, reinforcement learning).

Brief History of AI

1940s - 1950s

• 1945 – Enigma Machine (code breaker) created by Alan Turing
• 1950 – Turing Test for machine intelligence proposed
• 1956 – AI conference at Dartmouth – defining the field
• 1958 – First implementation of the “perceptron” – a computational model of a simplified artificial neuron.

Brief History of AI

1960 - 1974

DARPA funded several AI research institutes; Lack of computational power made it difficult to do anything interesting that resembled real human intelligence.

• 1964 – First chatbot, “Eliza,” invented (rules-based).
• 1970 – “From three to eight years we will have a machine with the general intelligence of an average human being.” �– Marvin Minsky, Life Magazine
• 1974–1980 – First AI “Winter” (no funding)

Brief History of AI

1980s

New round of public funding in early ‘80s – many of the methodological advances in AI happened during this time:

• Deep learning & neural networks (learning from experience)
• Expert systems (symbolic AI & logic programming) which mimic the decision making process of a human expert.

Brief History of AI

1990s & 2000s

Second AI “winter” for public funding, but many landmark AI accomplishments still achieved – in part because of the increased computational power available (Moore’s Law).

• 1997 – Deep Blue beats Kasparov; speech recognition on Windows (Dragon)
• 1998 – Kismet (robot with emotions) – MIT AI Laboratory
• 2002 - Roomba vacuum
• 2008 – Voice recognition on the iPhone

Brief History of AI

2010 - Present

Given increase in computational power and access to big data, many landmark achievements – and more public visibility & funding:

• 2011 – IBM Watson Introduced
• 2014 – Alexa introduced
• 2015 – Deep Learning paper published in Nature by Yann LeCun, Yoshua Bengio & Geoffrey Hinton
• 2020 – ChatGPT introduced

Narrow v. General AI

“Narrow AIs” perform specific tasks within a limited domain. All current applications of AI are generally considered “narrow”:

• Search engines (Google search)
• Speech recognition (Siri, Alexa)
• Self-driving cars (Waymo)
• Generative tools (ChatGPT)

“General AI” – refers to machines or systems that possess human-like intelligence. This concept is largely theoretical.

• Whether this is actually possible is widely debated

What is deep learning a big deal?

What is deep learning a big deal?

• Generalizability – The same basic approach can be used to train models across a wide variety of domains – text / language, image recognition, etc.
• Data & Computational Resources – Requires a lot of data and computational power – which different businesses and organizations are differentially able to leverage.

But isn’t this just an engineering problem?

How might other disciplines guide the direction of AI? Your thoughts here…

• Historians
• Sociologists
• Psychologists
• Political Scientists
• Economists
• Environmental Studies
• Biologists
• People / communities who are positively or �negatively impacted by AI

“Levels” of impact to consider

• Psychological
• Social
• Political impacts

​

What are examples of each (both good and bad)?

Moral Frameworks

• Utilitarianism, Rights, Virtue (next week)!
• Also Social Contract Theory and Libertarianism

Theoretical Ideas

• Social theory helps us analyze the world
• Gives us a way to name the things we’re seeing and put them into a broader framework for understanding the world
• Some theoretical ideas take a while to wrap your head around, but having exposure to a broad range of psychological, social, political, and economic theories will help you think in new ways
• We will draw from philosophy of technology, �engineering ethics, Science & Technology Studies, �Media Studies, and more!

Major Themes in AI Ethics

What are some examples of AI issues along each of these dimensions?

• Human Well-Being
• Safety
• Privacy
• Transparency
• Fairness
• Accountability

Outline

• A High-Level Overview of AI Ethics
• Discussion of the Black Mirror Episodes
• Small-group discussions (1:50PM - 2:10PM)
• Break (2:10-2:15PM)
• Whole-group discussion (2:15-3:00PM)

Activity: 1:50PM to 2:10PM

Split into two groups:

• Nosedive: move towards the door
• Joan is Awful: move towards the windows

​

Sit at a table with others who watched the same episode as you:

• Open the discussion questions document
• Sarah will assign your group a set of questions
• Follow the instructions
• Take notes and be prepared to share out