Lecture 27: Philosophy, Ethics, and Safety of AI

Lecture 27: Philosophy, Ethics, and Safety of AI¶

AIMA Chapter 27 — 1 hour¶

Learning Objectives¶

• Examine philosophical limits of AI

• Address ethical issues: bias, privacy, weapons

• Understand AI safety and alignment

• Consider fairness and transparency

Limits of AI¶

• Informality: Human knowledge not fully formalizable?

• Disability: Machines can’t do X?

• Mathematical: Gödel, undecidability

• Measurement: How to measure intelligence?

Can Machines Think?¶

• Turing test: Behavioral

• Chinese room: Syntax vs. semantics

• Consciousness: Qualia, hard problem

Lethal Autonomous Weapons¶

• LAWS: Weapons that select targets

• Concerns: Accountability, proportionality

• Ban campaigns: Stop Killer Robots

Bias and Fairness¶

• Data bias: Reflects historical bias

• Algorithmic bias: Disparate impact

• Fairness definitions: Demographic parity, equalized odds

• Mitigation: Data, algorithms, evaluation

Privacy and Surveillance¶

• Surveillance: Face recognition, tracking

• Privacy: Right to be forgotten

• Security: Adversarial attacks

Trust and Transparency¶

• Explainability: Why did model decide?

• Interpretability: Understand internals

• Black box: Medical, legal implications

Future of Work¶

• Automation: Which jobs?

• Augmentation: Human-AI collaboration

• Displacement: Retraining, safety nets

AI Safety¶

• Value alignment: AI goals with human values

• Robustness: Handle distribution shift

• Corrigibility: Allow human intervention

Summary¶

• Philosophy: Limits, consciousness

• Ethics: Bias, privacy, weapons

• Safety: Alignment, robustness

• Responsible AI: Design for good

References¶

• AIMA Ch. 27

• Russell & Norvig, AIMA 4e, Ch. 27

• Chapter PDF: chapters/chapter-27.pdf

Questions?¶

Next lecture: The Future of AI (Chapter 28)