Learning Bayesian Statistics BITESIZE | How AI is Redefining Human Interactions, with Tom Griffiths
May 21, 2025
In this discussion, Professor Tom Griffiths from Princeton University, an expert in psychology and computer science, shares insights on the interplay between human and artificial intelligence. He highlights key differences in learning processes, emphasizing that AI should enhance human capabilities rather than merely mimic them. Tom addresses how AI can help overcome human biases, improve decision-making, and align better with human cognition. The conversation underscores the need for AI models that reflect human understanding to make more effective systems.
AI Snips
Chapters
Transcript
Episode notes
Human Intelligence Constraints
- Human intelligence operates under constraints of limited data, fixed compute, and low communication bandwidth.
- These constraints shape cognition differently than AI systems which scale data and compute easily.
Current AI Scaling Approach
- AI currently improves by increasing data and compute rather than mimicking human inductive biases or resource efficiency.
- This approach avoids complex engineering needed for human-like reasoning but differs fundamentally from human cognition.
AI-Human Representational Alignment
- Training AI on human-generated data like language aligns their worldview partially with humans.
- More human-like AI requires deeper representational and conceptual alignment beyond surface similarities.