A talk at TASK by Christine Smoley
Notes
- “We shape our tools, and thereafter our tools shape us” - Marshall Berman
- Socrates believed tools made us stupider
- Modern LLM created by neural network, similar to human mind
- MIT analyzed students who use brain, students who use search engines, students who use LLM to write paper
- Socrates critiques writing:
- “If men learn this, it will implant forgetfulness; they will cease to exercise memory because they rely on that which is written, calling things to remembrance no longer from within themselves, but by means of external marks. What you have discovered is a recipe not for memory, but for reminder. And it is no true wisdom that you offer your disciples. We are filled with the conceit of wisdom rather than wisdom itself. He who reads in the belief that anything in writing will be clear and certain, would be an utterly simple person. If he thinks written words are of any use except to remind him who already knows the matter about which they are written”
- ANN vs BNN
- Both use Reinforcement Learning
- Both attempt to have an internal model of the worlds knowledge via generalization (mental model or Schema)
| X | BNN | ANN |
|---|---|---|
| Reinforcement learning | Continuous, adaptive reward | Feedback to update network weights, Fine tuned and uses RLHF |
| Weighting of neural pathways | Gradual, contextual, controlled by biochem | Results from mathematical weight adjustments |
| Cognitive Load | Limited cognitive resources requiring efficiency and Neurological Plasticity | Less energy constraints but dependent on fundamental datastructures and datasets |
| Working Memory | Declarative (short term) | Prompt/context window |
| Schemata | Operational memory / Intuition | Patterns based on statistics |
- Schemata Internalization
- Eligibility Trace
- Genie Wiley
- Google developed AlphaGo Zero which learned how to play Gomoku by absorbing rules and successful strategies
- Your Brain on ChatGPT
- The Memory Paradox - Why Our Brains Need Knowledge in an Age of AI
- Limbic System is the emotional reaction-based
- Basal Ganglia
- Prefrontal Cortex
- Consequences of AI:
- Cognitive impact, Basal Ganglia will weaken due to less practice, does not allow for procedural or automated learned responses
- Social Impact: Weakens Neurological Plasticity of interpersonal social adaptation
- Relation to self: Less emotionally regulated and impact of Cognitive Reappraisal
- Sycophancy in GPT 4o, they are too kind and are also manipulative
- GPT 5 reduced these sycophancy issues by 25%
- Adam Raine
- We can fix this via:
- Top-Down approach: Policy and governance
- Bottom-up approach:
- Dont need to learn everything, but learn one thing well
- Process things, go for walks, touch grass
- Collaborate with AI: use it as a starting board and then probe deep into responses