Interview, Fireside Chat
Cognition Is a Function of the Environment | Matt Botvinick and Lex Fridman
- A Turing machine is characterized as a universal machine emulator capable of emulating any other machine.
- Human brains are viewed as capacity-limited computational devices that adapt behaviors, yet the specific basic computations driving the full richness of human cognition remain undefined.
- There is a stated requirement to understand the computational mechanisms behind complex behaviors like arithmetic and chess, not just simple tasks, before human cognition is considered fully explained.
- Group behavior is identified as a form of collective intelligence where large populations converge on ideas without physical co-location, a phenomenon linked to the concept of "distributed convergence."
- The speaker expresses skepticism regarding the definition of intelligence as a property of communities, maintaining a preference for defining the individual human as the basic unit of intelligence.
- Counter-arguments suggest that human intelligence may not arise in isolation, noting that a single human in isolation would not develop the cognitive complexity of Homo sapiens.
- Understanding human intelligence requires equal attention to the structure of the environment and the cognizing system (brain or AI).
- Cognition is shaped significantly by the interaction between brain architecture/learning algorithms and the structure of the world in which they are embedded.
- In reinforcement learning, an agent can only learn to the extent that it can simulate its environment.
- DeepMind research highlights the self-play mechanism, where a competing agent acts as the environment, as an essential driver for successful learning.
- Competitive behavior between agents is posited as a fundamental requirement for learning in specific AI contexts.