Interview, Conference Presentation
What is Intelligence? - François Chollet and Lex Fridman | AI Podcast Clips
- Human intelligence is expected to be specialized for the human experience rather than universal, proving ineffective for programs involving very long-term problems due to limited working memory and difficulty projecting across scales of five, ten, years, or millennia.
- Civilization is anticipated to function as a collective problem-solving and artificial intelligence system operating on a network of brains and infrastructure, where science and computer science tackle problems at scales comparable to thousands or hundreds of thousands of humans.
- Individual human brains are projected to perform poorly on large-scale problems, with a significant portion of intelligence externalized into books, computers, the internet, and other humans, while no strong delimitation is expected between the brain and the body.
- Intelligent agents will not operate as isolated hierarchical systems where a brain controls a body, as the definition of an agent depends entirely on context rather than hard boundaries.
- Exponential growth in solving specific problems is expected within particular verticals, driven by a recursive cycle where science feeds technology to build better tools and instrumentation, which in turn accelerates scientific progress.
- Science is predicted to serve as the closest existing example of a self-improving, superhuman AI, with its progress serving as a basis to understand the behavior of future artificial systems.