François Chollet on Keras, Deep Learning, the Limits of AI, and the Myth of Intelligence Explosion

This episode features François Chollet, creator of Keras and a prominent AI researcher, discussing his controversial views on the future of artificial intelligence, particularly challenging the concept of an "intelligence explosion" or singularity. Chollet argues that intelligence is not an isolated property of a brain but emerges from the interaction between a brain, an embodied system, and its environment. He posits that improving one part of this complex system inevitably leads to new bottlenecks, preventing exponential growth in overall intelligence, even for highly intelligent humans whose brains are often not the primary bottleneck to their achievements.

Chollet further elaborates that all intelligence, including human intelligence, is specialized. Human intelligence is tailored to the human experience, making us adept at certain problems while being ill-suited for others, such as very long-term planning. He introduces the idea of "civilization" and "science" as superhuman, artificially intelligent problem-solving systems that operate on a network of brains and infrastructure, capable of tackling problems far beyond individual human capacity. However, even these recursively self-improving systems, like science, do not exhibit exponential progress in output.

Drawing an analogy to scientific progress, Chollet contends that while the resources consumed by science (e.g., number of researchers, papers, computational power) increase exponentially, the actual output in terms of significant discoveries and knowledge generation tends to be linear. This is attributed to "exponential friction" – the increasing difficulty of making new discoveries as a field matures, requiring exponentially more effort and resources for diminishing returns. He suggests that this dynamic would also apply to any recursively self-improving AI, preventing an uncontrolled intelligence explosion.

The conversation also touches on the psychological appeal of the singularity narrative, likening it to religious mythologies about world-ending events. Chollet shares his personal experience of being surprised by deep learning's capabilities in its early days but clarifies he never believed it was limitless or directly leading to human-level intelligence. He emphasizes that intelligence is multi-dimensional, distinguishing between "human-level" capabilities and "human-like" characteristics. Finally, he provides a brief history of Keras, highlighting its origins in addressing RNN/LSTM challenges, its Python-centric design, and its inspiration from Scikit-learn's user-friendliness.

"intelligence emerges from the interaction between a brain a body like embodied intelligence and an environment"

"I don't think even today for very smart people their brain is not the bottleneck to the sort of problems they can solve"

"intelligence is the meaning of great problem-solving capabilities with a great problem and if you don't have the problem you don't react spreche in intelligence"

"I do believe all intelligence is specialized intelligence even human intelligence has some degree of generality"

"we have this thing called civilization right which is itself a sort of problem solving system a sort of artificially intelligent system"

"a lot of our intelligence is externalized when you write down some notes that is externalized intelligence when you write the computer program you are externalizing cognition"

"science is probably the closest thing we have today to a recursively self-improving super human AI"

"what happens with recursively seven boring system it's typically not explosion because no system exists in isolation and so tweaking one part of the system means that suddenly another pollow system becomes a bottleneck"

"the resource consumption of science is exponential but the output in terms of progress in terms of significance is linear"

"as you make more and more progress this despoiling this recursive self-improvement component you are encountering exponential friction"

"AI is not just a subfield of computer science it's more like a belief system"

"intelligence is very multi-dimensional and so there's the question of capabilities but there's also the question is being human-like and two very different things"

"I loved psychically for its usability so I drew a lot of inspiration from psychic then when I went Cara's it's almost like cycling for neural networks"