Melanie Mitchell on the Nature of Intelligence, Analogy, and the Future of AI

This episode features Melanie Mitchell, a distinguished computer science professor and external professor at Santa Fe Institute, discussing the multifaceted field of artificial intelligence. The conversation begins by dissecting the term "Artificial Intelligence" itself, highlighting its inherent ambiguity and the historical context of its coining by John McCarthy, who later regretted it. Mitchell emphasizes that intelligence is not a clearly defined concept, leading to a moving goalpost where tasks once considered hallmarks of human-level intelligence (like playing chess) are re-evaluated once machines achieve them. She posits that understanding our own minds is a prerequisite for creating truly intelligent machines, suggesting that the process of building AI will inevitably reveal more about our own "machine-like qualities" and the emergent nature of intelligence from underlying cellular processes.

A significant portion of the discussion revolves around the distinction between narrow AI, exemplified by current deep learning successes, and the elusive goal of Artificial General Intelligence (AGI) or human-level intelligence. Mitchell expresses skepticism about whether current brute-force, big-data, and huge-network approaches alone can lead to AGI, suggesting fundamental limits and weaknesses, particularly in supervised learning. She introduces the "Copycat" cognitive architecture, developed with Douglas Hofstadter, which places analogy-making at the core of human cognition. This leads to Hofstadter's powerful mantra: "Without concepts there can be no thought and without analogies there can be no concepts," underscoring the critical, yet unsolved, problem of how AI can form and fluidly use concepts.

The podcast delves into the diverse opinions within the AI community, from the singularity/transhumanism camp predicting exponential progress to those who believe deep learning will scale all the way, and others like Yann LeCun who advocate for unsupervised learning breakthroughs. Mitchell also highlights the "hybrid view" (Gary Marcus) combining symbolic and connectionist approaches, and the push for incorporating causality, developmental learning, intuitive physics, and the crucial role of embodiment. She predicts that human-level intelligence is "more than a hundred years away," requiring "a hundred Nobel Prizes" worth of fundamental discoveries beyond mere scaling of current methods.

Broader implications are explored, including the psychological and philosophical challenges humanity faces when confronted with machines that surpass human capabilities in areas traditionally considered unique to humans, such as creating beautiful music or art. Mitchell reflects on the deep-seated human drive to create artificial life, suggesting it stems from a desire for self-understanding. She also broadens the definition of intelligence, viewing it as a continuum present in various complex systems like ant colonies or the immune system, while acknowledging human intelligence's unique capacity for self-reflection. The conversation ultimately frames AI development as a journey not just of technological advancement, but of profound self-discovery and redefinition of what it means to be intelligent and human.

"I'm not crazy about the term [Artificial Intelligence] I think it has a few problems because it it's means so many different things to different people and intelligence is one of those words that isn't very clearly defined either."

"once a machine can do some task we then have to look back and say oh well that changes my understanding of what intelligence is because I don't think that machine is intelligent at least that's not what I want to call intelligence"

"I don't see any reason why we couldn't in principle create something that we would consider intelligent"

"one of the things that that's going to produce is is making us sort of understand our own machine like qualities that we in a sense are mechanical"

"if I had to bet on it I would say no we we we do have to understand our own minds at least to some significant extent but it I think that's a really big open question"

"without concepts there can be no thought and without analogies there can be no concepts"

"how to form and fluidly use concepts is the most important open problem in AI"

"human level intelligence is a hundred Nobel Prizes away"

"I think there's some fundamental limits to how far they're gonna get [current deep learning approaches]"

"deep learning as its formulated a completely lacks any notion of causality and that's dooms it"

"without having a body it's gonna be very hard to learn what we need to learn about the world"