Show Notes
When an LLM answers a question, is it reasoning like humans, or just producing text that looks like reasoning? The distinction isn’t just philosophical, this determines what we can trust AI to do, how closely we need to supervise it, and ultimately what its real-world impact will turn out to be.
Melanie Mitchell at the Santa Fe Institute argues that we lack adequate methods for measuring machine cognition, and that AI is a form of “alien intelligence” that operates through non-human cognitive mechanisms. In this episode of The Joy of Why, Mitchell tells Steven Strogatz how methods that psychologists use to study cognition in other kinds of “alien intelligence” — babies and animals — can be adapted to probe AI, and she lays out six principles for better assessing machine cognition. Their conversation ranges from the challenge of interpreting what’s happening inside these systems, to recent AI-assisted breakthroughs in mathematics, to why a math-performing horse from the early 1900s offers a cautionary tale for how we assess intelligence.
More from The Joy of Why
People may think of soil as just dirt, but it is anything but inert. Beneath your feet, a vast network of mycorrhizal fungi is hard at work, trading minerals with plant roots. This trading system helps enrich the soil, and it sequesters 13 billion tons of atmospheric carbon dioxide each year — roughly equivalent to one-third of the world’s fossil-fuel emissions. Scientists know that these fungi form partnerships with 80% of all plant species on Earth, a relationship that evolved some 450 million years ago. What they have not understood is how organisms with no brain or central nervous system can decide whom to trade with, and on what terms.
We know all about how things exist in the three dimensions of length, width, and height. Physicists often talk about time as a fourth dimension, but what if there were a fourth spatial dimension — another direction entirely? How on earth would we picture that?
Quantum mechanics may be one of the most precisely tested paradigms in science, but there’s still no universally accepted interpretation of what it tells us about reality. How do we get from the wave-like behavior of quantum systems to the solid, macroscopic world of objects and the universe at large?
Mathematicians are witnessing a profound shift in their field. AI systems are now producing proofs, spotting connections between distant fields, and, in a few cases, solving problems that had stumped mathematicians for decades. The pace of progress over the past several months has raised challenging questions: What can AI systems actually do? And what happens to the more human, creative aspects of the field once machines can match – or exceed – people at problem-solving?
Pain and pleasure seem like simple facts of life, but they are far from it. Neuroscientists still cannot say why physical pain differs from psychological pain, for instance, nor why a loved one’s touch soothes while a stranger’s touch repels.