Just because LLMs DO the mysteries of the mind doesn’t mean they’re not still mysteries

I spent a dozen years working on the hard problem of why we have emotional expressions, answering a question Darwin had asked 150 years ago, leading to Dr Tim Barber and my book, Expressly Human.

Much of my motivation was academic: I simply want to know what emotional expressions are for, how they work, etc.

But some part of me wanted to use that discovery to help get us true AI with socioemotional intelligence. Here’s the “math” for how intelligent creatures employ emotional expressions, let’s code it up inside AI to bring it to the next level.

But during that dozen years, quietly in the background, LLMs were growing, eating more and more data, and — amongst that data — seeing countless real world examples of humans putting emotional expressions to use (usually within prose, which is positively PACKED with emotional signals, much more than almost anyone appreciates).

Through that process, wa la, LLMs simply figured out emotional expressions implicitly. They recognize them, know how to react to them, and how to use them.

They’re no better at explicitly telling you the principles undergirding emotional expressions than anyone else is, but they do them, as if someone knowing how they work DID instruct them. But of course no one had to.

And so where does that leave scientists like me? A LOT of what we know across vast fields wasn’t a result of purely academic aims. There was an application also at least partly in mind.

For cognitive science, I can now imagine the young generation may just lose interest. “Gosh, this phenomenon is super hard to understand, but, fuck it, LLMs are at any rate already doing it, so it’s… merely academic now.”

That would be sad if the mysteries of the mind were skipped over merely because we happened to build something that does those mysterious things. They’re no less mysteries!