When people assume that AI would have to perfectly replicate a certain brain area (like the prefrontal cortex) to actually be able to “think,” that intuition is wrong on multiple levels.

When we say artificial neural networks are inspired by the brain, we don’t mean a one-to-one structural copy. We mean a different way of solving the same problem.

We built systems that can implement higher-order cognitive functions without rebuilding the same biological parts (like how an airplane doesn’t need to flap its wings like a bird in order to fly).

And as it turns out, both fields are now showing that the functional level is the right level of description anyway.

These two studies recently came across my timeline back-to-back and the theme made me laugh.

The LLM paper basically says to stop looking for the one special part inside a model that handles a given task. The same task gets handled by multiple different internal pathways that overlap and back each other up. There is no single sacred wire.

The neuroscience paper says the same thing about brains. Stop treating named brain regions as the privileged units of cognition. Functions live across gradients, layers, networks, columns, and patches. There is no single box for “the part that does language” or “the part that does emotion.”

Both fields are coming to the same realization from opposite directions. Cognition is distributed.

Which makes the “AI doesn’t have the same specific structures humans have” objection weaker in both directions at once.

Neither field treats the discrete-module picture as the real architecture of cognition anymore.

The right comparison was always systems-level functional organization.

Again. Like I’ve been saying. This whole time. 🫠🙃