The Explanatory Power of Convergent Models

There’s interesting research emerging comparing our ever-improving machine learning models to data generated from brains. Not surprisingly, we do best at replicating what the brain can do when the computer models begin to model the underlying neural machinery. Now the substrate is entirely different, but the predictive approaches appear to be similar. We used to think that the brain was creating representations of the world, with features being abstracted at each level of cortical processing.

The problem that everyone saw from the beginning with this concept is that there’s no little man in the theater of the mind to look at the representations. Instead, the brain is representing hypotheses and the these predictions are constantly updated by the stream of incoming sensation from the world.

Those hypotheses — and not the sensory inputs themselves — give rise to perceptions in our mind’s eye. The more ambiguous the input, the greater the reliance on prior knowledge.

And so too with language, the brain guesses what word comes next, providing a fixed interpretation when we hear unclear or ambiguous language. Of course, we often make wrong decisions, famously about song lyrics (e.g. “There’s a bathroom on the right”) known as Mondegreens.

I’m beginning to appreciate just how important this is as our ability to look at brain activity improves just as our computational ability to create these models begins to match it. We’re not recreating higher brain function from the bottom up by understanding circuits and connections, but instead from the top down. Perhaps not surprisingly, as this is how physical sciences like chemistry and physics have advanced. They create formulas and equations that are mathematical models of the world that have remarkable predictable powers. Once systems get too complex, these methods seem to fall apart and numerical simulation seems to be needed, but nevertheless, when those models start converging on the behavior of the real thing, they seem to tell us about what’s actually going on in that complex system being modeled. Truly a remarkable time for brain science.

Leave a Comment

Your email address will not be published. Required fields are marked *