The brain is a Peirce engine

There comes from Scott Alexander’s blog news of a new unified theory of neural cognition called the “predictive processing model”. Read his review of the book “Surfing Uncertainty” before proceeding further.

This model seems to solve a whole raft of longstanding problems about how the brain does what it does, offer insight into how various neurotransmitters work in cognition, and even into how disorders such as autism can be understood as consequences of very specific processing failures with testable consequences.

Now excuse me while I spike a ball in the end zone and yell “YEEHAA!”. Because, although its framers seem still unaware, the predictive-processing model tends strongly to confirm a set of philosophical positions I’ve been taking (and taking flak for) for many years.

Specifically, under the predictive processing model, the brain is a Peirce engine. “Mind” is what we observe as the epiphenomenon of that engine running – its operating noise, more or less.

The Peirce I’m referring to is Charles Sanders Peirce. In his seminal 1878 paper On Making Our Ideas Clear he recast “truth” as predictive accuracy, asserting that our only (but sufficient warrant) for believing any theory is the extent to which it successfully anticipates future observations.

This insight was half-buried and corrupted by later analytic philosophy, notably when William James and John Dewey vulgarized “what is predictive” into “what is useful to believe” and invented the whole sorry mess called Pragmatism.

As a result, the incisiveness of Peirce’s insight was largely forgotten for most of a century except by specialists in the philosophy of science who used it to construct the new commonly accepted explanation of what we mean when we assert that a scientific theory is true, Or false; Karl Popper’s falsifiability criterion is another not-quite-right approximation of Peirce.

What Peirce tells us perhaps best expressed in an antinomious way: There is no “Truth”, only prediction and test.

And now it turns out this is what the brain is doing, all the time, at the neural-circuitry level. Endless waves of top-down expectations crashing against endless waves of bottom up sense data, interpreting predictive failure as unwelcome surprise. Knowledge emerging as a constant Bayesian update of priors at the collision face.

OK, that oversimplifies. What PP actually tells us is that there isn’t any one collision face but many, scattered all through the nervous system. feeding each other. Some in the retina of the eye, for example, where first-stage visual processing is done.

PP boldly says this collision-and-update is not a metaphor and not an approximation of lower-level neural processing of a different kind – it’s the actual computation that the actual meat substrate of your mind is doing in hardware.

I think this is right. It explains so much – and it’s Peirce having the last laugh.