Abstract

Many are worried that cognitive scientists have lost our way. Researchers anxious about the ‘theory crisis’ in psychology and neuroscience argue that while we have lots of clever tools for collecting and analysing data about the mind and brain, we urgently lack good theories about the systems we are studying, with current thinking about the mind often seeming wooly and confused. One proposed solution to this problem is to make our theories more ‘formal’ – using computational or mathematical models that are explicit about our assumptions, unlike verbal or narrative theories that are harder to pin down. The hope is that thinking through equations will make our theories more precise. But will it? Here, I will focus one computational idea that has had enormous influence on thinking about the minds of humans and other animals: the idea that the brain is Bayesian. A key idea in these models is that agents combine different sources of information according to their estimated reliability or precision – giving more weight to information that is believed to be more ‘precise’. This concept of precision-weighting is central to a host of classic and contemporary theories across cognitive science – including models of perception, prediction, learning, social interaction and self-awareness. What’s more, many have leveraged this computational principles to explain atypical minds - most prominently psychosis and autism - as arising from unusual forms of precision-weighted inference. However, a key ingredient that makes many Bayesian models work is the idea that agents can adopt beliefs about ‘precision’ that are false. This offers Bayesians enormous flexibility when modelling in health and disease – but may create an illusion of explanation, rather than genuinely ‘precise’ theories. I will argue that Bayesian models will only create useful theories for psychologists and neuroscientists if we can get more precise about how precision works. To illustrate how, I will describe some recent results from the lab which reveal how we form beliefs about the precision of our senses, and the impact this has on metacognition and awareness. This kind of work pinpoints how we can begin to constrain presently unshackled Bayesian models, and may more broadly illustrate how to navigate the promises and pitfalls of thinking through equations.

Reference

Daniel Yon (Birkbeck, University of London), “Explaining the mind with precision?”, IAST General Seminar, Toulouse: IAST, April 1, 2022, 11:30–12:30, room Auditorium 4.