Desire, belief, and predicting with our brains

Our brains build a model of ourselves and our world, but how do these models generate our perceptions, actions and decisions? New research from Goldsmiths, University of London investigates.

Dr Daniel Yon recently co-authored a research paper, Beliefs and desires in the predictive brain published in Nature Communications, that attracted widespread attention on social media, particularly from fellow scientists. 

With his co-authors, Daniel (a cognitive neuroscientist and Lecturer in Psychology) argues there is a flaw in influential Bayesian models of the brain – which claim to have found a hidden principle behind all brain functions. He suggests this idea doesn’t work because it leaves key elements of cognition and behaviour unexplained. 

We asked Daniel to explain why his proposal is different, and what this could mean for studying how the brain works. 

Sarah Cox: Can you explain what Bayesian brain theories are?

Daniel Yon: If you cast your mind back to secondary school maths you might remember studying probability and ‘Bayes Theorem’. This branch of maths gives us a way to make inferences about the world by combining new bits of evidence with our prior background knowledge. If you start to hear the pitter patter of water droplets outside but you know it’s going to be a scorching summer’s day you might infer that it’s more likely your neighbour turned their sprinklers, rather than think it’s the beginning of a rainstorm. 

Bayesian theories in psychology and neuroscience suggest this kind of process – testing predictions against incoming evidence – is a good model of how our brains actually work. When we perceive the world around us we combine the actual evidence arriving at our senses (e.g. our eyes and ears) with beliefs about what should be there based on our prior expectations. Some Bayesian models have become particularly ambitious in recent years, suggesting that everything the brain does can be thought of in predictive terms – perception, cognition and action are all ways to close the gap between our predictions about the world and the way it really is. 

SC: What might this ‘predictive processing’ model mean for cognitive science and the broader natural sciences? 

DY: Some hope that one day science will be ‘unified’ – with one framework explaining all aspects of biology, psychology and the social sciences. The concept of prediction is a real boon for this project, because many living systems besides the brain can be described in the same conceptual language.

For example, a biologist could model a single-celled organism like a bacterium as a simple prediction machine. In this way of thinking, we might say a microbe ‘predicts’ to find itself in certain environments (e.g. not too hot or too cold) and behaves in ways that cause this prediction to come true. In this way, both the brain and bacteria are thought as similar kinds of prediction engine. 

This attempt to unify cognitive science (the study of mind and brain) with the wider life sciences could have some downsides. For example, talking only in the language of ‘predictions’ makes it hard to account for some distinctive mental states that humans and other animals have but microbes probably don’t, like beliefs and desires. 

SC: Some neuroscientists and philosophers argue that “desires emerge as a web of prior beliefs”, why do you think this argument is flawed? 

DY: In everyday language we might think of beliefs and desires as explicit thoughts in our heads (“I think there is milk in the fridge”, “I would like to make a coffee”). But lots of cognitive scientists (like me) think that the beliefs and desires that control our decisions and behaviour could be ‘subpersonal’ – mental states that your brain encodes but which you might not necessarily be aware of. The main issue is that beliefs and desires play opposite roles in our mental life. 

SC: So what is the difference between our beliefs and our desires? 

DY: Whether they’re conscious or not, beliefs and desires are different because they have different ‘directions of fit’. A good belief tells you the way the world is, and you should update your beliefs to match the world. In contrast, a desire tells you how the world ought to be, and a desire is fulfilled when you change the world to get what you want.

SC: You argue that the theory that everything is based on prediction is problematic, and retaining the distinction between belief-like and desire-like states is important. Why?

DY: We think the problem arises because you can’t use a prediction in a belief-like and desire-like way at the same time. In these predictive processing theories, to explain how I make a morning coffee we’d say my brain has a prediction like “I am drinking coffee now”. If this state in my brain is like a desire I would need to start changing the world to make it come true - e.g., opening the fridge to get the milk. But if it is like a belief I would need to start changing the prediction so it matches the world as it is right now e.g. updating the prediction to say I’m not drinking coffee now, I’m just standing bleary eyed in the kitchen.  So these theories can’t yet explain how we can monitor our actions (beliefs) while we pursue our goals (desires).

This distinction might be especially important in explaining pathologies of cognition. In drug addiction or conditions like obsessive compulsive disorder there seems to be a decoupling of beliefs and desires about action. Somebody may believe that taking a drug will be unpleasant, or report knowing that a compulsive behaviour (such as flicking light switches) is pointless, but nonetheless still feel strong urges to perform the action. These dissociations are hard to account for without suggesting the brain compartmentalises what we believe from what we desire. 

SC: The new research paper was shared widely on Twitter. Why do you think people were so interested?

DY: The paper did attract quite a bit of attention. I think this is partly because ideas around predictive processing have been really influential in the cognitive sciences over the past few years – and many think this framework presents a ‘paradigm shift’ in how we should think about the mind.

I also think one reason the paper received positive feedback in some quarters was because it tried to bridge the gap between computational models and psychological theory. My hunch is lots gets lost in translation on both sides of this equation since most scientists expert in one of these topics don’t have much training in the other. Trying to put the two approaches side-by-side hopefully means different kinds of researchers can share and compare their perspectives rather than working away in their own disciplinary silos. 

Reaching across disciplines in this way will be really important to advance our understanding of the mind and brain. One of the things I have really valued about being in the Department of Psychology at Goldsmiths since joining last year has been the opportunity to interact with colleagues with different theoretical perspectives who are also very open minded about how ideas can be combined in new ways. 

Beliefs and desires in the predictive brain by Daniel Yon (Goldsmiths, University of London), Cecilia Heyes (University of Oxford) and Clare Press (Birkbeck, University of London) is available to read online: https://www.nature.com/articles/s41467-020-18332-9