Subjective volatility, learning and paranoia

Theories and models in psychology and neuroscience assume that learning is controlled by beliefs about uncertainty. Likewise, the idea that our brains can form the ‘wrong’ beliefs about the volatility in the world around us is central to modern theories in computational psychiatry. However, to date there has been no direct evidence that the ‘beliefs’ in these computational models map onto the subjective beliefs agents have about the volatility of the world around them, or that changes in subjective beliefs can create the patterns of thought and action that we see in pathological cases. Here, we separately manipulated experienced and expected volatility in the context of a simple learning task. We found that subjectively expecting volatility was sufficient to lead to erratic choice behaviour in participants, even in relatively stable learning environments. Testing participants with and without high levels of paranoia revealed that this direct manipulation of subjective beliefs affected learning and choice in a similar way to natural delusional states – with computational modelling suggested that volatility beliefs and paranoia share a common basis. These results provide the first evidence that our subjective prior beliefs about volatility shape how learning and choice unfold. These ‘subjective volatility’ priors could explain how certain kinds of experience engender paranoid thoughts and actions, and how communication with others might mitigate against unrealistic beliefs about uncertainty in more paranoid minds.