I will talk about a formal treatment of choice behaviour based on the premise that agents minimise the expected free energy of future outcomes. Crucially, the negative free energy or quality of a policy can be decomposed into extrinsic and epistemic (intrinsic) value. Minimising expected free energy is therefore equivalent to maximising extrinsic value or expected utility (defined in terms of prior preferences or goals), while maximising information gain or intrinsic value; i.e., reducing uncertainty about the causes of valuable outcomes. The resulting scheme resolves the exploration-exploitation dilemma: epistemic value is maximised until there is no further information gain, after which exploitation is assured through maximisation of extrinsic value. This is formally consistent with the Infomax principle, generalising formulations of active vision based upon salience (Bayesian surprise) and optimal decisions based on expected utility and risk sensitive (KL) control. Furthermore, as with previous active inference formulations of discrete (Markovian) problems; ad hoc softmax parameters become the expected (Bayes-optimal) precision of beliefs about – or confidence in – policies. We focus on the basic theory – illustrating the minimisation of expected free energy using simulations. A key aspect of this minimisation is the similarity of precision updates and dopaminergic discharges observed in conditioning paradigms.

By presenting sequences of sound containing pattern information we can study the brain’s sensitivity to both regularity and regularity violation. In two studies I will demonstrate that the first context in which we encounter a sound can generate enduring differences in cortical responsiveness to this sound. The data derive from studies using sequences in which two sounds of different frequency (study 1) and duration (study 2) alternate roles as a rare pattern violating “deviant” and a common repeating “standard”. The evoked response to the sound first heard as a rare unexpected pattern deviation exhibits larger error signals when encountered as a local deviant and more pronounced suppression when encountered as a local repetitious “standard” compared to the sound first heard as the repeating predictable standard. The data are consistent with a first-impression of sound probability leading to first deviants being tagged as potentially more informative leading to enhanced modulation of evoked responses over time.

The ability to learn about regularities in the environment and to make predictions about future events is fundamental for adaptive behaviour, as it may provide a competitive advantage for anticipating reward or avoiding punishment. In this talk I will demonstrate that people are able to encode statistical regularities in the sensory environment even while their cognitive resources are taxed by concurrent demands and that violations to these regularities evoke sensory prediction errors that engage fronto-temporal networks with recurrent interactions.

The free energy principle focuses on how agents minimize the long term average of prediction error and thereby approximate Bayesian inference. This long term perspective implies that aspects of hierarchical inference and learning will be slow and challenging. This leaves room for individual differences in how well the Bayesian ideal is approximated in the short and medium term, which in turn may be implicated in mental and developmental disorder, such as autism. We review various Bayesian theories of autism and present new data suggesting that non-hierarchical theories are too simple. Instead, there may be differences in more complex hierarchical filtering, suggesting difficulties with context-sensitive inference and action initiation.

The free energy principle focuses on how agents minimize the long term average of prediction error and thereby approximate Bayesian inference. This long term perspective implies that aspects of hierarchical inference and learning will be slow and challenging. This leaves room for individual differences in how well the Bayesian ideal is approximated in the short and medium term, which in turn may be implicated in mental and developmental disorder, such as autism. We review various Bayesian theories of autism and present new data suggesting that non-hierarchical theories are too simple. Instead, there may be differences in more complex hierarchical filtering, suggesting difficulties with context-sensitive inference and action initiation.