Summary: Researchers apply a mathematical framework to clarify healthy brain function and to explain how neuropsychiatric disorders may arise. They propose that problems in processing prior beliefs can cause incorrect interpretations of sensory data, producing pathological perceptions and behaviors.
Source: Frontiers
Every second, our brains receive a flood of information from the five senses. To act effectively, the brain must interpret these inputs, integrate them with past experience, and generate appropriate responses. Even simple actions rely on millions of neural connections and complex computations across many cell types.
While neuroscience has made great strides in understanding individual neurons, explaining how large networks of neurons work together to produce healthy cognition and behavior—and what goes wrong in brain disorders—remains a major challenge. A new review by researchers at University College London examines a promising computational framework that uses Bayesian inference to model brain function and to link neural computation with clinical symptoms.
Modeling brain function with computational theories
Early insights into brain function came from studying people with localized brain damage, a method that revealed functional specializations across regions. However, cognition depends on widely distributed networks, so damage in one area can disrupt distant functions, limiting what lesion studies alone can show.
Computational theories offer a complementary approach. By expressing perception and action as formal calculations, these models describe the computations the brain must perform to produce observable behavior. They let researchers simulate neural dynamics, generate testable predictions, and compare alternative hypotheses about how circuits implement cognition. As Professor Karl Friston and colleagues emphasize, computational models—especially those based on Bayesian principles—have become central to formal approaches for understanding psychiatric and neurological disease.
Prior beliefs and Bayesian inference
The review focuses on Bayesian inference: a statistical framework in which the brain combines incoming sensory data with prior beliefs to infer the most likely causes of sensations. In everyday life, perception is not a passive record of stimuli; it is an interpretation shaped by prior knowledge and expectations. For example, seeing a person with long hair from behind leads you to assume the person is female because that prior belief is often a good predictor. If later evidence (such as seeing the person enter a men’s restroom) contradicts that assumption, the brain updates its priors and revises the inference.
Bayesian models formalize this balance between priors and sensory evidence. They specify a generative model—how causes produce sensations—and then invert that model to infer causes given sensations. This framework captures both perception and action as inferential processes and provides a principled way to describe prediction, surprise, and learning in the brain.
When inference goes wrong: abnormal perceptions and behavior
Viewed through the lens of Bayesian inference, many neuropsychological symptoms can be conceptualized as errors in inference. These errors may stem from inappropriate prior beliefs (priors that poorly match the world) or from faulty processing of priors, leading to misattributions about the causes of sensory inputs. Such misinferences can produce pathological sensations, beliefs, or actions.
One illustrative example is phantom limb pain experienced by amputees. If the brain retains a strong prior belief that the limb remains present, the absence of expected sensory feedback cannot correct that belief, and the person continues to experience sensations from the missing limb. Similarly, hallucinations and delusions may arise when prior expectations exert excessive influence over perception, producing false positive inferences. Motor disorders such as Parkinson’s disease could also reflect aberrant beliefs about how and when the body should move, disrupting the brain’s predictive control of action.
Conversely, some theories propose that autism involves overly weak priors. If prior beliefs carry less weight, autistic individuals may rely more heavily on current sensory input than on past experience, which could explain differences in sensitivity to unexpected or ambiguous stimuli.
Implications for diagnosis and personalized treatment
One of the most exciting outcomes of this computational perspective is the possibility of quantitative, individualized characterization of neuropsychological disorders. If a patient’s symptoms can be described by a particular set of aberrant priors or inference processes, clinicians could use model-based measures to phenotype individuals and guide targeted interventions. This approach aligns with precision medicine: diagnoses and therapies tailored to the computational and biological profile of each person.
However, two major challenges must be addressed. First, researchers need robust methods to infer the specific prior beliefs that best explain a patient’s behavior. Second, it is essential to link those computational priors to their biological implementations in neural circuits and neurotransmitter systems. Progress on both fronts—identifying patient-specific priors and mapping them to brain mechanisms—will be required before computational neuropsychology can fully inform clinical practice.
Conclusions
Bayesian and generative models provide a powerful framework for understanding how the brain integrates sensory data and prior knowledge to generate perception and action. By framing neuropsychological symptoms as optimal inference under suboptimal priors, computational neuropsychology offers new insights into the origins of hallucinations, delusions, motor disorders, phantom sensations, and developmental differences such as autism. With continued advances in modeling and neural measurement, this approach holds promise for developing quantitative diagnostics and personalized treatments that target the computational roots of brain disorders.
About this research
Authors: Thomas Parr, Geraint Rees, Karl J. Friston.
Article: “Computational Neuropsychology and Bayesian Inference.”
Source: Frontiers in Human Neuroscience (open access).
Publisher: NeuroscienceNews.com adaptation.
DOI: 10.3389/fnhum.2018.00061