Pain in brains and robots

<h3>Biography</h3> Ben is a Wellcome Senior Clinical Fellow at Oxford University, working jointly at the Wellcome Centre for Integrative Neuroimaging and the Oxford Institute for Biomedical Engineering; and a visiting researcher at ATR labs (Kyoto). He is a Fellow at the Alan Turing Institute and an honorary consultant neurologist at Oxford University Hospitals NHS Trust with an interest in behavioural homeostasis, sleep, pain and fatigue. His lab addresses the computational and systems neuroscience of pain. This research is part theoretical: building realistic models of neuronal information processes to understand processes of pain perception and behaviour, and part experimental: testing these theories using a range of experimental methodologies, especially fMRI. His research aims to develop new technology-based therapies for treating pain in clinical populations. Pain has always been a difficult topic in neuroscience: it’s inherently subjective, private and self-intimating. This makes it very hard for clinicians to know the details of what someone’s pain actually feels like. To make matters worse, there are no simple objective biomarkers of pain - such as a blood measure or MRI scan feature. Indeed underlying this is a more fundamental problem: we understand very little about how pain is processed in the brain (much less than simple touch, for example). There is no unique ‘pain cortex’, and processing appears distributed across multiple distant cortical and subcortical regions. This has meant that solving pain is often viewed equivalently to solving the hard problem of consciousness. In this talk I’ll introduce a different approach, based on the fact that pain obviously serves a vital function in healthy humans. Since pain protects us from bodily harm, we can think of it as a control system in engineering. This is particularly important for autonomous systems such as robots, who need to learn based on experience by interacting with the world around them. These systems learn through making predictions - both of events outside of their control, and on the basis of actions they take. Since pain tells you about harmful or dangerous actions, it should shape our behaviour in definable and quantifiable ways. I will explain how building computational models of this, based on control systems engineering, has allowed us to understand much more about the brain circuits that underlie pain behaviour: effectively deconstructing pain into its computational parts. We can then start to reconstruct these parts together again in simulations. This approach also extends to understanding tonic (persistent physiological) pain, and clinical chronic pain. In this case, pain is often part of a broader suite of behaviours that typically accompany injury - such as fatigue, anxiety, lower mood, and sleep disruption. In this context, a lot of ‘multimorbidity’ in behavioural neuropsychiatry makes sense in terms of models of aberrant homeostasis, and allows us to build models of expected clinical symptoms based on normative computational foundations.