Pain is the pinnacle of phenomenal consciousness, and central to our core sense of self and self-preservation. But it still remains a neuroscience puzzle: although numerous brain regions respond to pain, we still have no idea what aspect of brain activity is necessary and sufficient for the conscious perception of pain.

Our research adopts an engineering-based approach to this question: to deconstruct pain into its component parts, and then reconstruct it to see how subjective pain might emerge. This approach considers pain from an information processing perspective, and focuses on the evolved function of pain as driving behaviour to protect and limit bodily harm. At the heart of this is learning, because what makes pain so useful is its ability to act as a teaching signal to minimise not just immediate harm, but also future occurrences when it can be predicted. Consequently, our research has focused on computational models of pain prediction and action (escape and avoidance). In particular, we've argued that Reinforcement Learning - a branch of control theory based on learning about the world through trial-and-error interaction - can offer insight into the computations the brain performs when learning about pain (much as it has been useful in understanding reward). Indeed this has allowed us to build models of the multiple value-systems that drive different control processes in pain-related prediction and control behaviour.

We've also argued that pain perception can be understood as a process of inference, based on a Bayes-approximate process that tries to infer the causes of sensory stimulation. Although this explains several incidences by which pain perception is biased by predictions and expectations, we've shown that it doesn't fully explain subjective perception. This has led to an integrated model of perception and control, in which behaviour is determined along a hierarchy of increasingly sophisticated computational inference-control modules: from simple reflexes in the dorsal horn, to conscious cognitive pain behaviour in the brain. This complex interplay between perception and control is manifest in the many different types of endogenous control - the pain systems ability to intrinsically control what we perceive. Ultimately the picture that emerges is that pain reflects a precise and finely-tuned control signal, that captures an estimate of the overall near-sighted and far-sighted significance of pain to harm mininisation. Accordingly, this signal must be constructed of distinct components of sensory inference, value-estimation, and information content.
From Pain: a precision signal for reinforcement learning and control, Neuron 101(6) 1029-41 (2019)
Practically, the labs work combines theoretical modelling with behavioural and neuroimaging research in humans. Although the focus is on basic science, we also do translational research with a particular interest in musculoskeletal pain, as well as fatigue and anxiety.
 

Current Funding:
  • Wellcome Trust (UK)
  • National Institute of Information and Communications Techology (Japan)
  • Versus Arhtirits (UK)
  • IITP (Korea)
  • MRC (UK)