"The Predictive Brain: From Perception to Awareness"
Tuesday 16th July 2019, 9am-5pm, at the University of Sussex
Arts A01, Arts Hallway, Quiet Room (Meeting House)
Attendance is free but don't forget to register! Registration is now closed
How is what we see affected by what we know, or think we know? To what extent do our conscious perceptual experiences track sensory data? The conference will bring together researchers investigating the predictive qualities of the brain, including inference mechanisms and the role of prior knowledge and expectations in perception and cognition, and researchers exploring the relationship between sensing, perception and awareness more generally.
Keynote speakers:
- Prof Andy Clark FBA (University of Sussex, Department of Philosophy, Cognitive Philosophy). Bayesing Qualia: Consciousness as Inference, not Raw Datum
- Dr Katerina Fotopoulou (UCL, Research Department of Clinical, Educational and Health Psychology). Mentalising Homeostasis: The Social Origins of Interoceptive Active Inference
- Dr Ryota Kanai (Founder and CEO of ARAYA Brain Imaging, external champion of Sussex's Leverhulme Doctoral Scholarship Programme "From Sensation and Perception to Awareness"). Information Generation as a Core Function of Consciousness
- Dr Rosalyn Moran (King's College London, Centre for Neuroimaging Science). Active Inference in Gaming Environments for Computational Psychiatry.
Click HERE to download the schedule
Click HERE to download the abstracts
The day will also feature doctoral research and post-/doctoral poster presentations.
Abstracts
Andy Clark: Bayesing Qualia: Consciousness as Inference, not Raw Datum
The meta-problem of consciousness is the problem of explaining the set of behaviors and verbal reports that constitute the so-called ‘hard problem of consciousness’. In this talk, I present and defend a solution to the meta-problem, and tentatively suggest that it makes progress with the first-order hard problem too. The solution takes as its starting point recent work in cognitive and computational neuroscience that depicts the human brain as a complex, multi-layer prediction engine. Such a device is constantly striving to generate the sensory signal ‘from the top-down’ using stored probabilistic knowledge about the world. In so doing, it crunches together prior expectations and sensory information from many sources including the physiological state of the body, our own current states of ‘action-readiness’, and distal states of affairs. The detail of this mix (though not the result) is hidden from introspection. This opacity, I argue, explains much that seems puzzling about so-called ‘qualia’. I show, in addition, how it is that advanced agents of this stripe obtain a grip on counterfactual scenarios in which perceptual appearance and reality come apart. Perhaps we being fooled by a Cartesian demon, or are locked in the Matrix? Such agents are then led to represent some of their own highly certain mid-level sensory re-codings as deeply special – known with great certainty while not quite fixing how the world really is - becoming increasingly puzzled and opening the door to all the demons of the Cartesian mindset.
Katerina Fotopoulou: Mentalising Homeostasis: The Social Origins of Interoceptive Active Inference
According to recent, active inference models of self-awareness, interoception has a central role in our sense of self. In this interdisciplinary talk, I will defend the claim that even some of the most minimal aspects of selfhood, namely the feeling qualities associated with being an embodied subject, are fundamentally shaped by embodied, 2nd-person, interactions with other people in early infancy and beyond. Human infants cannot satisfy homeostatic imperatives themselves. Instead, interoceptive predictions are formed only on the basis of interactions with caregivers. Such embodied interactions allow the developing organism to ‘close the loop’ in interoceptive inference. In other words, caregiving interactions are not only keeping an infant alive but they are also the only way for an infant to build predictive mental models of its physiological states, i.e. make interoceptive inferences. To support these theoretical claims I will present a set of neuropsychological and neuroimaging studies with healthy individuals, neurological patients with right-hemisphere damage and patients with anorexia nervosa, showing how self-experiences that rely on epistemically private (interoceptive and proprioceptive) and exteroceptive signals can dissociate and their integration relies on proximal, embodied experiences of affective congruency that act as the ‘emotional glue’ of the senses. Without such integration, self-awareness is either dominated by egocentric, interoceptive priors (as in anosognosia for hemiplegia), or third-person judgements lacking in affective anchoring to the body (as in anorexia nervosa).
Ryota Kanai: Information Generation as a Core Function of Consciousness
The notion that consciousness is a natural phenomenon suggests that consciousness is subject to a set of yet to be uncovered laws of nature. However, conditions in which consciousness occurs remain elusive. Here, we review existing literature in psychology and neuroscience to discern possible functions of consciousness. Drawing upon empirical literature, we propose that a key function of consciousness might be the ability to internally generate sensory representations of events even when they are not happening at the present moment. Such counterfactual representations can be constructed by generative models that an agent learned through sensory-motor interactions with the environment. This ability endows an agent with a variety of cognitive functions related to consciousness such as intention, imagination, planning, short-term memory, attention, curiosity, and creativity, all of which contribute to behavioural flexibility and efficient learning. Using variational autoencoder (VAE) as an example, we illustrate how information generation and processing respectively corresponds to decoding and encoding, or data decompression and compression in artificial neural networks. In biological neural networks, information generation corresponds to top-down prediction, which is compatible with the common observation that feedback is more relevant for consciousness than feedforward processing. Taken together, the information generation theory provides new perspectives on the relationship between information and consciousness.
Rosalyn Moran: Active Inference in Gaming Environments for Computational Psychiatry
The normative rules by which brains make decisions, act and interact with their environments can be formally expressed by mathematical and computational principles. This supports neuroscience efforts, for example in neuroimaging, by providing detailed and latent descriptions of behaviour. It further supports the understanding of abnormal behaviour and its treatment e.g. in the context of psychiatric disorders.
Under Active Inference (Friston 2009), a decision – such as that to move ones’ eyes - is driven by the imperative to minimise a bound on surprise known as the Free Energy. In the context of partially observable Markov decision processes (POMDPs), a model-based framework in which we can cast naturalistic decision-making tasks; the Free Energy of a policy (a sequence of actions) can be understood as a drive to both minimize cost (maximise the likelihood of achieving a goal) while maximising the information return from a given set of actions. This scheme has been used to model decision making in tasks such as ‘the urn task’ and also in reading.
In my talk I will explain technical framework of Free Energy minimization in the context of brain connectivity. I will describe the use of online gaming environments (designed to test artificial intelligence algorithms) and present data from decision-making simulations. For example, how would ‘optimal’ brains play the game ‘Doom’ and compare agents trained under Active Inference to agents trained to maximise reward. Linking these simulations to putative neurobiological substrates I will describe the potential links from brain and neuroimaging to the underlying drives that influence decisions. I will focus on two results from our simulated agents that describe normative rules to understand aging and anhedonia.