A Lack of Imagination in the Predictive Mind?

Max Jones is a Teaching Associate at the University of Bristol, lecturing on courses in Philosophy of Psychology and Metaphysics. His research focuses on the impact of recent developments in the cognitive sciences for traditional issues in metaphysics and epistemology. This post is based on a forthcoming paper written alongside Sam Wilkinson of the University of Exeter.

Max Jones is a Teaching Associate at the University of Bristol, lecturing on courses in Philosophy of Psychology and Metaphysics. His research focuses on the impact of recent developments in the cognitive sciences for traditional issues in metaphysics and epistemology. This post is based on a forthcoming paper written alongside Sam Wilkinson of the University of Exeter.

A post by Max Jones.

The Predictive Processing framework (PP) has become increasingly influential in recent years, with some claiming that it provides a grand-unifying theory of mental function, explaining perception, action and all cognitive processes in between (Clark 2015; Hohwy 2013). Some proponents of PP have claimed that it is particularly well-placed to explain imagination (Clark 2015 ch. 3; Kirchhoff 2018). This optimism is partly based on the idea that imagination-like processes, where agents endogenously generate content, are somewhat ubiquitous, playing a role in our everyday interactions with the world through perception and action.

The groundbreaking insight of PP is to turn the traditional conception of perceptual processing on its head. On a more traditional account, perception involves the bottom-up processing of information from sensory receptors so as to build an increasingly detailed model of the environment. According to PP, perception is primarily an anticipatory process, with the brain constantly generating top-down predictions about the sensory information it expects to receive, with bottom-up information merely serving to correct errors by conveying the discrepancies between the predictions and reality. One way of thinking about this is that we are constantly in the game of imagining how we expect the world to be and checking if our imagined world conforms to reality.

One might be tempted to think that PP’s commitment to the ubiquity of imagination-like processes would make an explanation of imagination relatively straightforward. After all, the notion of endogenously generating content need not be seen as mysterious is any way. However, the ubiquity of these imagination-like processes makes it hard for proponents to explain what’s special about the distinctive metal act of imagining (Jones & Wilkinson (forthcoming)).

One problem that arises for proponents of PP lies in explaining how we are able to maintain imaginative content. PP suggests that discrepancies between endogenously generated content and reality are corrected for by incoming sensory error signals. However, imaginative content, almost by definition, diverges from one’s current experience of the environment. One rarely, if ever, imagines precisely what one is seeing. The idea that sensory inputs interfere with imagination is certainly intuitive, since many find it easier to maintain an imagined image with their eyes closed. However, the problem for PP is that it seems unable to explain how imaginative episodes are ever maintained at all.

Proponents of PP have a readily available response to this problem. The brain does not just predict what it expects to receive from its sensory receptors, it also makes predictions about the quality of information that it will receive and weights the information accordingly. This supposedly accounts for the role of attention in perception. For example, at a noisy party, one will down-weight auditory information, expecting it to be less precise. In the case of imagination, this down-weighting mechanism could help to explain how imaginative content could be maintained. When imagining, the brain may down-weight incoming sensory information based on the expectation that it lacks precision relative to the content being imagined.

The problem with this response is that it makes it mysterious as to how deliberate acts of imagination can be controlled. An oft-cited slogan of PP is that “perception is controlled hallucination”. We are continually in the process of hallucinating a world, with our hallucinations being kept in check by our error-correcting contact with the world. Once the sensory signals are dampened, the question arises as to what keeps the content under control. This is particularly salient in the case of imagination, as deliberate acts of imagining are very different from wild hallucinations. We engage in deliberate acts of imagination to achieve specific goals. We can purposefully imagine in order to acquire knowledge or to create engaging fiction. Once the control of contact with the world is removed, what constrains our imaginative episodes so that they are able to fulfill relevant goals?

Again, PP can potentially respond to this problem. PP is not just a novel theory of perception, it also provides a novel account of action (see Adams, Shipp & Friston 2013). PP suggests that the fundamental aim of organisms’ neural systems is to reduce the discrepancy between their models of the world and the actual state of the world. However, updating the model in light of error signals, i.e. perception, is just one of two ways of accomplishing this goal. Another way to achieve this is to change the world to fit to one’s expectations by acting on it. PP provides an account of action where the brain predicts the sensory consequences of an action in order to bring the action to fruition. For example, by predicting that one’s arm is raised rather than by one’s side and by dampening the error signals from the body that tell the brain that one’s arm is still by one’s side, the motor system is activated so as to bring the arm into the expected position. At any one time, there will be a whole range of actions available, some of which will be mutually exclusive. To choose between actions, the brain must represent the consequences of various actions, some of which will never actually occur. Thus, basic motor control already involves representing the consequences of imaginary actions.

Rather than seeing imagination as perceptual prediction without control from sensory inputs, proponents of PP can see imagination as a form of offline action (Pezzulo 2012, 2017). Organisms can chain together series of predictions about what would happen if they were to act in a particular way, giving rise to imaginative episodes that diverge from their immediate surroundings. This picture seemingly works well for relatively mundane imaginative episodes. For example, when imagining whether a piece of furniture might fit through a door, one can chain together expectations regarding the results of carrying out relevant actions on the sofa.

This is certainly a more promising attempt to explain imagination through PP. However, it arguably misses out on the most interesting and important types of imaginative episode. Much of our imaginative engagement far transcends just imagining actions that one is able to do. Imagination encompasses so much more than mundane action planning. The most interesting cases of imagination, such as developing novel thought experiments to spur scientific discovery or creating fantastical worlds of fiction, involve radical departures from our everyday interactions with the world. Yet, by explaining imagination as mere simulation of action, proponents of PP miss out on the aspects of imagination that many hold most dear. It would seem that the radical departures from reality that are central to paradigmatic cases of deliberate use of the imagination would generate significant error signals as a result of diverging from the way in which our interactions with the world usually pan out. PP suggests that the brain aims to avoid surprising scenarios, but then one must ask how it is possible to radically depart from reality without surprise.

PP is, without doubt, a ground-breaking theory that is in the process of revolutionising our understanding of perception and action. However, as things stand, it falls short of being able to explain key features of human experience, such as imagination. PP does well to explain how we are able to stay in touch with the dynamic and chaotic environments that we inhabit. However, staying in touch with the world is not all that we need to do. Sometimes we explicitly aim to build worlds in our minds that diverge from our mundane contact with reality.


References

Adams, R. A., Shipp, S., & Friston, K. J. (2013). Predictions not commands: active inference in the motor system. Brain Structure and Function218(3), 611-643.

Clark, A. (2015). Surfing uncertainty: Prediction, action, and the embodied mind. Oxford University Press.

Hohwy, J. (2013). The predictive mind. Oxford University Press.

Jones, M. & Wilkinson, S. (forthcoming). From Prediction to Imagination. In A. Abraham (ed.) The Cambridge Handbook of the Imagination. Cambridge University Press.

Kirchhoff, M. D. (2018). Predictive processing, perceiving and imagining: Is to perceive to imagine, or something close to it?. Philosophical Studies175(3), 751-767.

Pezzulo, G. (2012). An Active Inference view of cognitive control. Frontiers in Psychology3, 478.

Pezzulo, G. (2017). Tracing the Roots of Cognition in Predictive Processing. In Metzinger & Wiese (eds.) Philosophy and Predictive Processing. Johannes Gutenberg-Universität Mainz.