Predictive processing in the brain

Post date: Oct 20, 2012 9:16:26 AM

Is the brain essentially a predictive machine?

Recently, several scientists have argued that the view of brains as stimulus-processing devices is too restrictive. Rather, the brain is dominated by internal processes of goal setting, preparation to action, production and evaluation of expectations, generation of short- and long-term predictions.

Many theories have been proposed that see the brain as an essentially prediction-based or prospective device (or at least that give prediction a prominent role). Below I list a few:

  • The “predictive coding” approach (which is nowadays popular in its “free energy” formulation, Friston, 2010) argues that the brain builds generative models of the body and the sensorium (using perceptual and proprioceptive information). The central concepts of this theory are that generative models produce predictions that guide perceptual processing in a top-down matter, that predictions can be set as “fake observations” in an inferential process, and thus become goals that steer goal-directed action (aka active inference). The idea of using generative models for perception entails that it is a prediction, not a sensation, which steers brain processing (see also Hinton, 2007 for generative models).

  • The idea of implementing planning and action by using “future observations” as goal states (that the system strives to achieve) has equivalents in the duality of inference and control (Todorov, 2004), “planning as probabilistic inference” (Toussaint, 2009), and the ideomotor theory (Hommel et al., 2001).

  • The theory of proactive brains (Bar, 2009) focuses on perceptual processing and argues that it continuously generates predictions and analogies with previous events. It emphasizes that the brain is not only predictive but also proactive in its continuous use of prediction to imagine and evaluate future hypotheses.

  • Our proposal of the mind as an anticipatory device (Pezzulo, 2008, 2011; Pezzulo and Castelfranchi, 2007, 2009) emphasizes proactivity, anticipatory behaviour, and goal-directedness of choice. Anticipations are defined as those predictions relative to events that have behavioural meaning for an organism, and an associated value.

  • Along similar lines, several scientists are now focusing on future-directed abilities such as prospection, mental time travel, and self-projection in the future, linking them to the reuse of episodic memories (Buckner et al., 2007; Schacter et al., 2007, 2012).

  • Another framework that is very popular in computational neuroscience is the inverse-forward modeling scheme (Wolpert and Ghahramani, 2000). In this view, forward models produce to internal loops that parallel the external, feedback loops, supporting state estimation, perceptual processing, and action control in the face of environmental uncertainties. The inverse model describes how the brain learns to control its actions, and the forward model describes how it predicts the sensory effects of such controls.

  • Motor theories of cognition use the theory of internal modeling to develop the idea of “simulations of actions” (Jeannerod, 2006). Simulations of action occur by reusing the internal models employed in motor control, and at the same time suppressing incoming sensory stimuli, and inhibiting the overt execution of motor commands. In brief, predictions generated by forward models are supplied to inverse models, which in turn supply efferent copies of motor commands to forward models; this process can chain multiple predictions.

  • Several scientists view "action simulation" as a unifying phenomenon that can explain multiple abilities, including planning, prediction of external events, visual and speech perception, mindreading, and imitation, and the interference between concurrently executed and perceived actions (Jeannerod, 2006). Philosophers have argued that internal modelling loops (called emulations) are representational (Grush, 2004).

  • Anticipatory brain dynamics (linked to both expectation and preparation) are studied by numerous scientists (see e.g. Engel et al., 2001).

  • A book by Chris Frith (2007) "Making up the Mind: How the Brain Creates Our Mental World" provides a simple introduction to some of the concepts discussed earlier.

The current literature on prediction and anticipation (in living organisms and robots) is much wider; see this webportal on anticipation. It touches many fields, including cognitive psychology and neuroscience, computational modeling, robotics, philosophy, etc.

I am actively promoting the idea of the anticipatory nature of brain and cognition.

I authored more than 30 publications on this theme (including 3 edited books), with theoretical, computational, and empirical studies. I promote several activities aimed at fostering the interdisciplinary study of prediction and cognition: (i) organizing three international workshops on “Anticipatory Behavior in Adaptive Learning Systems” ABiALS (2006, 2008, 2010); (ii) editing three special issues: “Anticipation and anticipatory behavior” in Cognitive Processing, 8(2) and 8(3), (2007); “Intentional action: from Anticipation to Goal-Directed Behavior” in Psychological Research, issues 4 and 5, 2009; “The anticipatory construction of reality” in new Ideas in Psychology (forthcoming); (iii) editing three books on the themes: “Anticipations, Brains, Individual and Social Behavior” Springer LNAI 4520, (2007); “The Challenge of Anticipation”, Springer LNAI 5225, (2008); “Anticipatory Behavior in Adaptive Learning Systems”, Springer LNAI 5499, (2009); (iv) maintaining the webportal anticipatorybehavior.org.

Selected pubs:

  • Pezzulo G., Rigoli, F. Friston, K. (2018) Hierarchical Active Inference: a Theory of Motivated Control. Trends in Cognitive Sciences [link]

  • Pezzulo, G., Rigoli, F., Friston, K. (2015) Active Inference, homeostatic regulation and adaptive behavioural control. Progress in Neurobiology 134, 17-35 [link][pdf]

  • Pezzulo, G. (2007). Anticipation and future-oriented capabilites in natural and artificial cognition. In 50 Years of AI, Festschrift, LNAI 4850, pages 258–271. [pdf]

  • Butz, M. V., Sigaud, O., Pezzulo, G., and Baldassarre, G., editors (2007). Anticipatory Behavior in Adaptive Learning Systems, From Brains to Individual and Social Behavior, volume 4520 of LNAI. Springer. [link]

  • Pezzulo, G., Butz, M. V., Castelfranchi, C., and Falcone, R., editors (2008). The Challenge of Anticipation: A Unifying Framework for the Analysis and Design of Artificial Cognitive Systems. LNAI 5225. Springer. [link]

  • Pezzulo, G., Butz, M. V., Sigaud, O., and Baldassarre, G., editors (2009). Anticipatory Behavior in Adaptive Learning Systems: From Psychological Theories to Artificial Cognitive Systems. LNAI 5499. Springer. [link]

Other pubs:

  • Engel, A. K.; Fries, P. & Singer, W. (2001) Dynamic predictions: oscillations and synchrony in top-down processing Nature Reviews Neuroscience, 2, 704-716

  • Friston, K. (2010) The free-energy principle: a unified brain theory? Nat Rev Neurosci, 11, 127-138

  • Jeannerod, M. (2006) Motor Cognition Oxford University Press

These are two of my (rather old) videolectures: Seeing the world through the lens of a predictive brain at BCBT 2012, Barcelona, and a Presentation of the Goal-Leaders (Goal-directed, Adaptive Builder Robots) EU-funded project at CogSys 2012, Vienna.

This is a podcast of my interview on predictive brains at BCBT 2012, Barcelona.