The Predictive Mind, Jakob Hohwy, Oxford University Press, 288 pp, £22, ISBN: 978-0199686735
How does the mind work and what makes people behave as they do? Most of us ask these questions. Philosophers, psychologists, cognitive neuroscientists and novelists all contribute, in different ways, to our understanding of the mental faculties. The interiorised fictional narratives of Henry James’s later books and the psychological writings of his elder brother William were part of the same movement towards modern conceptions of consciousness and motivation. Professional work as varied as those of psychiatrists, economists and politicians requires assumptions about the operational settings of the mind.
For human intelligence ever to succeed in exposing the source of its own power, many of these strands of thought, both scientific and non-scientific, will need to be drawn together. One asset of Jakob Hohwy’s new book, The Predictive Mind, is that it does employ such a multidisciplinary approach. His own academic background is philosophical, but he takes new knowledge about computational models of the brain and research in experimental neuropsychology to present an inclusive theory about the way that the higher centres in the cerebral hemispheres process incoming information. Recent advances in the field of artificial intelligence have shown how recognition and learning can best be replicated by “top down” models of organisation, and Hohwy offers an account of the brain thus configured as a hypothesis-testing mechanism to interpret the external world. This predictive character of perception may also, he proposes, explain how we experience consciousness itself.
Much is known about the anatomy of the nervous system, and of the electrical and chemical signalling by which groups of nerve cells interact. But the substrate of our conscious selves ‑ rational thought, emotions, sensations and memories ‑ is not well understood. The relatively young specialty of cognitive neuroscience seeks to bridge this gap, and it has two broad subdivisions. Cognition can be investigated with “biological” tools ‑ neuroimaging, neurophysiology (electrical recording from or electromagnetic stimulation of the cerebral hemispheres) and neurogenetics. Magnetic resonance imaging techniques show how areas of the brain are activated during certain mental tasks. Anatomical patterns of neurological disease are studied for clues about the localisation of cognitive deficits. But circuit diagram models of cerebral function that originate from this body of research probably underestimate the degree to which the brain is performing as an integral network. The other school, to which some of the research cited in The Predictive Mind belongs, considers the brain as a “black box” independent of its biological identity, and tries to approximate facets of human cognition using mathematical symbols and computer programme language. Both approaches rely on the behavioural science of psychology to isolate and test elements of higher brain function so that experiments can be devised.
The subject of sensory perception is central to The Predictive Mind, and a brief description will help to introduce the book’s main hypothesis. A conventional notion of the sensory system is of an ascending stream of information, analysed then passed upwards at a number of levels. To take vision as the example: the optic nerve transmits comparatively simple messages about the topography of light falling on the retina. The nerve cells in the primary visual cortex on the surface of the brain that receive this input are wired to discern linear orientations of light and dark. Adjacent areas of the cortex contain cells that respond to increasingly complex shapes and eventually, enough reprocessing occurs for the brain to recognise the image. Hohwy asserts that this hierarchical scheme is much more bidirectional, with the highest centres in the cerebral cortex actively involved in predicting the external causes of incoming signals and modifying the subordinate processing levels to minimise divergence between expectation and input. To do this, the nervous system has the capacity to apply Bayes’ probability theorem. On encountering something that looks like a bicycle, the theorem would say: the likelihood that the decoding of the visual image as a bicycle is correct depends on the degree of fit of the incoming signal with internal templates of bicycles multiplied by the prior probability of seeing a bicycle. When reasoning or making decisions, we often use an interpretation of Bayes’ formula to evaluate our confidence in a proposition before and after encountering some new piece of evidence. But Hohwy has the theorem built into our nerve circuitry, operating in an automatic, unconscious fashion. A similar Bayesian statistical principle is used in computer programmes for image recognition, allowing adjustment of the probability estimates of hypotheses as new data is being acquired. According to The Predictive Mind, the brain, at its highest levels, is involved in perceptual inference and a constant quest to minimise prediction error by modulating the lower processing areas, tuning and weighting their output against an internal model of the world. This precis does not really convey how dynamic, intricate and simultaneous the system has to be, but in Hohwy’s hypothetical framework, perception depends strongly on prior belief.
In his introduction, Hohwy lists philosophers, neuroscientists, psychologists, psychiatrists, cognitive and computer scientists, and anyone interested in the nature of the mind as his audience. It is not an easy book to read without some grasp of several of these areas, and The Predictive Mind is written at a level well above a popular science publication. The statistical and computer science that underpin his model are simplified in summary or appear in footnotes, but a little knowledge of mathematics and thermodynamics is useful. There are many philosophical themes, which Hohwy is good at explaining, often in historical context. Research on visual and tactile sensory illusions provides support to the prediction error minimisation hypothesis. The sections about this evidence are interesting, and the relevant technical material is quite well described and illustrated. There is only a modest amount of correlation with other areas of brain research, and attempts to link his conclusions with scientific knowledge about the structure and function of the brain do not carry a great deal of authority.
The opening third of the book outlines the predictive perception theory, and the idea is then extended to action and expected experience. Passive and active cerebral tasks that are usually considered to be separate are conjoined by the presumption that in both, the brain is trying to minimise surprise from incoming information. When the heavy suitcase that you go to lift turns out to be empty, you are unprepared for the sense of mismatch between the applied force and its effect. The inclusion of voluntary movement does reinforce the plausibility of the whole idea of the predictive mind. Movement is much easier to observe than perception, and plainly involves both prediction and feedback. Normal movement control implies that the brain has capacity for integral calculus involving distance, velocity and acceleration, and could easily be using similar mechanisms to integrate the Bayesian probability estimates of different incoming sensory streams.
Good ideas can create their own momentum. The stated aim of The Predictive Mind is to outline a unified explanation for many puzzling aspects of the mind. In Parts II and III, the scope of the book’s ambition becomes clearer. Two controversial topics sitting in the border zone between philosophy and cognitive neuroscience are “the binding problem” (how does the brain assign stable identities to objects from disparate sources of sensation?) and “cognitive penetrability” (of perception by mood, belief or desire). Hohwy brings a perceptual inference angle to the debates about both phenomena. The hallucinations and delusions of schizophrenia and the behavioural abnormalities of autism spectrum disorders are analysed in terms of failure of prediction error minimisation systems. Next, philosophy’s most elusive subjects—consciousness, attention, introspection and emotional states—come under the umbrella of the Bayesian predictive hypothesis. Prioritising an internal model of the world at any one time corresponds to the unitary character of conscious perception, and perhaps all consciousness occupies a virtual global workspace of the brain that has access to the predictive modelling. By this stage, some readers will notice the prediction error minimisation framework starting to strain under the load of all of these applications. Could this account explain the greatest mysteries of an organ as complicated and versatile as the human brain? Even validated theories and discoveries in neuroscience usually turn out to reveal less about the brain than initially promised. There is a problem with such a comprehensive principle of cognition being driven by the cerebral cortex without much regard for the importance of older, deeper nerve circuitry. While computers have evolved from adding machines, the human brain is really a vastly improved version of the tribolite brain, and it shares goals that enabled those animals to negotiate daily life in the Cambrian seas.
Had I the skills to design a computerised system to model human intelligence, I would not start with its sophisticated perceptual properties. I would first build a device that could examine any given set of circumstances for survival threat, for reproductive opportunity and for energy maximisation (the input from food against the energy expended in obtaining it). On this foundation, more elaborate layers of computer programming could be superimposed to interpret drives from below in increasingly subtle and abstract ways. Hohwy’s book gives the impression of a predictive mind that operates at a remove from its basic motivators. Hallucinations, which he attributes to excessive perceptual prediction errors, also reveal something of the “primitive” brain when caused by organic cerebral disease. If the cerebral visual processing centres are interrupted by damage from a stroke or infiltrated by the microscopic pathology of Parkinson’s disease, some individuals experience complex visual hallucinations. Their content is more likely to come from an older store of images of the natural world—insects, reptiles, small mammals—than from defective predictions about a modern visual environment. In severe visual hallucinosis, the animals are more frightening and faceless human intruders more menacing, seeming to imitate ancient fears of hostile environments that most of us have experienced only in dreams or on television.
Right at the end of the work comes its most interesting philosophical insight, which concerns our interactions with each another. The Law of the Jungle is so named because jungles are populated by solitary animals who must ruthlessly pursue their own biological priorities in order to prosper. The Law of the Plain, where the pack and herd animals live, has to work differently. Social animals have the same drives to maximise their individual survival chances, reproductive prospects and energy status but, in addition, their brains must identify situations where group interests override individual ones and modify behaviour accordingly. Humans have the most complex social interactions and, as with all social animals, these need to be sustained by some sort of meeting of minds. While individual biological priorities are deeply engrained, there is no reason why this meeting of minds should not occur at much higher levels.
Hohwy’s argument runs like this: the private nature of our conscious experiences is a key to their social function. Our brains are constantly using their Bayesian strategies to refine models of the outside world, but the results are shielded from others. This creates a conditional independence that allows the interaction of separate minds also to have a Bayesian character, improving and aligning beliefs about their world. Implicitly, our brains understand that other brains are using similar predictive processes, a point that Hohwy makes by referring to John Stuart Mill’s statement on privacy of mind and inference about others’ inner thoughts. Had the nineteenth century’s clear-headed champion of personal liberty been aware of this part of the predictive mind theory, he might have been able to avoid a mistake. Mill’s writings on political economy failed sufficiently to appreciate that both the Law of the Jungle and the Law of the Plain are important in determining economic behaviour, and that the pursuit of individual needs only achieves advantageous outcomes for a society ‑ the “invisible hand” of a market economy ‑ because none of its members are really acting as if their own interests are all that matters.
Peter Kempster lives in Melbourne where he works as a neurologist. Some literary interests, which relate to his professional learning, include the use of narrative techniques as an aid to neurological diagnosis and a neurobiographical understanding of the life and work of John Ruskin.
