Many philosophers and scientists still believe that sentience serves no physical purpose. My task in this book is to persuade you of the plausibility of an alternative interpretation. This requires me to convince you that feelings are part of nature, that they are not fundamentally different from other natural phenomena, and that they do something within the causal matrix of things. Consciousness, I will demonstrate, is about feeling, and feeling, in turn, is about how well or badly you are doing in life. Consciousness exists to help you do better.
Consciousness should not be confused with intelligence. It is perfectly possible to feel pain without any reflection as to what the pain is about. Likewise, the urge to eat – a feeling of hunger – need not imply any intellectual comprehension of the exigencies of life. Consciousness in its elemental form, namely raw feeling, is a surprisingly simple function
was drawn to Panksepp, Damasio and Merker because they believed, as I do, that what is lacking in the neuroscience of our time is a clear focus on the embodied nature of lived experience. It
The nature of consciousness may be the most difficult topic in science. It matters because you are your consciousness, but it is controversial because of two puzzles that have bedevilled thinkers for centuries. The first is the question of how the mind relates to the body – or, for those of a materialist bent (which is almost all neuroscientists), how the brain gives rise to the mind. This is called ‘the mind/body problem’. How does the physical brain produce your phenomenal experience? Equally confoundingly, how does the non-physical stuff called consciousness control the physical body?
The question of what we can know about minds from the outside – how we can even tell when they are present, for that matter – is the second puzzle. It is called ‘the problem of other minds’. Simply put: if minds are subjective, then you can only observe your own. How, then, can we know whether other people (or creatures, or machines) have one at all, let alone discern any objective laws governing how minds in general work?
Starting in the first half of the twentieth century, a school of psychology called ‘behaviourism’ began systematically to apply the experimental method to the mind. Its starting point was to disregard everything except empirically observable events. The behaviourists threw out all ‘mentalistic’ talk of beliefs and ideas, feelings and desires, and restricted their field of study to the subject’s visible and tangible responses to objective stimuli. They were fanatically uninterested in subjective reports about what was going on inside. They treated the mind as a ‘black box’, whose inputs and outputs were all that could be known of it. Why did they take such an extreme stance? Partly, of course, it was an attempt to navigate around the problem of other minds. If they refused to countenance any talk of minds in the first place, it stood to reason that their theories could not be afflicted by the philosophical doubts endemic to psychology. In effect, they excluded the psyche from psychology.
Despite the austerity of their programme, they were in fact able to infer causal relations between certain types of mental stimuli and responses. Not only that: they could also manipulate the inputs to elicit predictable changes in the outputs. In doing so, they discovered some of the fundamental laws of learning
Such discoveries were no small achievement; they showed that the mind is subject to natural laws, like everything else
The great neurologist Jean-Martin Charcot once said: ‘theory is good, but it doesn’t prevent things from existing’.2 Since internal mental events clearly do exist and causally influence behaviour, the behaviourist approach was gradually eclipsed in the second half of the twentieth century by another approach. It was called ‘cognitive’ psychology, which was able to accommodate
internal mental processes – in a manner of speaking.
What is information processing? I will say a lot about it later, but the most interesting thing for our present purposes is that it can be implemented with vastly different kinds of physical equipment. This casts new light on the physical nature of the mind. It suggests that the mind (construed as information processing) is a function rather than a structure
This is the power of what came to be called the ‘functionalist’ approach, but it is also its weakness. If the same functions can be performed by computers, which presumably are not sentient beings, then are we really justified in reducing the mind to mere information processing? Even your phone has memory, perceptual and executive functions.
The third major scientific response to mind/body metaphysics developed in tandem with cognitive psychology, but by the end of the last century it had grown to overshadow it. I am referring to an approach that is broadly termed ‘cognitive neuroscience’. It focuses on the hardware of the mind, and it arose with the development of a plethora of physiological techniques that make it possible for us to observe and measure the dynamics of the living brain directly.
These techniques render the inner workings of the organ of the mind plainly visible – thereby realising the wildest empiricist dreams of the behaviourists without limiting the scope of psychology to stimuli and responses.
Neuropsychology is admirable, but it excludes the psyche – it excludes the experiencing, active, living ‘I’
These books were quite unlike my neuropsychological textbooks, which dissected mental functions as we would the functions of any bodily organ
Was the brain really no different from the stomach and lungs? The obvious thing that set it apart was the fact that there is ‘something it is like’ to be a brain. This did not apply to any other part of the body
Those who did take the subjective perspective into account were not taken seriously by proper neuroscientists. I am not sure how many people know that Sacks’s publications were widely derided by his colleagues. One commentator went so far as to call him ‘The man who mistook his patients for a literary career’
This caused him a good deal of distress
You are warned not to pursue such intractable questions – they are ‘bad for your career’. And so, most students of neuroscience gradually forget why they entered the field, and come to identify with the dogma of cognitivism, which approaches the brain as though it were no different from a mobile phone
The one aspect of consciousness that was a respectable scientific topic in the 1980s was the brain mechanism of wakefulness versus sleep. In other words, the ‘level’ of consciousness was a respectable topic but not its ‘contents’. So, I decided to focus my doctoral research on an aspect of sleep. In particular, I chose to study the subjective aspect of sleep, namely the brain mechanisms of dreaming. Dreaming, after all, is nothing but a paradoxical intrusion of consciousness (‘wakefulness’) into sleep. Amazingly, there was a huge gap in the literature on this topic: nobody had systematically described how damage to different parts of the brain affected dreaming. So, this is what I set out to do.
When the brain descends into what is now called ‘slow wave’ sleep, we see exactly what Aserinsky and Kleitman predicted. Their hypothesis was confirmed. The surprise is what happens next: within about ninety minutes of drifting off (and roughly every ninety minutes thereafter, in regular cycles) the brainwaves speed up again, almost reaching waking levels, even though the person from whom the recordings are being obtained remains asleep
Various other things happen in this peculiar state. The eyes move rapidly (which is why paradoxical sleep was later renamed ‘rapid eye movement’ or REM sleep), yet the body below the neck is temporarily paralysed. There are dramatic autonomic changes, too, such as reduced control
of core body temperature and engorgement of the genitals leading to visible erections in men. How science managed not to notice all this until 1953 is mind-boggling.
it: whereas approximately 80 per cent of awakenings from REM sleep produced dream reports, fewer than 10 per cent of awakenings from non-REM sleep did so. From that moment onward, REM sleep was considered to be synonymous with dreaming
Unlike the mainstream responses to mind/body metaphysics that characterised mental science in the second half of the twentieth century, psychoanalysts had no qualms about treating introspective reports as data. In fact, reports elicited by ‘free association’ (unstructured sampling of the stream of consciousness) were the primary data of psychoanalytic research
The discovery of REM sleep in the 1950s triggered a race to identify its neurological basis, since the function of REM sleep could reveal the objective mechanism of dreams, whose elucidation would place the psychiatry of the time on a more respectable scientific footing. (This research was made easier by virtue of the fact that REM sleep occurs in all mammals.) The race was won by Michel Jouvet, in 1965. In a series of surgical experiments on cats, he demonstrated that REM sleep was generated not by the forebrain (which includes the cortex, the upper part of the brain that is so impressively large in humans and partly for that reason is considered the organ of the mind) but rather by the brainstem, a supposedly much humbler structure of exceedingly ancient
evolutionary origin.11
Combining these findings with the fact that REM sleep switches on and off automatically, roughly every ninety minutes, like clockwork, Hobson wasted no time in drawing the inevitable conclusion: ‘The primary motivating force for dreaming is not psychological but physiological since the time of occurrence and duration of dreaming sleep are quite constant, suggesting a preprogrammed, neurally determined genesis
Therefore, according to Hobson, Freud’s view that dreams were driven by latent desires
must be completely wrong. The meaning that Freud saw in dreams was no more intrinsic to them than it is to inkblots. It was projected onto them; it was not in the dream itself. From the scientific point of view, dream interpretation was no better than reading tea leaves.
To my amazement, however, alongside all the obvious things I observed, I found also that patients with damage to the part of the brain that generates REM sleep still experienced dreams. Moreover, patients in whom dreaming was abolished had damage to a completely different part of the brain. Dreaming and REM sleep were therefore what we call ‘doubly dissociable’ phenomena.18 They were correlated with each other (i.e. they usually happened at the same time) but they were not the same thing
When, in the early 1990s, I first reported that dreaming was obliterated by damage in a different part of the brain from the part that generates REM sleep, I took pains to stress that the critical area was not in the brainstem.20 This was because I wanted to emphasise the mental nature of dreaming, and we all knew that mental functions reside in the cortex.
In fact, I found two areas of damage that caused loss of dreaming with preservation of REM sleep. The first was in the cortex, in the inferior parietal lobule (see Figure 2). That finding was not surprising, as the parietal lobe is important for short-term memory. If a patient cannot hold the contents of their memory in the buffer of consciousness, how can they experience a dream? Far more interesting was the second brain area, namely the white matter of the ventromesial quadrant of the frontal lobes, which connects the frontal cortex to various subcortical structures. This finding was totally unexpected; nothing about the functions of this part of the brain is obviously connected with the manifest experience of dreaming, and yet it must contribute something crucial to the process, because damage there reliably caused a total cessation of dreaming
What they found is that prefrontal leucotomy had three main psychological effects. First, it reduced positive psychotic symptoms (hallucinations and delusions). Second, it reduced motivation. Third, it caused loss of dreaming. In fact, one of the early psychosurgical investigators went so far as to suggest that preservation of dreaming after the operation was a bad prognostic sign
What are dreams if not hallucinations and delusions?
As it happens, the neurosurgical treatment of hallucinations and delusions was not abandoned for ethical reasons; it fell out of favour when it became apparent that equivalent therapeutic results could be obtained with less morbidity and mortality by using some drugs which first became widely available in the 1950s, namely ‘major tranquillisers’. What these drugs did, and modern ‘antipsychotics’ still do, was block the neurochemical dopamine at the terminals of a brain circuit
known as the mesocortical-mesolimbic dopamine system
It is therefore now widely accepted that dreaming can occur independently of REM sleep and that the mesocortical-mesolimbic dopamine circuit is indeed the major driver of dreaming
It rapidly became clear that neuroscience owed Freud an apology. If there is one part of the brain that might be considered responsible for ‘wishes’, it is the mesocortical-mesolimbic dopamine circuit. It is anything but motivationally neutral. Edmund Rolls (and many others) calls this circuit the brain’s ‘reward’ system.29 Kent Berridge calls it the ‘wanting’ system. Jaak Panksepp calls it the SEEKING system – and foregrounds its role in the function of foraging.30 This is the brain circuit responsible for ‘the most energised exploratory and search behaviours an animal is capable of exhibiting’.31 It is also the circuit that drives dreaming
From 1999 onwards, partly prompted by Braun’s comments about the implications of my discovery, I directed my attention to the other arousal systems of the brainstem. The most interesting work in this area was being done by Jaak Panksepp, whose encyclopaedic book Affective Neuroscience (1998) laid out in exquisite detail a vast array of evidence for his view that these supposedly mindless systems, responsible for regulating only the ‘level’ of consciousness, generated a ‘content’ of their own.
In the 1990s, in common with the rest of neuropsychology, I thought the cortex was where all the psychological action was, so I focused on the fact that the white matter tracts that interested me were in the frontal lobes, which is where the damage in my nine cases was located. But all the core nuclei of the brainstem send long axons upwards into the forebrain (see Figure 2). The cell bodies of these neurons are located in the brainstem, although their output fibres (the axons) terminate in the cortex. This underpins the main arousal function of these brainstem nuclei, known collectively as the reticular activating system. It was these activating pathways that were damaged in my nine patients, and in the hundreds of documented nondreaming leucotomy patients who preceded them
He concluded that what ultimately underpinned feelings were bodily needs; that human mental life, no less than that of animals, was driven by the biological imperatives to survive and reproduce. These imperatives, for Freud, provided the link between the feeling mind and the physical body.
Although his quest from the first had been to discern the laws underpinning our rich inner life of
subjective experience, nevertheless mental life remained a biological problem for him
These were what he called ‘metapsychological’ forces, the forces that lie behind psychological phenomena. He clarified that he wanted to ‘transform metaphysics into metapsychology’.18 In other words, he wanted to replace philosophy with science – a science of subjectivity
Foremost among the new forces that Freud was obliged to infer was the concept of ‘drive’, which he defined as ‘the psychical representative of the stimuli originating from within the organism and reaching the mind, as a measure of the demand made upon the mind for work in consequence of its connection with the body’
Freud’s notion of ‘drive’ – which he considered to be the source of all ‘psychical energy’ – was not unlike Müller’s ‘vital energy’, but it was rooted in bodily needs. Freud described the causal mechanisms by which drives become intentional cognition as an ‘economics of nerve-force’.21 Still, he freely admitted that he was ‘totally unable to form a conception’ of how bodily needs become a mental energy
The subjective ‘I’ was never excluded from psychoanalysis, for all its faults. It had pride of place, no matter how embarrassing that was for the rest of science. Many scientific colleagues advised me not to associate what I was doing with psychoanalysis, given the historical baggage the word carried. They said it was like an astronomer associating himself with astrology. But I considered it intellectually dishonest to not give Freud his due. No matter how incompletely he achieved his goals, they were clearly the correct goals for a science of the mind. So, I called my approach ‘neuropsychoanalysis’. I have said that the neuropsychology I was taught might as well have been called neurobehaviourism, such was its attitude to subjectivity. I wanted to be clear that the neuropsychology I was developing pivoted on lived experience. In that spirit, after writing a programmatic paper on the relationship between psychoanalysis and neuroscience, I set to work
But when Mr S’s confabulations are considered from the subjective point of view, additional facts emerge. Imagine what it feels like to suddenly realise that you do not recognise the clinician who just walked into the room, although he seems to be responsible for your care; that you do not know what room (or even which city) you are in; that you have a huge scar over the top of your head, and you do not know where it comes from; that – in fact – you do not remember what happened just two minutes ago, let alone over the days and months preceding the present moment. You would probably feel something like panic, wondering whether this doctor might have performed an operation on your head, as a result of which you no longer remember anything from one moment to the next. This is what missing memory search and monitoring mechanisms feel like to the intentional subject of the mind – to the living I.
Considering Mr S’s confabulations from the first-person perspective clearly reveals something new about them: the content of his misrememberings is tendentiously motivated. These are far from being random search errors. They contain a clear self-serving bias; they have the aim and purpose of recasting his anxiety-ridden situation into a reassuring, safe and familiar one. So, just as Freud inferred in the case of dreams, confabulations are motivated. The mental processes in confabulatory amnesia are wishful. But this fact becomes apparent only when the emotional context and personal meaning (experienced by Mr S alone) of dental implants (‘the implants work fine’) and cardiac pacemakers (‘it never misses a beat’) are taken into account – as a psychoanalyst would do. This is what neuropsychologists fail to see when they aim to be entirely objective; as Sacks put it, when they exclude the psyche
These were my first steps. Naturally, such broad generalisations cannot be based on purely clinical evidence in a single case. Having formulated my impression of Mr S, therefore, I enlisted ‘blind’ raters (colleagues unfamiliar with my hypothesis) to measure, on a seven-point Likert scale, the degree of pleasantness versus unpleasantness in a continuous unselected sample of 155 of his confabulations. The results were statistically (highly) significant: when compared to the target memories they replaced, Mr S’s confabulations substantially improved his situation from the emotional point of view.31 Next, my research collaborators and I demonstrated the same strong effect in studies involving numerous other patients with confabulations. In subsequent empirical studies, the mood-regulating effects of confabulation that I inferred clinically in the case of Mr S were statistically validated.32 This programme of research opened a whole new approach to the neuropsychology of confabulation,33 and related disorders such as anosognosia.34 It also laid the foundations for a novel approach to common psychiatric disorders such as addiction and major depression.35 I have spent the last three decades developing this ‘neuropsychoanalytic’ approach to mental illness, trying to return subjectivity to neuroscience.36
This is the part of the brain that obeys the ‘pleasure principle’. But how can feelings of pleasure be unconscious? As we saw with Damasio’s patient, drives such as hunger and thirst and the desire to void are felt. Of course they are. Yet Freud said the id – the seat of the drives – was unconscious. He had imbibed the same classical doctrine as Craig (as had I, at least initially) and had therefore located consciousness in the cerebral cortex
For Freud, clearly, conscious feelings, no less than perceptions, are generated in the ‘ego’ (the part of the mind that he identified with the cortex),45 not in the unconscious ‘id’ – which I was now obliged to locate in the brainstem and hypothalamus. In short, it seemed that Freud got the functional relationship between the ‘id’ (brainstem) and the ‘ego’ (cortex) the wrong way round, at least insofar as feelings are concerned. He thought the perceiving ego was conscious and the feeling id was unconscious. Could he have got his model of the mind upside down?46
Merker’s most crucial observation is not that the patients lose and regain alertness but that they show ‘responsiveness to their surroundings in the form of emotional or orienting reactions to environmental events’. This is the defining feature of what vegetative patients are supposed to lack: intentionality
It is generally accepted that these patients register visual, auditory and tactile stimuli unconsciously, but Merker saw the children expressing pleasure by smiling and laughter, and aversion by fussing, arching their backs and crying, ‘their faces being animated by these emotional states
without a cortex. This is usually due to a massive stroke in utero, which results in reabsorption of the forebrain, so that the baby’s cranium is filled with cerebrospinal fluid instead of brain tissue. Hence the term ‘hydranencephaly’ – which means ‘water instead of encephalon’
Where in the brain is consciousness generated? For the past 150 years, the almost universal answer to this question has been ‘in the cortex’. This was the only point on which Freud and the mainstream tradition in twentieth-century mental science could agree. And yet, if that’s right, when the cortex is absent, consciousness ought to disappear. In the case of hydranencephalic children this appears not to happen. All the behavioural evidence suggests that they are, in fact, conscious. They are not comatose and nor do they exist in a vegetative state.
What happens when this is done? Obviously, such experimental procedures cannot be performed in human children, but they have often been performed in other newborn mammals, such as dogs, cats and rats. The outcome is always the same: by the objective behavioural criteria that we normally use to measure it, consciousness is preserved.* The post-operative behaviour of these animals cannot by any stretch of the definitions be described as ‘comatose’ or ‘vegetative’. Merker writes that they show ‘no gross abnormalities in behaviour that would allow a casual observer to identify them as impaired’. Antonio Damasio concurs: ‘Decorticated mammals exhibit a remarkable persistence of coherent, goal-oriented behaviour that is consistent with feelings and consciousness.’3 Neonatally decorticate rats, for example, stand, rear, climb, hang from bars and sleep with normal postures. They groom, play, swim, eat and defend themselves. Either sex is capable of mating successfully when paired with normal cage mates. When they grow up, the females show the essentials of maternal behaviour, which, though deficient in some respects, allow them to raise pups to maturity
Take the famous phenomenon of ‘blindsight’, which occurs when the visual cortex is destroyed.6 These patients are able to respond to visual stimuli, and yet when asked to describe what their ‘seeing’ is like, they report that they do not experience any visual images at all; instead, if asked, they use gut feelings to guess where the visual stimuli are located, which they do with remarkable accuracy. Consider the case of the patient known as TN as reported by the neuroscientist Lawrence Weiskrantz: although totally blind – in other words, entirely devoid of conscious visual experience – TN deftly manoeuvred around obstacles placed in his way along a corridor. Questioned afterwards, he reported having no idea that he was avoiding anything
The little girl pictured in Figure 5 possesses no functional cortex at all. If the argument above is right, it must be that she unconsciously senses her baby brother being placed on her lap, without generating any conscious awareness of the situation. In fact, she must be incapable of any sort of conscious experience whatsoever. She is something like what the philosopher David Chalmers calls a ‘zombie’. Though in certain respects she acts normally, all is dark within
being conscious in the behavioural sense of being awake and responsive is significantly different from having consciousness in the phenomenological sense – that is, being a subject of experience
if we are to accept that someone who seems to be conscious actually isn’t, we should require an extremely convincing argument. Merely raising philosophical doubt isn’t enough. We need very good grounds to think that the two sorts of consciousness have come apart in such people, as they seemingly never do in us
The essence of homeostasis is that living organisms must occupy a limited range of physical states: their viable states, or valued or preferred states, or what Friston calls (referring to all of the above) their ‘expected’ states
Homeostasis runs in the opposite direction. It resists entropy. It ensures that you occupy a limited range of states
Living things must resist one of the fundamental principles of physics: the Second Law of Thermodynamics
In thermodynamics, there are two conditions of energy: useful and useless. The ‘usefulness’ of energy is defined by its capacity to perform work
Combining these facts, we learn the following: as the useful energy in a system runs down, its entropy increases. This means that the capacity of the system to perform work always decreases. Entropy is therefore associated with loss of useful energy, because the energy is no longer available to perform work. The Second Law is a statement of the ineluctable fact that some energy will be lost to useful work during any natural process
Think of entropy in terms of the number of locations that each molecule might be in at a given time. This turns out to be a statement about probability
This is important, because, unlike the other laws of thermodynamics, the laws of probability apply to all things, not just material things. Just as the entropy of a gas in a chamber can be defined probabilistically, so too can the entropy associated with a psychological decision-making process. In both cases the entropy increases with the randomness of possible outcomes. The ‘entropy’ associated with expanding gases and expanding options is the same thing
In my physical example of entropy, when a gas was initially compressed into its container and its molecules were packed close together, fewer bits of information were required to describe the actual location of each molecule than were required after it was released to fill all the available space in the large chamber
to put it more simply still: the more yes/no questions one needs to answer to describe a system, the greater its entropy
Entropy measures the average amount of information you get upon multiple measurements of a system. Thus, the entropy of a series of measurements is its average information, its average uncertainty.
EEG entropy values are higher in minimally conscious than in vegetative patients
The relationship between entropy and information was formalised in a famous equation by the electrical engineer and mathematician Claude Shannon. With this breakthrough, Shannon single-handedly incorporated ‘information’ into physics, where it has since become a basic concept, especially in quantum mechanics
The laws of thermodynamics could therefore be seen as a special case of the deeper laws of probability
This omission of the experiencing subject raises a bigger question than that, however – perhaps the biggest one faced by cognitive science today: without an observer, how and why does information processing (i.e. question asking and answering) come about in the first place
Experience itself therefore arises from communication between an information receiver (a participant observer) and an information source; between a questioner and the answers they register. But this still leaves the question: where do questioners come from?
I have conveyed three important points. The first is that the average information of a system is the entropy of that system (i.e. the entropy in a system is a measure of the amount of information needed to describe its physical state). The second is that living systems must resist entropy. These two facts together imply that we must minimise the information that we process. (Here I mean information in Shannon’s sense, of course; in other words, we must minimise our uncertainty.) Everything else I am going to say in this chapter, and the next two, follows from this simple but startling conclusion
This leads to the third important thing we have learnt so far: we living systems resist entropy through the mechanism of homeostasis. In short, we receive information about our likely survival by asking questions (i.e. taking measurements) of our biological state in relation to unfolding events. The more uncertain the answers are (i.e. the more information they contain) the worse for us; it means we are failing in our homeostatic obligation to occupy limited states (our expected states).
With Darwin’s insight, the question of the origin and composition of teleological beings became tractable to science
An important further step came in the mid-twentieth century, when Norbert Wiener, the mathematician who founded the discipline of ‘cybernetics’, added the notion of feedback to Shannon’s understanding of information. According to Wiener, a system could attain its goal (its ‘reference state’) by receiving feedback about the consequences of its actions. The feedback includes error signals – measuring deviations from the reference state – which would be used to adjust the system’s actions, and keep it on course. Homeostasis thus turns out to be a specific case of a more general cybernetic principle: it is a kind of negative feedback.
William Ross Ashby used this notion of feedback, combined with the statistical physics introduced above, to reveal how self-organisation develops naturally.35 Ashby showed that many complex dynamical systems automatically evolve towards a settling point, which he described as an ‘attractor’ in a ‘basin’ of surrounding states. The further evolution of such systems then tends to occupy limited states
Hopefully, this tendency to occupy limited states sounds familiar to you: it is nothing other than a tendency to resist entropy. According to Friston, it is this tendency that triggers ever more elaborate forms of self-organisation
This arrangement of causal dependencies defines the properties of what is known as a Markov blanket
The ‘Markov blanket’ is a statistical concept which separates two sets of states from each other. Such formations induce a partitioning of states into internal and external ones, i.e. into a system and a not-system, in such a way that the internal states are insulated from the ones that are external to the system
The formation of a Markov blanket thus divides a system’s states into four types – internal, active, sensory and external – where the external states are not part of the self-organising entity. Crucially, the dependencies between these four types of state create a circular causality. The external states influence the internal ones via the sensory states of the blanket, while the internal states couple back to the external ones through its active states. In this way, the internal and external states cause each other in a circular fashion. Put differently, sensory states feed back the consequences of the effect on the external states of the active states, and thereby adjust the subsequent actions of the system
Self-organising systems can always be composed of smaller self-organising systems – not all the way down, but certainly a dizzyingly long way. That is the basic fabric of life: billions of little homeostats wrapped in their Markov blankets
from what I have told you so far it is reasonable to conclude that the very selfhood of a complex dynamical system is constituted by its blanket. Such self-organising systems come into being by separating themselves from everything else. Thereafter, they can only register their own states: the not-system world can only be ‘known’ vicariously, via the sensory states of the system’s blanket. I propose that these properties of self-organisation are in fact the essential preconditions for subjectivity.
It is the self-preservative nature of such systems, their tendency to separate from their environments and then actively maintain their own existence, that provides the elemental basis of selfhood. And it is the insulated nature of such systems, the fact that they can only register the not-self world via the sensory states of their own blankets, that constitutes the elemental basis of subjectivity – the ‘point of view’ of a sequestered self
In thermodynamics, the free energy of a system is equal to the total amount of energy contained in the system minus the part of that energy that is already employed in effective work and therefore is not free
Free energy is what is left when you take away the energy that isn’t free (i.e. when you take away the ‘bound energy’)
Friston uses a third version of the same equation, to quantify free energy in information contexts. I will call this type of free energy ‘Friston free energy’. The relevant equation says that ‘Friston free energy is equal to average energy minus entropy’.46 Here, ‘average energy’ means the expected probability of an event happening under a model, and ‘entropy’ means the actual incidence of it happening. So Friston free energy is the difference between the amount of information you expect to obtain from a data sample – from a sequence of events – and the amount of information you actually obtain from it
surprisal, like model entropy, is a bad thing for living organisms
at the most brutally basic level, you are in a surprising state if you move outside the set of states you biologically expect to be in (e.g. below or above 36.5–37.5 ˚C, or breathing underwater), precisely because there is a low probability of your being in that state
Friston free energy is a quantifiable measure of the difference between the way the world is modelled by a system and the way the world really behaves. Therefore, we must minimise this difference
which means that it must minimise the difference between the sensory data that it samples from the world and the sensory data that were predicted by its model
One way to do this is by improving the system’s model of the world. Prediction errors can be fed back to the generative model, so that it generates better predictions next time
quantifying the gap between the sensory states predicted by an action and the sensory states that actually flow from that action
Sensory evidence (received in the form of spike trains generated inside our heads – in effect, billions of ones and zeros) are the only data we can get. From that data we must infer the causal structure of the world. Markov-blanketed beings that we are, we are obliged to rely on things like probability distributions rather than absolute truths. That is why it is useful to know that Friston free energy is always greater than surprisal; it enables our brains to approximate unknowable truths using statistical calculations
As we saw in Friston’s soup experiment, generative models come into being with self-organising systems. For that reason, they are sometimes called ‘self-evidencing’ systems, because they model the world in relation to their own viability and then seek evidence for their models. It is as if they say not ‘I think, therefore I am’ but ‘I am, therefore my model is viable’
The Free
Energy Principle thus explains in mathematical terms how living systems like you and I resist the Second Law of Thermodynamics through homeostasis-maintaining work
The most important thing about Bayes’s theorem for the purposes of neuroscience is that it explains how perceptual inference – an unconscious process – actually works in real life, and how signal transmission actually works in real sensory-motor processing. Brain circuits literally do compute prior probability distributions and then send predictive messages to sensory neurons, in an endless effort to dampen the incoming signals; and perception literally does involve comparisons between the predicted and actual distributions, resulting in computations of posterior probability. The resultant inferences are what perception actually is.59 Perception is an endeavour to self-generate the incoming sensory signals and thereby explain them away. That is why so many neuroscientists nowadays speak of the ‘Bayesian brain’.
minimising prediction error minimises information flow, and reducing information flow reduces metabolic expenditure on the part of the brain and the body as a whole
The predictive brain is thus revealed to be ‘lazy’ (over the long term): vigilant for every opportunity to achieve more by doing less
We have learnt that the suppression of prediction error is the essential mechanism of homeostasis. Therefore, minimising free energy becomes the basic task of all homeostatic systems. Friston’s free-energy equation turns out to be a reformulation, in quantifiable terms, of Freud’s definition of ‘drive’: ‘a measure of the demand made upon the mind for work in consequence of its connection with the body’ – a measure that Freud considered to be unquantifiable. Now we can quantify it. The fundamental driving force behind the volitional behaviour of all life forms is that they are obliged to minimise their own free energy. This principle governs everything they do
It has become evident, thanks to Friston’s work, that the nervous system implements an iterative predictive hierarchy that works very much like the one that Eve Periaqueduct established over time