The Hidden Spring

1. The Stuff of Dreams

• Two puzzles that have bedevilled thinkers for centuries:
  • The mind/body problem - How the mind relates to the body, or how the brain gives rise to the mind.
  • The problem of other minds - What can we tell about what happens in other people's minds.
• In the first half of the 20th C, behaviorism worked with classical conditioning and operant conditioning (the Law of Effect)
• In the second half, behavorism was gradually eclipsed by "cognitive" psychology, which formulates models of the information processing that goes on within minds. It suggests that the mind is a function rather than a structure. The software of the mind is implemented by the hardware of the brain and could be implemented elsewhere. Both brains and computers perform:
  • Memory functions (they encode and store information)
  • Perceptual functions (they classify patterns of incoming information by comparing them with stored information)
  • Executive functions (they execute decisions about what to do in response to such information)
• In parallel with cognitive psychology developed "cognitive neuroscience", which focuses on the hardware of the mind.
• Paradoxical sleep, where the brain is physiologically aroused despite being fast asleep.
• The whole sleep/waking cycle - including REM sleep and dreams as well as the different stages of non-REM sleep - is orchestrated by a small number of brainstem nuclei interacting with each other.

2. Before and After Freud

• Freud argued that the erratic train of our conscious thoughts can be explained only if we assume implicit intervening links of which we are unaware. This gave rise to the notion of latent mental functions and, in turn, to Freud's famous conjecture of "unconscious" intentionality.
• He concluded that what ultimately underpinned feelings were bodily needs; that human mental life, no less that 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.
• Freud wanted to find metapsychological functional laws. Trying to skip over the functional level of analysis, jumping directly from psychology to physiology, is nowadays called the localisationist fallacy.
• Freud's "project" spoke of the forebrain as a sympathetic ganglion monitoring and regulating the needs of the body and of these needs as the driving force of mental life, "the mainspring of the psychical mechanism":
  • Drive - 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. He described the causal mechanisms by which drives become intentional cognition as an "economics of nerve-force".
  • "The Id, cut off from the external world, has a world of perception of its own. It detects with extraordinary acuteness certain changes in its interior, especially oscillations in the tension of its drive needs, and these changes become conscious as feelings in the pleasure-unpleasure series. It is hard to say, to be sure, by what means and with the help of what sensory terminal organs these perceptions come about. But it is an established fact that self-perceptions - coenaesthetic feelings and feelings of pleasure-unpleasure - govern the passage of events in the id with despotic force. The id obeys the inexorable pleasure principle."
  • "What consciousness yields consists essentially of perceptions of excitations coming from the external world and of feelings of pleasure and unpleasure which can only arise from within the mental apparatus; it is therefore possible to assign to the system Pcpt-Consciousness a position in space. It must lie on the borderline between inside and outside; it must be turned towards the external world and must envelop the other psychical systems.
• 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), not in the unconscious "id" - which I was now obliged to locate in the brainstem and hypothalmus. 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?

3. The Cortical Fallacy

• Hydranencephalic children are born without a cortex.
• Decorticated mammals exhibit a remarkable persistence of coherent, goal-oriented behavior that is consistent with feelings and consciousness. In many respects, decorticate mammals are in fact more active, emotional and responsive than normal ones.
• Information from the sense organs is fed not only to the cortex but also to the superior colliculi of the brainstem, via a set of subcortical connections.
• Two principal meanings of the term consciousness:
  • Consciousness as the waking state - A necessary condition for consciousness in the second sense
  • Consciousness as experience
• The way in which ideas become conscious in response to an external stimulus was called apperception (which roughly means perceiving the present through the lens of past experience).
• Meynert in 1884 identified the mind with the totality of memory images of objects produced by projection of the sensory-motor periphery onto the cortex plus transcortical associations between them and those memory images that constituted abstract ideas.
• Penfield and Jasper concluded that cortical resections do not interrupt sentient being; they merely deprive patients of certain forms of information.

4. What is Experienced?

• The scientific evidence showing that we are unaware of most of what we perceive and learn is now overwhelming. Perception and memory are not inherently conscious brain functions.
• If the cortex is not connected directly with the body periphery, but rather via intermediate subcortical links, then the memory images deposited in the cortex cannot be literal projections of the world outside. They must be the end products of multi-stage information processing.
• It makes no sense, Freud argued, to draw an artificial line between the subcortical and cortical parts of the processing and claim that only the final product is "mental".
• Freud concluded that both conscious and unconscious memory images are formed in sympathy with the demands of the body - that we only perceptually represent and learn about the outside world because we must meet our biological needs there.
• Patients and research participants are conscious of their feelings; they are unconscious only of where the feelings came from. Apparently alone among mental functions, feeling is necessarily conscious.
• What turns up in consciousness? "Representations" of the outside world and feelings:
  • About what is going on in the world, about our thoughts about that world, about ourselves, including feelings that seem to be reports on the conditions of our bodies.
  • Free-floating feelings, the emotions and moods that qualify our experience of the world and shape our behavior within it. Sometimes they register as bodily sensations; still, many moods seem attributable neither to the condition of our bodies not to anything we can put our fingers on in the world outside. Isn't consciousness full of feelings like this? And yet, to an amazing degree, neuroscientists searching for an explanation of consciousness have ignored them.
• Feeding behavior is regulated by two interacting brain mechanisms: a "homeostatic" system, which regulates energy stores, and a "hedonic" system, which mediates appetite. And just as with bodily affects like hunger, might not the prohibition of emotional words like "sadness" and "fear" delay the development of antidepressant and anti-anxiety treatments?
• Feelings are real, and we know about them because they permeate our consciousness. They are, in fact, for the reasons I will now explain, the wellspring of sentient being - in a sense that seems to me barely metaphorical. From their origin in some of the most ancient strata of the brain, they irrigate the dead soil of unconscious representations and bring them to mental life

5. Feelings

• Isn't this how our feelings seem to motivate us, somewhere below the threshold of our awareness?
• Vasovagal syncope causes you to faint because your brain reacts to something alarming, usually the sight of blood or some other perceived risk of physical injury. This trigger (registered by the amygdala) activates the solitary nucleus in your brainstem, which causes your heart rate and blood pressure to drop suddenly. That in turn leads to reduced perfusion of your brain; and you lose consciousness.
• Why do we have this innate reflex? It reduces blood flow and thereby staunches haemorrhaging, in anticipation of injury. It is only in us humans that the reflex causes fainting, due to our upright posture and large brains, which requires more cardiac effort.
• Respiratory control is normally automatic: so long as the levels of oxygen and carbon dioxide in your blood stay within viable bounds, you don't have to be aware of your breathing in order to breathe. When blood gases exceed these normal limits, however, respiratory control intrudes upon consciousness in the form of an acute feeling called "air hunger". Unexpected blood gas values are an indication that action is required. It is urgently necessary to remove an airway obstruction or to get out of a carbon-dioxide-filled room. At this point, respiratory control enters your consciousness, via an inner warning system that we experience as alarm.
• The simplest forms of feeling - hunger, thirst, sleepiness, muscle fatigue, nausea, coldness, urinary urgency, the need to defecate, and the like - might not seem like affects, but that is what they are. What distinguishes affective states from other mental states is that they are hedonically valenced: they feel good or bad. This is how affective sensations such as hunger and thirst differ from sensory ones like vision and hearing. Sight and sound do not possess intrinsic value - but feeling does.
• Pleasure and unpleasure tell you how you are doing in relation to your biological needs. Valence reflects the value system underwriting all biological life, namely that it is good to survive and to reproduce and bad not to do so.
  • Hunger feels bad, and it feels good to relieve it by eating
  • A distended bowel feels bad, and it feels good to relieve it be defecating
  • Pain feels bad, and it feels good to withdraw from the source of it
  • Separation distress feels bad and we escape it by seeking reunion
  • Fear feels bad and we escape it by fleeing the danger (and sometimes by fainting
• Feelings make creatures like us do something necessary. In that sense, they are measures of demands for work.
• In the jargon of control theory, blood gas imbalances, temperature undershoots, missing caregivers and approaching predators are "error signals," and the actions they give rise to are meant to correct the errors.
• Affects are how we become aware of our drives; they tell us how well or badly things are going in relation to the specific needs they measure.
• If you swapped subjective redness with blueness there would be no consequences, but if you swapped the feeling of fear with separation distress (or hunger with urinary urgency), it would kill you.
• It makes a difference whether a need is felt or not. Your water-to-salt ratio may be sliding all the time, in the background, but when you feel it, you want to drink. You might objectively be in danger without noticing it, but when you feel it you look for ways to escape.
• Needs are different from affects. Bodily needs can be registered and regulated autonomically, as in the examples of cardiovascular and respiratory control, thermoregulation and glucose metabolism. These are called vegetative functions: there is nothing conscious about them. Hence the term autonomic reflex. Consciousness enters the equation only when needs are felt. This is when they make demands on you for work.
• Emotional needs, too, can be managed automatically, by means of behavioral stereotypes such as "instincts" (inborn survival and reproductive strategies, which Freud placed at the center of his conception of the unconscious mind). But emotional needs are usually more difficult to satisfy than bodily ones. That is why the feelings they evoke are typically more sustained. A feeling disappears from consciousness when the need it announces has been met.
• Felt needs are prioritized over unfelt ones. Priorities are determined by the relative strengths of your needs (the size of the error signals) in relation to the range of opportunities afforded by your current circumstances.
• When you become aware of a need, when it is felt, it governs your voluntary behavior. Choices can be made only if they are grounded in a value system - the thing that determines goodness vs badness. Otherwise, your responses to unfamiliar events would be random.
• You decide what to do and what not to do on the basis of the felt consequences of your actions. This is the Law of Affect. Voluntary behavior, guided by affect, thereby bestows an enormous adaptive advantage over involuntary behavior: it liberates us from the shackles of automaticity and enables us to survive in unpredicted situations.
• The fact that voluntary behavior must be conscious reveals the deepest biological function of feeling: it guides our behavior in conditions of uncertainty. It enables us to determine in the heat of the moment whether one course of action is better or worse than another.
• Natural selection determined our autonomous survival mechanisms, but once feelings evolved - that is the unique ability we have as complex organisms to register our own states - something utterly new appeared in the universe - subjective being.
• I think the "dawn of consciousness" involved nothing more elaborate than valenced somatic sensations and human emotions are complex versions of the same type of thing. They too are ultimately "error" signals which register deviations from your biologically preferred states, which tell you whether the steps you are taking are making things better or worse for you.
• There are seven emotional affects that can be reliably reproduced in all mammals. Many can be evoked in birds too, and some in all vertebrates, suggesting that they are at least 200m years old. These are the basic ingredients of the entire human emotional repertoire.
• Our reflexes and instincts provide rough-and-ready tools for survival and reproductive success, but they cannot possibly equip us adequately for the multiplicity of unpredicted situations and environments that we find ourselves in.
• The whole of psychoanalytic theory rests un the insight that if you take the trouble to find them, implicit instinctual tendencies can always be discerned behind explicit intentions. These are the seven:
  • Lust - We need to reproduce, at least on average. That is why sex became subjectively pleasurable in the first place, through natural selection. To satisfy sxual needs, we must supplement our innate knowledge with other skills acquired through learning. This explains the wide variety of sexual activities be indulge in, alongside the "average" form that was bequeathed by natural selection
  • Seeking - Generates exploratory foraging behavior, accompanied by a conscious feeling state that may be characterized as expectancy, interest, curiosity, enthusiasm, or optimism. Almost everything we living creatures need is "out there"; through foraging we learn, almost accidentally, what things in the world satisfy each of our needs. In this way we encode their cause-and-effect relations. Seeking proactively engages with uncertainty. It is our default emotion and we tend towards this generalized sense of interest in the world. It can be aroused even during sleep by demands made upon the mind for work, leading to problem-solving activities which must be guided by conscious feelings. Hence we dream.
  • Rage - Is triggered by anything that gets between us and whatever could otherwise meet our current needs - and causes our consciousness to feel irritated frustration up to blind fury. The feelings tell you how you are doing, whether things are going well or badly, as you try to rid yourself of an obstacle - one that is often simultaneously trying to get rid of you. If we couldn't become frustrated, irritated or angry, we wouldn't be inclined to fight for what we need; in which case, sooner or later, we'd be dead. Conscious thinking requires cortex. But the feelings that guide it do not. The circuit mediating rage is almost entirely subcortical, and, like all the other affective circuits, its final destination is the brainstem PAG.
  • Fear - The contextual factors separating fight from flight are encoded in the amygdala, which mediates both rage and fear. We humans fear dangers like heights, dark places and creatures that slither and crawl towards us, and we avoid them by the same instincts and reflexes as other mammals: freezing and fleeing. Escape behavior are facilitated by rapid breathing, increased heart rate and redirection of blood from the gut to the muscles (hence the loss of bowel control). The conscious feeling of fear tells you whether you are heading towards or away from safety. In addition to instinctual fears, we must learn what else to fear and what else to do when fearful. And once you have learned to fear something - especially if you do not consciously know why - it is very difficult to unlearn. Fear memories are "indelible".
  • Panic/Grief - Panic can frequently combine with anger: "where is she?" - The feelings I want here to be close but I also want to destroy her can lead to Guilt, a secondary emotion which inhibits Rage. Depression is characterized by the mirror opposites of the feelings that characterize Seeking. The mental anguish of loss is an elaboration of the bodily mechanisms for sensory pain.
  • Care - The maternal instinct exists in all of us to some degree, mediated by chemicals found at higher levels (on average) in females. There is an overlap between the brain chemistry and circuitry for Care, Panic/Grief and female-typical Lust.
  • Play - We need to play. It is the medium through which territories are claimed and defended, social hierarchies are formed, and in-group and out-group boundaries are forged and maintained. All juvenile mammals engage in vigorous rough-and-tumble. The associated feeling state is equally universal: fun. Biologically speaking, Play is about finding the limits of what is socially tolerable and permissible. Dominance: The 60:40 rule of reciprocity states that the submissive playmate continues playing so long as they are given sufficient opportunities to take the lead. We humans engage in preten play in which the participants try out different social roles with every-present status and power hierarchies. Play requires (and conditions) you to take into account the feelings of others. It is a major vehicle for developing empathy. Play hovers between all the other instinctual emotions - trying them out and learning their limits - and it probably recruits all parts of the brain.
    • The as if quality of Play suggests it may even be the precursor of thinking in general. Some scientists believe that dreaming is nocturnal play.
• Thinking is virtual action; the capacity to try things out in imagination; a capacity which, for obvious biological reasons, saves lives.
• Feeling is all that is required to guide voluntary behavior.
• It is not the emotions that are unconscious so much as the cognitive things they are about.
• Secondary emotions (guilt, shame, envy, jealousy) arise from conflictual situations and are learnt constructs - hybrids of emotion and cognition.
• Learning how to reconcile the various emotional needs with each other in flexible ways determines the bedrock of mental health and maturity. To manage life's problems we use emotions as a compass.

6. The Source

• The reticulate (net-like) core of the brainstem must be about 525m years old, because it is shared by all vertebrates - from fishes to humans.
• Most antidepressants - serotonin boosters - act on neurons whose cell bodies are located in a region of the reticular activating system called the raphe nuclei.
• The reticulate core of the brainstem generates affect.
• The neurological sources of affect and of consciousness are, at a minimum, deeply entangled with one another, and they may in fact be the very same machinery.
• An EEG produces graphic tracings of cortical electrical activity:
  • Delta (2Hz) waves - When the cortex is unstimulated, it produces a series of high-amplitude waves occurring roughly twice a second.
  • Theta (4-7Hz) or Alpha (8-13Hz) waves - When the cortex is stimulated by the reticular activating system in the absence of sensory input, it produces desynchronized or erratic waves.
  • Beta (14-24) or Gamma (25-100) waves - When the cortex is actively processing external information. Gamma is the rhythm most commonly associated with consciousness.
• The cortex becomes conscious only to the extent that it is aroused by the brainstem.
• Two ways in which neurons communicate with each other:
  • Synaptic transmission - Neurotransmitters (glutamate and aspartate are excitatory and gamma-aminobutyric or GABA is inhibitory) are passed from one synapse to the next. This transmission is target, binary (yes/no), and rapid.
  • Post-synaptic modulation - Neuromodulators spread diffusely through the brain. Instead of passing messages along specific "channels", they wash over swathes of the network, thereby regulating the overall "state" of the cortex.
• The distinction between "channel" and "state" is a useful shorthand for the two ways in which neurons communicate with each other. Synaptic transmission is binary but post-synaptic neuromodulation grades the likelihood that a given set of neurons will fire. It shifts the statistical odds that something will happen in them.
• Neuromodulators come from all over the body, including the pituitary, adrenal, thyroid and sex glands (which produce various hormones) and the hypothalmus (which produces innumerable peptides). But the central source of arousal from the brain's point of view is the reticular activating system. Recticular brainstem arousal releases the five best-known neuromodulators:
  • Dopamine - Sourced mainly in the ventral tegmental area and substantia nigra
  • Noradrenaline - Sourced mainly in the locus coeruleus complex
  • Acetycholine - Sourced mainly in the mesopontine tegmentum and basal forebrain nuclei
  • Serotonin - Sourced mainly in the raphe nuclei
  • Histamine - Sourced mainly in the tuberomammillary hypothalmus
  • and many others - mainly slow-acting hormones and peptides (over 100 in the brain), which modulate highly specific neural systems
• Arousal is generated mainly, but not exclusively in the brainstem and hypothalamus, and it arouses the forebrain by modulating neurotransmission.
• The shift from vegetative wakefulness to affective arousal appears to depend upon the integrity of a small, tightly packed knot of neurons surrounding the central canal of the midbrain, the periaqueductal grey (PAG), where all the brain's affective circuitry converges. We might think of the reticular activating system and PAG, respectively, as the origin and destination of forebrain arousal.
• All affective circuits converge on the PAG, which is the main output center for feelings and emotional behaviors. It divides into two groups of functional columns:
  • FEAR, RAGE and PANIC/GRIEF - The back one is for active "coping strategies" or defensive behaviors such as fight-or-flight reactions, increased blood pressure and non-opioid pain relief.
  • LUST, CARE and SEEKING - The front one is for passive coping/defensive strategies such as freezing with hyporeactivity, long-term sick behavior, decreased blood pressure and opioid pain relief.
• The PAG must set priorities for the next action sequence. It renders its verdict with the help of an adjacent midbrain structure, known as the superior colliculi.
• Bjorn Merker calls this affective/sensory/motor interface between the PAG, the superior colliculi and the midbrain locomotor region the brain's "decision triangle". Panksepp called it the primal SELF, the very source of our sentient being.
• The deepest layer of the superior colliculi consists in a map that controls eye movements.
• Once the midbrain decision triangle has evaluated the compressed feedback flowing in from each previous action, what it activates is an expanded feedforward process which unfolds in the reverse direction, through the forebrain's memory systems, generating an expected context for the selected motor sequence. This is the product of all our learning. In other words, when a need propels us into the world, we do not discover the world afresh with each new cycle. It activates a set of predictions about the likely sensory consequences of our actions, based upon our past experience of how to meet the selected need in the prevaling circumstances.
• "Predicting the Present": Jackob Hohwy's term for the mental process that controls voluntary behavior.
• Most people don't realize that our here-and-now perceptions are constantly guided by predictions, generated mainly from long-term memory. But they are. That is why far fewer neurons propagate signals from the external sense organs to the internal memory systems than the other way round.
• Why treat everything in the world as if you'd never encountered it before? Instead, what the brain does is propagate invards only that portion of the incoming information which does not match its expectations. That is why perception is nowadays sometimes described as "fantasy" and "controlled hallucination"; it begins with an expected scenario which is then adjusted to match the incoming signal. In this sense, the classical anatomists were right: cortical processing consists mainly in the activation of "memory images" suitably rearranged to predict the next cycle of perception and action.
• Perception, action, and cognition are only ever felt because they contextualize affect. It's as if our perceptual experience says: "I feel like this about that."
• Perception and action are an ongoing process of hypothesis testing in which the brain constantly tries to suppress errror signals and confirm its hypotheses. The more your hypotheses are confirmed, the more confident you are, and the less aroused - less conscious - you need to be. You can automatize your action sequences and drift off into the default mode. But if you find yourself in an unexpected situation - one in which your predictive model appears to shed no reliable light - the consequences of your actions become highly salient. You switch out of autopilot and become hyper-aware: the decision triangle carefully adjusts your predictions as you feel your way through the consequences of your actions and make new choices.

7. The Free Energy Principle

• Karl Friston explains that biological systems such as cells must have emerged through complex versions of the same process that formed simpler self-organizing systems such as crystals from liquid, because they share a common mechanism - "free energy minimization". All self-organizing systems, including you and me, have one fundamental task in common: to keep existing and Friston believes that we do this by minimizing our free energy.
• Remaining within the viable bounds of our emotions requires us to work: to maintain close proximity with our caregivers, to escape from predators, to get rid of frustrating obstacles and so on. Beyond a certain level of predictability, the work required to do these things is regulated by feelings.
• Every homeostat consists of three components:
  • A receptor
  • A control center
  • An effector
• Homeostasis runs in the opposite direction to disorder, dissipation, dissolution. It resists entropy. It ensures that you occupy a limited range of states. That is how it maintains your required temperature, and how it keeps you alive - how it prevents you from dissipating. Living things must resist one of the fundamental principles of physics: The Second Law of Thermodynamics.
• Entropy always increases on the large scale. It may in fact be the physical basis for the fact that time itself appears to have a direction and a flow.
• 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.
• The fewer the possible states, the lower the entropy.
• The most basic function of living things is to resist entropy.
• Increasing entropy means decreasing predictability. The entropy associated with expanding gases and expanding options is the same thing.
• The more information required to describe the microstate of a system (ie the state of each and every molecule), the greater the thermodynamic entropy.
• Entropy is minimal when the answer to every yes/no question is entirely predictable, ie when nothing is learnt and there is no information gained.
• 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.
• The EEG entropy values are higher in minimally conscious than in vegetative patients. That makes sense: cortical activity in the conscious brain communicates more information thatn it does during deep sleep. But here comes the strange part: if more information means more uncertainty and therefore more entropy, then - since living things must resist entropy - waking activity is less desirable, biologically speaking, than deep sleep.
• Probability is not quite the same as information in Shannon's sense, which entails the additional factor of communication. Unlike probabilities - which exist in and of themselves - communication requires both an information source and an information receiver.
• John Wheeler: "That which we call reality arises in the last analysis from the posing of yes/no questions and the registering of equipment-evoked responses; in short... all things physical are information-theoretic in origin.
• So:
  • The average information of a system is the entropy of that system (ie the entropy is a measure of the amount of information needed to describe its physical state)
  • Living systems must resist entropy. We must minimise the information (in Shannon's sense) that we process, ie our uncertainty.
  • We living systems resist entropy through the mechanism of homeostasis. We receive information about our likely survival by asking questions (ie taking measurements) of our biological state in relation to unfolding events. The more uncertain the answers are (ie 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)
• Natural selection fitted each species to its ecological niche: each creature's survival depends only on things that are in fact reliably found in its natural habitat. So, we need air because we can expect it.
• Kant first spoke of self-organization. Then Darwin discovered natural selection. Then Norbert Wiener founded the discipline of cybernetics, adding the notion of feedback to Shannon's understanding of information. William Ross Ashby used this notion of feedback combined with statistical physics to show 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 (ie to resist entropy).
• Markov blanket - A statistical concept which separates two sets of states from each other.
  • Such formations induce a partitioning of states into internal and external ones, ie 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 external states can only be "sense" vicariously by the internal ones as states of the blanket.
  • Moreover, a Markov blanket is itself partitioned into subsets that are, and subsets that are not causally dependent (directly) upon the states of the external set. These states of the blanket are called sensory and active states. Thus we have internal, active, sensory, and external states where the external states are not part of the self-organizing 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. 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.
  • Once you start looking, Markov blankets are everywhere - cell membranes and the skin and musculoskeletal system of the body as a whole, every organelle, organ, and physiological system. The brain (actually, the entire nervous system) - which regulates the body's other systems - therefore possesses a Markov blanket. In fact, it is a meta-blanket, since it surrounds all the other blankets. Self-organizing systems can always be composed of smaller self-organizing 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.
  • The very selfhood of a complex dynamical system is constituted by its blanket. Such self-organizing 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-organization are in fact the essential preconditions for subjectivity.
  • The very nature of a Markov blanket is to induce a partitioning of states into "system" and not-system ones, in such a way that not-system states are hidden from the system's interior and vice-versa.
  • The Markov blanket endows the internal states of self-organizing systems with a capacity to represent hidden external states probabilistically, so that the system can infer the hidden causes of its own sensory states, which is something akin to the function of perception. This capacity, in turn, enables it to act purposively upon the external milieu, on the basis of its internal states - which actions are akin to motor activity
  • The system maintains and renews itself in the face of external perturbations. Merely being a self-organizing system is sufficient to confer a purpose on it and on each of its parts, and that is the function of the active states of the blanket: they manipulate the environment in order to maintain the integrity of the system. Which means that, along with an enclosed self, a subjective point of view, a goal and the capacity both to sense and act, the mere fact of a Markov blanket brings about something akin to agency.
  • This is where the concept of "expected states" comes from, and why biological self-organizing systems are homeostatic. Homeostasis seems to have arisen with self-organization. The sensory and active states of a Markov blanket are nothing other than a self-organizing system's receptors and effectors, and the model of external states that it generates is its control center.
  • Biological self-organizing systems must test their models of the world, and if the world does not return the answers they expect they must urgently do something differently or they will die. Deviations from expected states are, therefore a foundational form of Wheeler's equipment-evoked responses. This is how question-asking arises; self-organization beings participant observers into being. The question that a self-organizing system is always asking itself is simply this: "Will I survive if I do that?" The more uncertain the answer, the worse for the system.
• Friston's four fundamental properties of all biological self-organizing systems:
  • They are ergodic (occupy limited states
  • They are equipped with a Markov blanket
  • They exhibit active inference
  • There are self-preservative
• The equation is A = U - TS (free energy is equal to the total internal energy minus the energy already employed):
  • A is free energy
  • U is total internal energy
  • T is temperature
  • S is entropy
• The are three types of free energy:
  • Helmholtz - Classical thermodynamic free energy
  • Gibbs - Chemical-ensemble free energy
  • Friston - Information free energy - Friston free energy is equal to average energy minus entropy
    • Average energy means the expected probability of an event happening under a model
    • 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.
• If biological systems must minimize their entropy, and entropy is average information, then it follows that they must keep the flow of information they process to a minimum. They must minimize unexpected events. This is technically known as "surprisal". Like entropy, surprisal is a declining function of probability: as the probability goes down, the surprisal goes up. Surprisal measures how unlikely it is expected to be (on average)
• Self-organizing systems must minimize information flow, because increasing information demand implies increasing uncertainty in the predictive world.
• Friston free energy is a quantifiable measure of the difference between the way the world is modeled by a system and the way the world really behaves. Therefore, we must minimize this difference. A system's model of the world must match the real world as closely as possible, which means that it must minimize 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. Because we are insulated form the world by our Markov blankets, we must bring the whole process of minimizing surprisal inside our heads, and become both the source and receiver of the information that flows from our question asking. We do this by measuring relative entropies - by quantifying the gap between the sensory states predicted by an action and the sensory states that actually flow from that action. This yields the quantity called Friston free energy, which is always a positive value greater than the actual surprisal.
• Generative models come into being with self-organizing 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 test of a good model of the self-in-the-world is how well it enables the self to engage the world in ways that keep it within its viable bounds. The better these engagements are, the lower its free energy will be. The lower its free energy, the more of the system's energy is being put to effective, self-preserving work. The Free Energy Principle thus explains in mathematical terms how living systems resist the Second Law of Thermodynamics through homeostasis-maintaining work.
• Self-organizing systems are obliged to ask questions of themselves about their own states. Specifically: "What will happen to my free energy if I do that?" The answer to this question will always determine what the system does next, over a suitable time period. This is the causal mechanism behind all voluntary behavior.
• 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. Perception is an endeavor to self-generate the incoming sensory signals and thereby explain them away. That is why so many neuroscientists nowadays speak of the Bayesian brain.
• 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". The obligation to minimize our free energy is the principle that governs everything we do.
• All the quantities in a self-organizing system that can changes will change to minimize free energy and everything that we call mental life becomes mathematically tractable.

8. A Predictive Hierarchy

• The brain's many complex functions really can, ultimately, be reduced to a few simple mechanisms.
• Jakob Hohwy: "The brain is somewhat desperately, but expertly, trying to contain the long and short-term effects of environmental causes on the organism in order to preserve its integrity. In doing so, a rich, layered representation of the world implicitly (unconsciously) emerges. This is a beautiful and humbling picture of the mind and our place in nature."
• Principles of a predictive hierarchy:
  • The brain conspires to anticipate and thus "explain away" events in the world. It suppresses predictable, uninformative incoming signals that it would otherwise have to process pointlessly. Each level in its hierarchy receives only the newsworthy, unexpected information transmitted from the level immediately beyond it. These feedback reports are prediction errors.
  • This hierarchy unfolds over progressively smaller temporal and spatial scales. The core predictions apply in all circumstances, whereas the more peripheral ones are fleeing and focal.
    • A predictive sequence unfolds from body-monitoring nuclei located in the brainstem and diencephalon, via the basal ganglia and limbic system, through the neocortex, to the modelity-specific sensory receptors located in the end organs (ie, the rods and cones of the retina), which have very narrow receptive fields.
    • At the periphery, short-term accuracy and complexity prevail at the cost of long-term generalizability, which is enjoyed by the deeper predictions.
  • A hierarchy of plasticity exists in terms of which the core predictions cannot change but the peripheral ones can and do; they are subject to instantaneous updating, with the intermediate degree of plasticity.
  • Perception (as opposed to learning) reverses the direction of information processing. Perception proceeds from the inside outwards, always from the viewpoint of the subject. What you see is your "best guess" as to what is actually out there; it is your proposed answer to the questions you are currently putting to the world.
    • Actions should therefore be viewed as experiments that test hypotheses arising from the generative model. If an experiment does not yield the predicted sensory data, then the system either must change its prediction to better explain the data or, if it remains confident about the original prediction, must obtain better data; that is, it must perform actions that will change its sensory input.
    • These two options - changing the prediction or the input - are the fundamental mechanisms of perception and action respectively.
    • In some respects, perception and action are more similar than they seem.
    • Suppressing prediction error is what controls action, no less than perception.
    • The multiple bodily homeostats regulated and orchestrated by the midbrain's meta-homeostat are the pivot of the mechanism by which we stay alive, for the simple reason that homeostatic regulation maintains our bodies within their viable bounds. These bounds cannot be changed. This means that something else in the system must change. This is the formal, mechanistic explanation of the imperative link that exists between drive and action, and it is why there must be a hierarchy of prior prediction, some of which can be changed and some of which cannot.
    • It is only action that can increase the probabilities of prior predictions - some of which simply cannot be changed.
• Where does the background knowledge come from, at the outset, before the system has gathered any evidence about the world? Our core "expected states" are encoded by our species as innate homeostatic settling points - quantities that were determined by what worked effectively for our evolutionary ancestors. We are beneficiaries of the biological successes of past generations, which fix the most basic premises of our existence.
• Prediction errors are the sensory signals that were not predicted by a current hypothesis, ie the ones that were not self-generated. This is the salient part of the data.

9. Why and How Consciousness Arises

• The basic question that living things must always ask themselves is "What will happen to my free energy if I do that?
• There is an executive bottleneck: you can only do one or two things at once. This means that, to select your next action, you must rank your current needs by urgency. That is why internal needs must be prioritized in relation to prevailing external conditions.
• We are born with species-specific predictions about what to do in states like hunger, thirst, fear, and rage. These innate predictions are called "reflexes" and "instincts".
• Affective valence - our feelings about what is biologically good and bad for us - guides us in unexpected situation. We concluded that this way of feeling our way through life's unpredicted problems, using voluntary behavior, is the biological function of consciousness. It guides our choices when we find ourselves in the dark.
• Statisticians call the exponential increase in computational resources necessitated by a linear increase in model complexity the "combinatorial explosion".
• On the principle of Ockham's razor (the Law of Parsimony), we want simple predictive models. Simplification is essential if our models are going to apply in a wide range of situations. They must be serviceable, not only here and now but also in many other contexts.
• Affects are always subjective, valenced, and qualitative. They have to be, given the control problem they evolved to handle.
• This sort of thing determines what a system does next. In other words, it determines which active states will be selected by the generative model to resolve the prioritized category of uncertainty. It is as if the system says: under present conditions, this is the category of prediction-error processing in which computational complexity cannot be sacrificed.
• Crucially, shifting into FEAR mode means that the prioritized need has become an affect. In other words, it has become conscious. Why? It becomes conscious so that deviations from expected outcomes in the most salient category of need will be felt throughout the predictive hierarchy. That is what affect is. It is the "equipment-evoked response" to the question the system asked of itself: "which of these converging error signals provides the greatest opportunity for minimizing my free energy?
• The purpose of precision modulation is to ensure that the inferences made by predictive models are driven by reliable learning signals (trustworthy news): If there is high confidence in a signal then it should be allowed to revise a prior hypothesis, and if there is low confidence then it should not.
• This means that we must minimize precise error signals. Again, that sounds paradoxical, until you realize that it just means we must avoid making glaring mistakes. The only way this can be achieved is by improving our generative models, thus increasing the mutual information between our models of the world and the sensory samples we obtain from it. In other words, we must maximize the precision of our predictions and then seek precise confirmatory data.
• We must maximize our confidence in the beliefs that guide our actions. This is called "precision optimization".
• Precision values quantify expectations about variability. So, they are representations of uncertainty. How confident am I about this error signal in the present context? How much weight should I give to it, right now? Is 8/10 for A worth more or less than 8/10 for B under present conditions?
• So much of our experience just is little pulses of feeling, as you notice things that aren't quite as you expected them to be, followed by cognitive castings around for ways to close the gap. You remember an email you need to send: it is only when your hand fails to detect the heard screen of your phone that you realize you were already reaching for it - but if it isn't right there beside you, where did you leave it? In the kitchen, where you were five minutes ago?
• A prioritized need (in this case LUST) is the currently most salient source of uncertainty. Inferences about its causes become conscious as affect, because fluctuations in your confidence level concerning the possible actions required to meet this need must be modulated by feelings. The feelings tell you how well or badly you are doing.
• It is of capital importance to note that the statement "the unfolding context giving rise to the fluctuations must become conscious too" explains why experience has dual aspects. It is not merely a matter of "I feel like this" but rather "I feel like this about that". The "about that" must be felt too, using a common currency (applied uncertainty) - because context is the main source of uncertainty over free energy. The economics of free energy minimization demands a common currency.
• Simply put, relatively strong signals attract attention: they are assigned higher precision.
• That is how saliency works. "Salient" features of the world are features that, when sampled, minimize uncertainty concerning the system's currently prioritized hypothesis: they are the ones that, when things unfold as expected, maximize our confidence in the hypothesis. Active agents are thus driven to sample the world so as to (attempt to) confirm their own hypotheses.
• The perceptual orientation of each species is dictated by the things that matter to it.
• Precision cannot be determined passively; we cannot just wait and see which signals are strong without any expectations either way. It must be inferred and then assigned by the generative model. Attention - which has everything to do with precision - can accordingly be both "grabbed" and "directed".
• Two equations:
  • Free energy is (approximately) the negative logarithm of the probability of encountering some actively authored sensory states.
  • The expected free energy decreases in (approximate) proportion to negative log precision.
• There are three ways for a self-evidencing system to reduce prediction error and thereby minimize free energy:
  • It can act to alter sensations so that they match the system's predictions (action)
  • It can change its representation of the world to produce a better prediction (perception)
  • It can adjust precision to optimally match the amplitude of the incoming prediction errors.
• Consciousness if this final optimization process, the optimization of the system's confidence, that we associate with the evaluation of free energy that underpins felt experience.
• The rate of change of precision over time depends on how much free energy changes when you change precision. This means that precision will look as if it is trying to minimize free energy.
• Exteroception, proprioception and interoception can all occur without consciousness, but consciousness if the feeling of these things.
• Feelings are fluctuating, existentially valued, subjective states with differentiated qualities and degrees of confidence. This is the stuff of consciousness.

10. Back to the Cortex

• The world as we experience it is literally generated from cortical representations. Within the predictive coding framework, odd as it seems, what we perceive is a virtual reality constructed from the mind's own building materials.
• What you perceive is not the same thing as the input that arrives from your senses. What you perceive is an inference. And the materials from which that inference is derived are for the most part your cortical predictive model derived from past (ie expected) experiences.
• This is the whole point of consciousness in cognition. You arrive at a situation in which you aren't sure what to do. Consciousness comes to the rescue: you feel your way through the scenario, noting the voluntary actions that work for you. Then, gradually, the successful lessons become automatized and consciousness is no longer needed.
• I want to emphasize that the cognitive work I have just described slows down the otherwise automatic business of acting in the world. This is the essential difference between voluntary and involuntary action, conscious vs unconscious cognition, felt drive vs autonomic reflex; the voluntary type is less certain and therefore requires more time.
• What all of this implies is that the conscious state is undesirable from the viewpoint of a self-organizing system.
• In that theoretical ideal state, in which our needs are met automatically, we feel nothing. (This is how most of our bodily needs are met: they are regulated autonomically). I say "theoretical ideal" because, in respect of many of our needs, especially the emotional one, we never get there. The SEEKING drive alone ensures that.
• Freud's aphorism, "consciousness arises instead of a memory trace", should make more sense now. It means that consciousness arises when automatic behavior leads to error, in other words, when the memory trace (a prediction) producing a behavior does not have the expected outcome. This means that the prediction in question must be updated to accommodate the error. Cortical consciousness may therefore be described as "predictive work in progress". A memory trace that is conscious is in the process of being updated. It is no longer a memory trace. Hence: consciousness arises instead of a memory trace.
• An activated memory is an aroused memory; and an aroused memory is a memory no longer - it is in a state of uncertainty. All I am trying to convey here is that cognitive consciousness boils down to a rendering labile of cortical memory traces, and that this liability is a product of arousal. We keep arriving from different directions at the same insight: cortical processes are fundamentally unconscious things (they are simply algorithms, if left to their own devices). Consciousness - all of it - comes from the brainstem.
• Learning requires consciousness, as we gradually improve confidence in our newly acquired predictions. But the ideal of all learning is to automatize these acquired predictions too, to make them behave like reflexes and instincts.
• The most important fact about non-declarative memory is that it is non-declarative. It generates procedural responses, whereas declarative memory generates experienced images.
• Subcortical memory traces cannot be retrieved in the form of images for the reason that they do not consist in cortical mappings of the sensory-motor end organs.
• The cortical memory system, by contrast, are always ready to revive the predictive scenarios they represent - literally to re-experience them.
• The cortex specializes in contexts; it restores model accuracy in unpredictable situations. A trade-off is inevitable. The more potential for conscious experience, the less automaticity, which means more plasticity but also more cognitive work. That costs energy, and it generates feelings, so the brain does as little of it as it can get away with. Even to the point of fading out a stimulus that is right before your eyes.
• By activating memories, we can strengthen, alter, and even erase them. Three types of thinking help do this
  • Mind-wandering is one means by which this is achieved. It involves spontaneous forebrain activity (also know as the resting state or default mode), which occurs in the absence of any specific external stimulus. This kind of activity goes on much of the time in the background, through an "imaginative exploration of our own mental space". There is a good deal of overlap between this form of thinking and dreaming, which seems to occur in all creatures equipped with a cortex; any animal with the capacity to generate images of itself acting in the world can also meander through endless simulated worlds as its circumstances permit. Meandering is tightly bound up with the SEEKING drive, which continues with its demands as we sleep.
  • What all conscious cognitive processes have in common is that they entail the necessary mental work of reconsolidation - the returning of consolidated predictions to states of uncertainty. That is why dreams (which are a form of problem-solving) are conscious.
  • Deliberate imagining - to imagine doing things in order to gauge in advance the probable consequences of actually doing them.
    • Contemporary research on episodic memory reveals that the hippocampus is in fact just as involved in imagining the future a it is in reliving the past. David Ingvar speaks of "remembering the future".
  • Thinking with words - Though language is a tool of communication it is, first and foremost, a tool for abstraction. With its marvelous gradations of generality and specificity, language lets us project something of the structure of the predictive hierarchy itself into consciousness. These powerful aids to cognition are not available to non-symbolic species. It is very difficult to imagine the whole of science, technology, and culture without language.
• The cortex contributes more to PLAY than to any of the other basic emotions.

11. The Hard Problem

• Ribot's Law - Your internally experienced memory of what happened ten years ago is more securely consolidated than what happened ten minutes ago. This is why elderly people are more likely to forget recent events than remote ones.
• Miller's Law - At any given moment, you can only retain seven units of information (plus or minus two). The duration of short-term memories can also be measured: they typically last between 15-30 seconds.
• Both laws are psychological and physiological (ie bound by chemical limits)

12. Making a Mind

• We may suppose that consciousness did not exist on earth before brains evolved - and perhaps only when vertebrate brains evolved; therefore, about 525m years ago. I suspect it arose in rudimentary form before that; that a precursor of affect gradually became felt affect, with no sharp dividing line between them, in tandem with the evolution of increasingly complex organisms with multiple competing needs.
• What emerged with the evolution of cortex was cognitive consciousness - that is the additional capacity to contextualize affect exteroceptively and hold it in mind:
  • Before the cortex evolved, an animal might just feel "bad" (affect). With a cortex, the brain can link that "bad" feeling to something specific in the environment—like a predator or a rotten piece of fruit. It puts the feeling into a context based on what is happening around you.
  • Without this capacity, an organism reacts purely in the "now." With a cortex, you can keep a feeling and the reason for it active in your brain even after the stimulus is gone. You can "stew" on a problem or plan a response because you can hold the information in your conscious workspace.
  • "basic" consciousness is just feeling things (being awake and reactive). "Cognitive" consciousness is the higher-level ability to think about those feelings, label them, and use them to make complex decisions.
• David Chalmers three principles for solving the "hard problem":
  • The Principle of Structural Coherence
  • The Double Aspect Principle
  • The Principle of Organizational Invariance
• Solms does not conceive of consciousness as being particularly intelligent - at least not in its elementary form.
• Sheep and cows and pigs are fellow mammals and are subject to the same basic emotions that we are, such as FEAR, PANIC/GRIEF, and CARE. Mammals possess a cortex too, which means they are capable - all of them to some degree - of consciously "remembering the future" and feeling their way through its probabilities and likelihoods.
• As we relinquish the familiar illusion that consciousness flows in through our sense, and the misconception that it is synonymous with understanding, let us take comfort in the fact that it actually comes spontaneously from our inmost interior. It dawns within us even before we are born. At its source, we are guided by a constant stream of feelings, flowing from a wellspring of intuition, arising from we know not where. Each of us individually does not know the causes, but we feel them. Feelings are a legacy that the whole history of life has bestowed upon us, to steel us for the uncertainties to come.

Postscript

• Self-organizing systems survive because they occupy limited states; they do not disperse themselves. This survival imperative led gradually to the evolution of the complex dynamical mechanisms that underwrite intentionality. Crucially, the selfhood of self-organizing systems grants them a point of view. That is why it becomes meaningful to speak of the subjectivity of such a system: deviations from their viable states are registered by the system, for the system, as needs.
• Feeling by an organism of fluctuations in its own needs enables choice and thereby supports survival in unpredicted contexts. This is the biological function of experience.
• Needs cannot all be felt at once. They are prioritized by a midbrain decision triangle, where current needs (residual prediction errors, quantified as free energy) converging on the periaqueductal grey are ranked in relation to current opportunities (displayed in the form of a 2D "saliency map" in the superior colliculi). This triggers conditioned action programmes, which unfold in expected contexts over a deep hierarchy of predictions (the generative model of the expanded forebrain). The actions that are generated by prioritized affects are voluntary, which means they are subject to here-and-now choices rather than pre-established algorithms. Such choices are felt in exteroceptive consciousness, which contextualizes affect.

Appendix: Arousal and Information

• Arousal, fueling drive mechanisms, potentiates behavior, while specific motives and incentives explain why an animal does one thing and not another.
• Generations of behavioral scientists have both theorized and experimentally confirmed that a concept like arousal is necessary to explain the initiation, strength, and persistence of behavioral responses. Arousal provides the fundamental forces that makes animals and humans active and responsive so they will perform instinctive behaviors or learned behaviors directed toward goal objects.
• Pfaff's own principal component analyses suggest that the proportion of behavior across a wide range of data that can be accounted for by "generalized arousal" is between 30-45%.
• How do internal and external influences wake up brain and behavior, whether in humans or in other animals?
• Generalized arousal is higher in an animal or human being who is:
  • S - More alert to sensory stimuli of all sorts
  • M - More motorically active
  • E - More reactive emotionally.
• This is a concrete definition of the most fundamental force in the nervous system. All three components can be measured with precision. Clearly there is a neuroanatomy of generalized arousal, there are neurons whose firing patterns lead to it, and genes whose loss disrupts it. Therefore generalized arousal is the behavioral state produced by arousal pathways, their electrophysiological mechanisms and genetic influences. The fact that these mechanisms, produce the same sensory alertness (S), motor reactivity (M), and emotional reactivity (E) as our definition affirms the existence of a generalized arousal function and the accuracy of its operational definition.
• In Shannon's equation, the information in any event is in inverse proportion to its probability. Put another way, the more uncertain we are about the occurrence of that event, the more information is transmitted, inherently, when it does happen. When all events in an array of events are equally probable, information is at its top value. Disorder maximizes information flow.
• For a lower animal or human to be aroused, there must be some change in the (interoceptive or exteroceptive) environment. If there is change, there must be some uncertainty about the state of the environment. Quantitatively, to the degree that there is uncertainty, predictability is decreased.
• Arousal of brain and behavior, and information calculations are inseparably united.
• Unknown, unexpected, disordered, and unusual (high-information) stimuli produce and sustain arousal responses.
• Central nervous system arousal systems battle heroically against the Second Law of Thermodynamics in a very special way. They respond selectively to environmental situations that have an inherently high entropy - a high degree of uncertainty and therefore information content. But in responding, CNS arousal systems effectively reduce entropy by compressing all of that information into a single, lawful response. Arousal neurobiology is the neuroscience of change, uncertainty, unpredictability, and surprise - that is of information science.
• The huge phenomenon called habituation, a decline in response amplitude on repetition of the same stimulus, pervades neurophysiology, behavioral science, and autonomic physiology; and it shows us how declining information content leads to declining CNS arousal. Thus arousal theory and information theory were made for each other.
• It is important to recognize that the "mathematics of information" explains the behavior of neurons in both arousal processes and learning processes, which, combined, determine what the brain does. Therefore, although "information" is not a physiological construct, it lawfully explains the physiological activity of the brain. It is the function that is selected by evolution; the physiological phenotypes follow.