<p><em>When we see, hear, smell and touch, we’re not receiving reality—we’re creating it. Consciousness is a tapestry of predictions and fantasies, stitched together in the brain. A world outside of our heads may exist but is forever out of reach. These are radical conclusions of the world’s most influential and highly cited neuroscientist, <strong>Karl Friston</strong>. In this exclusive interview with the Institute of Art and Ideas, he explains how his Free Energy Principle unifies physics and psychology, explains agency and dreaming, and promises new ways to understand and treat mental illness.</em></p><p> </p><p><strong>Alasdair Craig: You say that conscious experience is a prediction generated by the brain. What does this mean?</strong></p><p>Karl Friston: You could probably trace that back to the days of Plato, but to my mind it was beautifully articulated by the German physicist and physician <a href="https://plato.stanford.edu/entries/hermann-helmholtz/" target="_blank">Hermann von Helmholtz</a>, in terms of the notion of unconscious inference. The basic idea is that the brain is a very constructive organ. It <a href="http://iai.tv/iai-academy/courses/info?course=predicting-ourselves" target="_blank">generates predictions</a>, hypotheses, <a href="http://iai.tv/video/down-the-rabbit-hole" target="_blank">fantasies</a>, explanations that best explain the sensory impressions on our eyes and our ears and our bodies. So this is very much, as people like <a href="http://iai.tv/articles/anil-seth-the-hallucination-of-consciousness-auid-2525" target="_blank">Anil Seth</a> and <a href="http://iai.tv/articles/where-is-my-mind-an-interview-with-andy-clark-auid-1162" target="_blank">Andy Clark</a> would say, an <a href="https://doi.org/10.1111/sjp.12120" target="_blank">inside-out, a constructive, active process of generating predictions</a>, as opposed to an outside-in process of passively absorbing data.</p><p>The idea is that if I’ve got a good grip on the world, if I’m doing good sense-making, then I should be able to predict exactly what is arriving on my sensory epithelia. In other words, if I’m tracking the world in the right kind of way, there will be no prediction error, there will be no mismatch between my predictions and what I’m actually sensing. And that leads to the notion of <a href="https://doi.org/10.1038/s41593-018-0200-7" target="_blank">predictive coding</a>, the imperative to minimize one’s prediction errors—the mismatch between what we predict should happen and the sensory evidence we receive. If the predictions are not correct then we have a prediction error and we have to update our beliefs.</p><p> </p><p><strong>Most of us probably assume that in perception you first get sensory input, and then you construct the world on that basis. But you’re saying it’s the other way round, you first construct a prediction, which is your world-model, then you get sensory input, and then you update your prediction.</strong></p><p>You test that sensory input against your predictions. Literally, your brain cells can compare what’s coming in and what you thought should be coming in, and then the difference is the newsworthy information, it’s the bit you couldn’t predict. And then those prediction errors—the surprising sensory information—you use to update your beliefs, to drive changes in your brain, literally changing your mind, so that you adjust your model of the world to then generate better predictions.</p><p class="article-plus-content--header" style="text-align: center;">___</p><p class="article-plus-content--header" style="text-align: center;">We are indeed trapped within our own heads, in many senses.</p><p class="article-plus-content--header" style="text-align: center;">___</p><p> </p><p><strong>Why is the brain doing it this way round? What problems does it solve by doing this?</strong></p><p>The problem it solves is that we do not have access to the causes of our sensations. This is very often articulated in terms of the idea that <a href="https://doi.org/10.1017/S0140525X12002191" target="_blank">the brain is skull-bound</a>, it lives in the dark. It just has very sparse information through the sensory organs, sampling very small parts of the world.</p><p>To take an example, the way that we actually look at the world is by palpitating it every 250 milliseconds, four times a second. We’re only able to see a very small part of the world at any one time, so we have to stitch this together, and the way that we do this is by predicting: what would I see if I looked over there? So I have a prediction for what I would see over there, and if I’m right, then I can accumulate the evidence for my internal construction, my internal hypothesis. You don’t have direct access to what’s out there. So in a sense you’re solving the problem of living in the world where you only have very limited access to the actual causes of your sensations.</p><p> <span class="article-content-box"> <a href="video/consciousness-in-the-clouds" target="_blank" class="iai-related-in-article click_on_suggestion_link--gtm-track"> <span class="iai-card"> <span class="iai-card--image iai-related--video-play" style="display: block;"> <img src="/assets/Uploads/246.Consciousness-in-the-clouds-copy.webp" class="iai-related--primary-image" alt="related-video-image"> </span> <span class="iai-card--content" style="display: block;"> <span class="iai-card--title" style="display: block;">SUGGESTED VIEWING</span> <span class="iai-card--heading" style="display: block;">Consciousness in the clouds</span> <span style="display: block;">With Massimo Pigliucci, Shini Somara, Anders Sandberg, Nadine Dijkstra, Roman Yampolskiy</span> </span> </span> </a> </span> </p><p> </p><p><strong>Does this mean that we never really access reality outside of our heads?</strong></p><p>We are indeed trapped within our own heads, in many senses. This is not to deny the fact that there is some metaphysical world out there. But perhaps a nice way to think of this is that there’s the world out there, which we will never know, and then there’s the world on the inside—our brains and our neuronal dynamics—and they somehow become <a href="https://doi.org/10.1016/j.concog.2014.12.003" target="_blank">mutually entrained, they become synchronised</a>, so that knowing the brain state will tell you something about the state of the world that’s generating the sensations. This happens because the two systems—me and the rest of the world—talk to each other through our sensations.</p><p>So everything plays out <a href="http://iai.tv/video/the-dance-of-life" target="_blank">at the level of sensory impressions</a>, and that’s where prediction gets into the game—if you can on the inside always keep track of what’s going on, just by predicting what you’re sensing, ultimately you become synchronised with the outside world. So, they’re separate yet fundamentally joined together through the sensorium.</p><p>So, which is the reality? <a href="http://iai.tv/video/perception-as-a-fantasy-donald-hoffman" target="_blank">Well, it’s your reality on the inside</a>. There might be another reality, but we could only ever really know that if there was someone else observing the same reality, and magically we could somehow communicate that to each other. </p><p> </p><p><strong>And this is captured in the phrase that perception is a “</strong><a href="https://doi.org/10.1111/phib.12268" target="_blank"><strong>controlled hallucination</strong></a><strong>”?</strong></p><p>Yes. But just to push back a little bit on that… I know exactly what people mean by that. But the “control,” I think, slightly gives you a false impression: it’s not that reality is controlling what’s happening on the inside. This is genuinely a balanced dance: it’s a mutual entrainment. Let’s take vision again, I choose what I am sampling, so I am actually causing my own sensations. So I would replace “control” with “entrainment.”</p><p>I don’t know if you remember Huygens’ clocks—this notion of <a href="https://doi.org/10.3390/e17041701" target="_blank">generalized synchronicity</a>? Huygens was a seventeenth-century theoretician, who observed that if you went to a clockmaker’s workshop, after a sufficient period of time, two clocks suspended from the same wall would eventually come to swing in synchrony. This is called generalized synchrony or synchronization of chaos. It says that the only solution—from a physics perspective, the only free-energy-minimizing solution—for something that is loosely coupled to something else is this entrainment and synchronization. So I think this is what people mean by “controlled hallucination”: the world out there is somehow controlling your internal neuronal dynamics, your brain dynamics, but at the same time you are controlling the world out there.</p><p>You know, the world includes my eyes which I move around, so that speaks to another perspective on perception—that it is quintessentially active. It’s more than just constructing predictions: we sample the world actively.</p><p> </p><p><strong>How should we think about conscious experience on this picture? When I look at the table, it appears to be outside of my head. Are you saying that it’s not? Is it floating around in my head somewhere? Is it a physical thing?</strong></p><p>My view is not terribly well informed because I’m not a philosopher, but I will appeal to a lot of my friends who are philosophers. But, first of all, I’ll appeal to Helmholtz. This picture of sense-making he termed “<a href="https://philpapers.org/rec/HATPAU" target="_blank">unconscious inference</a>,” so there was no implication that we would be explicitly aware that we were perceiving. So, I think you have to bring something else to the table to talk about that kind of consciousness.</p><p>My preferred take on this phenomenology is due to people like <a href="http://iai.tv/home/speakers-and-authors/thomas-metzinger" target="_blank">Thomas Metzinger</a>, where there’s different kinds of sense-making that can be <a href="https://doi.org/10.1023/B:PHEN.0000007366.42918.eb" target="_blank">opaque or transparent</a>. I can <a href="https://doi.org/10.2307/2214186" target="_blank">just see something</a> (which is transparent sense-making), or I can know <em>that</em> I am seeing something (which would be the opaque version). Rather than just <a href="https://doi.org/10.1111/1468-0017.00205" target="_blank">seeing something through a transparent window</a>, I now realise that I’m looking at a picture of something. So that suggests that there are special kinds of systems—basically things like you and me—that have the ability to go beyond transparent sensation, transparent qualitative experiences, and actually know that they are looking at something. You’re actually seeing your own perception on the inside.</p><p class="article-plus-content--header" style="text-align: center;">___</p><p class="article-plus-content--header" style="text-align: center;">You have a seamless, unitary explanation for both action and perception, both in the service of minimizing free energy. Once you look at it like that you suddenly realize you’re talking about a fundamental phenomenon that characterizes all self-organization.</p><p class="article-plus-content--header" style="text-align: center;">___</p><p>This immediately tells us that there is internal action and internal perception, which means that the brain must have a hierarchical structure, the kind of hierarchy you’d see in deep learning, for example. The very word “deep” means that there’s a hierarchical depth, so that one level of sense-making can look at the level below and be informed by the level above.</p><p>If you’re a psychologist, this is <a href="https://doi.org/10.3389/fnhum.2010.00215" target="_blank"><em>attention</em></a>. There’s a deep link between the mechanics of sense-making of the kind implied by Helmholtz—modern-day versions of this would be the <a href="https://doi.org/10.1016/j.concog.2023.103510" target="_blank">Bayesian brain hypothesis</a>, for example—and attention, which is not just accumulating evidence from the world and assimilating it, but actually choosing what to ignore and what not to ignore in doing that. And that equips the brain with a kind of action-on-the-inside: you’re actually doing stuff to all the message-passing on the inside of your brain. And that kind of mental action gets you one step closer to the kind of conscious perception and awareness that we’re talking about. It’s not <a href="https://doi.org/10.3389/fpsyg.2018.00579" target="_blank">self-awareness</a>—we don’t know that we’re doing this at this stage—but there is an ability to act upon and render that which was transparent now opaque.</p><p>  <span class="article-content-box"> <a href="video/darwin-vs-consciousness" target="_blank" class="iai-related-in-article click_on_suggestion_link--gtm-track"> <span class="iai-card"> <span class="iai-card--image iai-related--video-play" style="display: block;"> <img src="/assets/Uploads/darwin-vs-consciousness.webp" class="iai-related--primary-image" alt="related-video-image"> </span> <span class="iai-card--content" style="display: block;"> <span class="iai-card--title" style="display: block;">SUGGESTED VIEWING</span> <span class="iai-card--heading" style="display: block;">Darwin vs consciousness</span> <span style="display: block;">With Güneş Taylor, Stuart Hameroff, Denis Noble, Antonella Tramacere</span> </span> </span> </a> </span> </p><p><strong>Now is a good time to introduce </strong><a href="https://doi.org/10.1038/nrn2787" target="_blank"><strong>the free energy principle</strong></a><strong>. Can you say what it is, and explain how it relates not only to perception but also to how the brain acts in the world?</strong></p><p>Right, so looking at what we’ve just said through the lens of physics and self-organizing systems, what we’re saying is that any system, including particularly the brain, is in the game of <em>minimizing its free energy</em>, where free energy is read as prediction error. So this is the most stable or the most likely way of being. Mathematically, it’s just called the principle of least action, where action is the average free energy or the average prediction error.</p><p>And that provides a more general perspective on <em>anything</em> that self-organizes. With respect to the brain, there is this bi-directional exchange between me and the world, in the sense that if I am wanting to minimize my prediction error or my free energy, then there are two ways that I can do that. Firstly, I can change my mind to make my predictions more like what I am sensing, and that resolves the prediction error. That would be perception. But there’s another way I can resolve prediction errors. That would be the active way, basically sampling a different kind of data, palpating the world with my eyes, with my fingers, in a different kind of way, to select and sample those data that I predicted. I change the world to make it conform with my predictions. That would not change the predictions, but it would minimize the prediction error.</p><p>So now you have a seamless, unitary explanation for both action and perception, both in the service of minimizing free energy. Once you look at it like that you suddenly realize you’re talking about a fundamental phenomenon that characterizes all self-organization. If I was a physiologist, it would just be homeostasis, it would just be the tendency of all self-maintaining systems to limit their prediction errors in the sense of restricting deviations from their homeostatic set points. You could even apply it to economics: you could say that things like expected utility theory is just an expression of rendering the world, making choices, acting upon the world in order to realize my predictions—usually optimistic predictions that I’m wealthy, loved and have a certain temperature.</p><p> </p><p><strong>Can you give us a sense of the variety of these self-organizing systems, from the very small to the very large?</strong></p><p>As soon as you can disambiguate anything from its environment—the self and self-organization from the non-self—you induce something called a <a href="http://iai.tv/articles/friston-vs-brogaard-down-the-rabbit-hole-part-1-auid-654" target="_blank">Markov Blanket</a>, a boundary that <a href="https://doi.org/10.1007/s11229-019-02370-y" target="_blank">separates me from everything that’s not me</a>. And that definitional approach to thingness implies free energy minimization.</p><p>In existing as separate from its environment—in existing in a way that it does not dissolve into the environment, so that it maintains the integrity of the boundary between it and everything else—a thing is engaging in a free-energy-minimizing process and therefore we can talk about it in terms of action and perception of a very elemental sort. So that would apply to anything. If you’ve got a name for it then it does this kind of thing.</p><p>But clearly that’s not a sufficient explanation for things like you and me. There are different kinds of things, and when we get to the level of things that are sufficiently large—large enough that quantum mechanics no longer apply but not so large that conservative dynamics and the motion of the heavenly bodies kick in—something really interesting happens. In the interface between me and not-me that mediates the perception and the action—the bidirectional traffic that does this entrainment and generalized synchronization—the active part becomes a cause of the sensation. And that’s really interesting because if my action is now a cause of my sensation, and if my brain is constantly inferring the causes of my sensations (as we think it is), then that means that my brain is inferring its own actions. And this is what means that we constitute a different sort of thing: the sort of thing that can <em>plan</em>.</p><p>So we now get this much deeper kind of action, where you have on the inside the consequences of your own actions. There’s now a separation between what I’m physically doing—all my autonomic reflexes that cause me to physically move—and what I think I am actually doing and what I am choosing to do and thereby predicting what I am going to do. So, I think planning would be definitive of the kinds of things that can possibly be called true agents.</p><p> </p><p><strong>Suppose I imagine something which I have no intention of making true, for example I imagine flying elephants. Haven’t I increased free energy in doing that, and how does the free energy principle, which says that things are always trying to minimize their free energy, accommodate that kind of case?</strong></p><p>That’s a lovely question. The free energy principle can be applied at different scales. What we’ve been talking about in terms of perception and active inference is an application at quite a fast time scale, several hundred milliseconds. But you could also apply it over days and weeks, in terms of learning. And you can also apply it at the level of neurodevelopment, of developmental psychology, and even at an evolutionary scale.</p><p>I think imagining flying elephants probably speaks to a timescale within which you could fit, say, dreaming. Why is it that we spend a good 20-40% of our life with our eyes closed, dreaming in a fantastical way, sequestered from the actual world? From the point of view of natural selection, you’d think this a very precarious policy to adopt.</p><p>However, if you look at the imperatives for minimizing free energy, then something really interesting happens. Minimizing free energy involves striving to model the world in a way that provides as accurate an account as possible <em>as simply as possible</em>.</p><p class="article-plus-content--header" style="text-align: center;">___</p><p class="article-plus-content--header" style="text-align: center;">There’s a vast range of psychiatric symptoms, ranging from psychosis to neglect symptoms (and you could include attention deficit disorder within that as well), that fall nicely under this notion of false inference.</p><p class="article-plus-content--header" style="text-align: center;">___</p><p>That means there’s an imperative to minimize complexity, to become wise and focused and sparse and non-redundant in the way that we understand our world. We do that by getting rid of all the redundant associations—all the synaptic connections, the wiring in our brains—that we don’t need. And we do this by trying to explain more simply some imaginary or fantasy sensorium or sensory input that we ourselves create. That is basically the argument for sleep, for dreaming, and one could also argue for many aspects of creative thought.</p><p>So I would contend that you flying the elephant is basically a way of just testing, “Do I need these constraints, these parameters, these contingencies, that are installed in my brain and my generative model to explain stuff or not?” And if you don’t need them, then as people like Giulio Tononi would say, you <a href="https://doi.org/10.1016/j.brainresbull.2003.09.004" target="_blank">engage in something called synaptic regression, a pruning</a>.</p><p>So the idea is that when we have our eyes open, we’re engaging with the world, accumulating all sorts of explanations and associations, and literally consolidating and reinforcing synaptic or wiring connections in the brain. And then we have a little break, and say, “Do we really need all that?” and we have a house-cleaning that minimizes the complexity.</p><p> </p><p><strong>You’ve applied the free energy principle to thinking about mental illness. Could you say why you find this such a promising application of the principle?</strong></p><p>Yes, and indeed much of this theoretical work was motivated by <a href="https://doi.org/10.1038/s41380-022-01743-z" target="_blank">trying to understand schizophrenia and psychosis</a>. A brief answer is that the free energy principle is basically a methodology that allows you to talk about sense-making in terms of inference. That allows us to see psychopathology as false inference, and we can start to talk about both the physiological and pharmacological processes that underwrite false inference, which is basically bad belief-updating.</p><p>That’s really useful because there are two ways that you can make a false inference. If you’re a statistician, these would be type 1 and type 2 errors. You can infer something is there when it’s not (a type 1 error, a positive error), and that’s a beautiful summary of hallucinations and delusions. And you can also have type 2 errors, when you infer that something’s not there when it is, and this would subsume what in the old days would have been called hysteria or dissociative symptoms or neglect symptoms. So there’s a vast range of psychiatric symptoms, ranging from psychosis to neglect symptoms (and you could include attention deficit disorder within that as well), that fall nicely under this notion of false inference.</p><p>So then you ask: what are the mechanisms that are consistent with an application of the free energy principle that could possibly explain this kind of false inference? And it all comes back to what we were talking about before, which is this ability to act on the inside by selecting the newsworthy information, the prediction errors, that drive my belief-updating. If you get that wrong, then you are susceptible to false inference. So you can look at this as a kind of mathematical <em>attention disorder</em>: a failure to ignore what is not relevant or informative. So, say I was watching television and I have this unignorable evidence from my body that is suggesting that there is some personal arousal that has been elicited by whatever’s happening, and I happen to be watching the newscaster on the evening news. To explain this unattenuated information, I might have the hypothesis—which would be completely Bayes-optimal—that the newscaster is talking to me personally. So you can see how easy it would be to fall into a delusional system or bring false explanations to the table if you can’t ignore the data at hand.</p>

Why emotion belongs to the body, not to "vibration"—a neuroscience-grounded reframing

MY METALLIC BRAIN makes it hard to remember anything. I wake up sometimes feeling like a cold wind is blowing through my head. I sense a dull ache in my teeth, too, which keeps me vigilant to pain as....

YouTube video by Nabaghan Ojha

This webinar was designed to help prospective applicants understand whether the AI Sentience Scholars program is the right fit for them. We introduce the goals and structure of the program, clarify…

Free masterclass: Explore near-death experiences with science, spirituality & lived insight—find peace about life, death, and what lies beyond.

Toward a Unified Language of Experience, Prediction, and Reality Rendering