IF I EXIST, I HAVE ADAPTED KARL FRISTON ON MIND AND THE FREE ENERGY PRINCIPLE By Karl Friston and Tony Sobrado *** The Montréal Review, June 2026 |
|||
Tony Sobrado: Why is there any kind of experience at all? In a world that's laden with physical objects, we have one physical object - the brain - that produces experience. So what? Why is this a problem? Many neuroscientists like Michael Graziano don't think it's a particularly exotic problem nor is consciousness enigmatic. Consciousness and experience is simply a model that the brain produces to understand itself. It is software running on hardware and that's why we have qualia and sensations. Daniel Dennett famously said there is no such thing as formal consciousness, it's an illusion. The Churchlands shared similar sentiments in terms of eliminating the mental. Essentially the hard problem of consciousness is why, in a physicalist world, is there any kind of subjective first person experience at all? The marriage of subjective experience and the physical brain is the problem. You are a world leader in the field of neuroscience and have thought about this issue prominently. However, before we get into the elements of the hard problem of consciousness specifically, as the world’s most cited neuroscientist, how do you think of, and define, consciousness itself? Karl Friston: I normally read everything—including consciousness—as a process of inference. One then asks what is special about consciousness? The answer will probably rest upon what kind of consciousness we want to talk about. Are we talking about experience? Are we talking about quality of experience? Are we talking about phenomenology? Are we talking about self-awareness? Are we talking about the loss of phenomenal transparency? I suspect we are talking about experience that is quintessentially subjective and therefore entails a subject being the owner of that experience. So, how would I think about that from the point of view of a physicist? A psychologist? A psychiatrist? A neuroscientist? Etc. Well, I would be asking what kind of natural things—like physical objects, as you put it—would be able to support forms of self awareness and experience that one would associate with qualitative experience? Tony Sobrado: Is the issue of experience and self awareness, and their constituent explanans, a philosophical or scientific one? And is it even answerable in any capacity? This is something we have discussed previously. So your paper The Hard Problem of Consciousness and the Free Energy Principle states that the Free Energy Principle answers the hard problem of consciousness in terms of experience from a scientific perspective (we will get onto this later). I don't think there's that much of a divide between a scientific perspective and a philosophical perspective in the way David Chalmers frames the hard problem of consciousness. Do you think answering why we have subjective experience in the first place is a philosophical or a scientific question or mixture of both? Karl Friston: I think it's a mixture of both. I'm not qualified to address the philosophical part because I have no foundational training in philosophy, but I certainly think it's a very interesting question that, in principle, can be naturalized under the natural sciences—that would include physics and psychology—when addressing why we have subjective experience. Tony Sobrado: A question that in principle can be gauged by the natural sciences primarily entails a scientific approach to answering why we have consciousness? And in this particular area it’s subjective experience. So give me an overview of the Free Energy Principle from your perspective, and how it serves to answer, perhaps in a scientific sense, why we have subjective experience? Karl Friston: I think the first thing to say is that the Free Energy Principle is not a theory and furthermore it is not a theory of consciousness. It is a principle in the sense of physics. This means it's a tool, a rule, a method, that may frame the space within which to constrain questions about consciousness. So, from my perspective, the Free Energy Principle is a principle of least action. It is akin to Hamilton's principle of least action that applies to certain kinds of things—like particles or people—that are defined by having characteristic states. From the point of physics, you can read this as attaining some kind of non-equilibrium steady state. Anthropologically, for humans, you can read that as a form of surviving. From an evolutionary perspective, you can read that as persisting but at the end of the day, mathematically, it’s the possession of characteristic states of being. So, once one works through the mathematical requirements—that would describe such systems—one can derive a mechanics, often called a Bayesian mechanics, that licenses a functionalist or teleological interpretation of the way that certain kinds of things behave. And that behavior can be described in terms of sense making and inference, given the fact that you and I as certain kinds of things are open to our environment. Essentially, there's a two-way traffic between us and our environment. Furthermore, this traffic can be expressed mathematically as minimizing variational free energy. The applications of the Free Energy Principle, however, are multiple and so too are the interpretations. My favorite is self-evidencing, which I borrow from Jacob Hohwy in the philosophy literature. Here, free energy represents a bound on something called model evidence. The model in question is our generative world model that we bring to the table to do our sense-making and to engage actively with the world. So, you can now describe behavior under the Free Energy Principle as essentially seeking evidence for ‘my world’ models. If I am an embodiment of ‘my world’ models, I am effectively seeking evidence for my own existence.
Robert Longo, Untitled (Freud's Desk by Window, 1938) (2004), from The Freud Cycle. Charcoal on mounted paper. The desk where the inner life was first mapped as a science — and the window that reminds us the mind is never sealed from the world it seeks to understand. Courtesy the artist and Pace Gallery.
That is the Free Energy Principle. The game — in applying the Free Energy Principle — is to demonstrate that many functionalist and teleological explanations are themselves emergent properties. This is because the Free Energy Principle turns explanation on its head: it doesn't say the reason for me doing X is so that I can survive, or that I can adapt, or that I am fit for purpose; instead, the Free Energy Principle says ‘If I exist then I have adapted...and I have this ability to maintain myself in characteristic states’. These then are the kind of behaviors I must exhibit to exist and adapt in characteristic states. You can then describe the characteristics themselves. Here, the Bayesian mechanics fosters many kinds of interpretations depending on whether you're a philosopher or a psychologist or a neuropharmacologist, etc. They all have their different ways of reading/interpreting the framework and putting a functionalist account of simply existing, in the sense of possessing characteristic states. Tony Sobrado: How does the Bayesian statistics translate? So we've got the principles in place, which is basically elements of biological evolution i.e. an organism must have these attributes to exist in its environment. How does the math behind the Free Energy Principle then dictate and arrive at the conclusion that this is why we have subjective experience? Is this even demonstrably possible? Karl Friston: I'm not sure that one would be able to demonstrate that ‘This is why we have subjective experience’ but you can certainly reduce the hard problem of consciousness to the generative model. Previously, I said that you can always describe the behavior of anything — that has characteristic states — in terms of seeking evidence for its world or generative model. That begs the question: What different kinds of generative models could be in play? What would be apt to describe this kind of thing or that kind of thing? And more specifically, what kinds of generative models would you need in order to infer that somebody has subjective experience, and indeed, that somebody could be me or you? So, the question then reduces to what are the special characteristics of generative models associated with subjective experience? There are a number of characteristics which I will now briefly cover:
These are a few examples of applying the Free Energy Principle — and specifically in this instance, active inference — that emphasizes the notion of action, the consequences of overt action upon the world but also covert actions, such as attending to particular sensations. Then, we have framed the question in terms of what things could be described as having subjective experiences and what things could not? And here there are clear divides. For example, you would not expect an insect to have this kind of consciousness simply because the math requires there to be a disconnect between the internal machinations of the thing in question and its interface or action upon the world. As soon as that happens you are licensed to describe this or that thing as effectively inferring its own actions. This means you are looking at things beyond thermostats, viruses and small insects. Instead, you are looking at larger systems, where deep inside these systems there is the latitude to understand their internal dynamics as modelling their own actions. Tony Sobrado: I think the reference point and example you make in terms of looking at the Free Energy Principle and, where applicable, the world/generative models is excellent: That there is obviously a clear difference between why Homo Sapiens would need a subjective inner world for agency and planning versus an insect. Ultimately, is this fundamentally just an elaborate functionalist account, just with more mathematical detail? Karl Friston: Yes! But I want to qualify that the Free Energy Principle can be applied in many different contexts. It can be applied to computational phenotyping, neurology, psychiatry, machine learning, cell cultures, etc. And you don't need to tell a functionalist story in its application. The Free Energy Principle in and of itself is quite agnostic to functionalist or teleological interpretations. When it comes to subjective experience, therefore, there's only a need for subjective experience in a particular context and it may not just be the nature of the thing — as determined by its generative model — but the relation of that thing to the world. It is about modelling and specifically the fact that it must model other things that are very similar to itself. For example, there are other things like me and you, and this induces an attribution problem. And so, to be a subject means that I have made an inference ‘I did that’ as opposed to ‘you did that’: I have solved an attribution and agency problem. To do that, however, I need the basic concept that there is a self and there are others. In the absence of others, however, there will be no need for subjective experience itself. Tony Sobrado: So I probed about functionalism and the Free Energy Principle being a functionalist paradigm - just with more mathematical detail. What you are saying, more specifically, however is that because the Free Energy Principle can be derived from physics and because maxims in physics are about physical properties of the universe like particles, mass and charge etc. this means it does not logically necessitate functionalist properties per se because functionalist elements come in later, particularly at the stage of evolutionary biology. That is an important nuance that needs noting! In a certain respect the application of the Free Energy Principle can answer the why question behind the hard problem of consciousness - why we have consciousness and subjective experience. The how question behind the hard problem of consciousness is more challenging and often involves, and requires, physical brain mechanisms that can explain the element of experience. David Chalmers, in a formal sense, wants a chemical equation or mathematical model akin to how and why H₂O = Water. This also relates to the Explanatory Gap articulated by Joseph Levine: Why is it that neuronal state X = consciousness? Chalmers’ position encompasses the theoretical, conceptual and explanatory problems of reduction and ultimately he argues that the scientific reduction of subjective experience to physical properties is impossible by definition. In terms of functionalism and answering the why question behind the hard problem of consciousness via the application of the Free Energy Principle, this might still be insufficient. Thinkers like Philip Goff, David Chalmers and Thomas Nagel, to name a few, are anti-functionalists. They argue that if you can explain consciousness as being something that's functional, then evolution could have selected other functional mechanisms that do not possess sentient experience but still do the same job. Contemporarily, one only has to think about AI to make their point. Here, however, the retort can be levied that evolution randomly selects natural functions that serve species’ fitness so yes it could have selected other functions but sentient experience was selected because evolution relies on random processes to generate variation. Functionalism is, nonetheless from my perspective, a powerful explanation as to why we have subjective experience. I noticed in your paper that you discussed dual‑aspect monism which bears a similarity to property dualism, and you also touched on illusionist ideas. My question for you is this: Do you think the mental and the conscious form their own ontological realm? Or do you think they are purely physical – thus the mental has no separate domain at all? Karl Friston: The key idea is modeling: as soon as you have an internal model—your brain’s representation of the world—you have neuronal dynamics on the inside that mathematically corresponds to things on the outside. This creates an inherently relational structure. You infer external causes you can’t directly access, so there is no absolute truth—only the best explanation for your sensory data. Mathematically, this lets you describe two kinds of information geometries, based on the same underlying dynamics. That’s the mathematical form of monism you mentioned: there is just one set of internal physical processes—such as neural dynamics—but they can be expressed in two complementary ways. For now, however, let's just keep it to the neural dynamics. There are neural dynamics that, metaphysically speaking, form the basis of everything we’re talking about. These internal neuronal processes have an information geometry that can be read mathematically as the thermodynamics of the brain—the energy use, dynamical stability, and physical behavior of neural activity. Researchers like Gavin Crooks have shown beautifully how these thermodynamic descriptions map onto information processing in population activity and cell firing. At the same time, the very same trajectory through neuronal state-space can be interpreted differently: each point in that space can be read as encoding a belief about the external world. This creates a second information geometry—one that describes belief updating. Every physical brain state corresponds to a probability distribution over the hidden causes of sensations. This gives rise to a special kind of space called a statistical manifold. Moving through this manifold literally corresponds to updating beliefs in response to sensory evidence. So, mathematically, there is room to describe the same neural dynamics in two complementary ways:
Both descriptions are grounded in the same physics. This dual reading isn’t something that fits neatly into traditional philosophical debates—say, with someone like David Chalmers—but many philosophically informed physicists appreciate that one set of dynamics can have two complementary interpretations. This is part of the utility of the Free Energy Principle: it provides a mechanics of beliefs, where “beliefs” are understood mathematically as probability distributions with attributes of uncertainty and confidence. These distributions are parameterized by the physical states of the brain—the same states we measure with EEG, brain imaging, or observe through neurological interventions. Tony Sobrado: Given your earlier remarks, I would like to pose the question in a more precise way: Do you hold that subjective experience is caused by the physical brain, or do you take subjective experience to be identical to the brain’s physical processes? Since your account assumes a single underlying physical base—one that can be expressed mathematically and that supports the notion of mathematical beliefs—the issue becomes whether consciousness is a causal product of neural activity or whether it is simply identical to those neural dynamics. Karl Friston: If I had to choose—and you’re clearly making me choose—I would say that subjective experience is identical to the physical brain’s activity. In this view, the only real notion of “cause” comes from physics itself. Under the Free Energy Principle, causation refers to the lawful way a system’s current state determines how it changes over time. In physics, this is expressed through something like a Langevin equation, where the rate of change of a state is a lawful function of its current position in state space. This gives you a generic random dynamical system. From this starting point, you can derive Bayesian mechanics, classical mechanics, or even quantum mechanics. The only true causality here is the way the present state of the universe determines its future trajectory. Because of this, you can’t say that “belief updating causes experience” or “experience causes belief updating.” These are just different functionalist descriptions of the same underlying dynamics. The real cause–effect structure is simply the evolution of the system through time. So, the Free Energy Principle allows you to describe the same physical process in two complementary ways:
Both descriptions, however, refer to the same underlying physical process, not two separate causal layers.
*** MORE CONVERSATIONS
The Montréal Review, February 2026 The Montréal Review, February 2026 The Montréal Review, October 2025 The Montréal Review, June 2025 The Montréal Review, April 2025 The Montréal Review, January 2025 The Montréal Review, July 2024 |
Robert Longo, View of Study Room with Books, Desk and Window, 1938 (2002), from The Freud Cycle. Charcoal on mounted paper. The room where the inner life became a science — rendered in the same medium that conceals as it reveals. Courtesy the artist and Pace Gallery.
