What do you think?
Rate this book
Active Inference: The Free Energy Principle in Mind, Brain, and Behavior
The first comprehensive treatment of active inference, an integrative perspective on brain, cognition, and behavior used across multiple disciplines.
Active inference is a way of understanding sentient behavior—a theory that characterizes perception, planning, and action in terms of probabilistic inference. Developed by theoretical neuroscientist Karl Friston over years of groundbreaking research, active inference provides an integrated perspective on brain, cognition, and behavior that is increasingly used across multiple disciplines including neuroscience, psychology, and philosophy. Active inference puts the action into perception. This book offers the first comprehensive treatment of active inference, covering theory, applications, and cognitive domains.
Active inference is a “first principles” approach to understanding behavior and the brain, framed in terms of a single imperative to minimize free energy. The book emphasizes the implications of the free energy principle for understanding how the brain works. It first introduces active inference both conceptually and formally, contextualizing it within current theories of cognition. It then provides specific examples of computational models that use active inference to explain such cognitive phenomena as perception, attention, memory, and planning.
Active inference is a way of understanding sentient behavior—a theory that characterizes perception, planning, and action in terms of probabilistic inference. Developed by theoretical neuroscientist Karl Friston over years of groundbreaking research, active inference provides an integrated perspective on brain, cognition, and behavior that is increasingly used across multiple disciplines including neuroscience, psychology, and philosophy. Active inference puts the action into perception. This book offers the first comprehensive treatment of active inference, covering theory, applications, and cognitive domains.
Active inference is a “first principles” approach to understanding behavior and the brain, framed in terms of a single imperative to minimize free energy. The book emphasizes the implications of the free energy principle for understanding how the brain works. It first introduces active inference both conceptually and formally, contextualizing it within current theories of cognition. It then provides specific examples of computational models that use active inference to explain such cognitive phenomena as perception, attention, memory, and planning.
312 pages, Hardcover
Published March 29, 2022
157 people are currently reading
1015 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 16 of 16 reviews
Read
October 18, 2025Could we solve the age-old and vexing mind-body problem with the help of mathematics, information theory and statistical physics? Perhaps I'm just hallucinating ...
Helmholtz already suspected this in the 19th century: we do not simply perceive the world through our senses, we try to predict it. To put it simple, we bet on the most probable scenario in our conception of the world. If our bet fails, we either correct our conception or adapt the world to our conception. This is the path of least effort or minimization of entropy.
Active inference (or predictive processing) is the conceptional framework for this process. It is based on stochastics, information theory and the free energy principle, a term from statistical physics. It represents not only a fundamental turn in our understanding of perception and action, but also in the mechanisms of survival and adaptation.
The most obvious applications are in neuroscience, e.g. it sheds new light on psychosomatic illnesses or consciousness training such as meditation, it bridges the gap between neurology and psychiatry, and opens up a new perspective on mental disorders and neurodiversity. In AI development, models are already being developed that use active inference or related approaches and go far beyond reinforcement learning.
But beware! This book deals with the mathematical and theoretical foundations of active inference and the free energy principle. It has stretched my understanding to its limits. Authors such as Anil Seth, Andy Clark, Lisa Feldman Barrett, Jakob Hohwy and Thomas Metzinger have written more accessible books and also go into greater detail on the philosophical and practical implications.
See also:
Being You: A New Science of Consciousness (Anil Seth)
Surfing Uncertainty: Prediction, Action, and the Embodied Mind (Andy Clark)
The Philosophy and Science of Predictive Processing (Dr Dina Mendonca, editor)
Helmholtz already suspected this in the 19th century: we do not simply perceive the world through our senses, we try to predict it. To put it simple, we bet on the most probable scenario in our conception of the world. If our bet fails, we either correct our conception or adapt the world to our conception. This is the path of least effort or minimization of entropy.
Active inference (or predictive processing) is the conceptional framework for this process. It is based on stochastics, information theory and the free energy principle, a term from statistical physics. It represents not only a fundamental turn in our understanding of perception and action, but also in the mechanisms of survival and adaptation.
The most obvious applications are in neuroscience, e.g. it sheds new light on psychosomatic illnesses or consciousness training such as meditation, it bridges the gap between neurology and psychiatry, and opens up a new perspective on mental disorders and neurodiversity. In AI development, models are already being developed that use active inference or related approaches and go far beyond reinforcement learning.
But beware! This book deals with the mathematical and theoretical foundations of active inference and the free energy principle. It has stretched my understanding to its limits. Authors such as Anil Seth, Andy Clark, Lisa Feldman Barrett, Jakob Hohwy and Thomas Metzinger have written more accessible books and also go into greater detail on the philosophical and practical implications.
See also:
Being You: A New Science of Consciousness (Anil Seth)
Surfing Uncertainty: Prediction, Action, and the Embodied Mind (Andy Clark)
The Philosophy and Science of Predictive Processing (Dr Dina Mendonca, editor)
December 30, 2022
Belief-based computational phenotyping holds promise in the emerging fields of computational psychiatry, neuropsychology, and neurology.
These methods can be used to simulate not only pathology but also the influence of therapeutics, e.g. it should be possible to simulate the consequences of pharmacological manipulations.
These methods can be used to simulate not only pathology but also the influence of therapeutics, e.g. it should be possible to simulate the consequences of pharmacological manipulations.
August 1, 2024
I was delighted to read at the beginning that Friston had stopped himself from writing much of the book so that it would be clear and concise. Even then, I could understand only about 25 per cent of it and ended up crying a lot into my Markov Blanket.
The beginning and ending chapters provide a good overview of Active Inference, Bayesian Brain, Predictive Coding, and related topics.
The middle part of the book is dense and specialist. Thankfully, there is a valuable appendix to help brush up the reader's algebra, and the start of the book gives a good reminder of Bayes Theorem and conditional probability.
I recommend people interested in the topic also read Anil Seth's Being You: A New Science of Consciousness and Andy Clark’s The Experience Machine. Both books give a good overview of mind, brain, and behaviour, but are targeted at a broader readership, so sacrifice the maths.
Seth introduced me to Karl Friston’s work, but even his overview was beyond me, and I had to read it a few times. I think the ideas and implications in Active Inference are appealing, and this book gave so much more insight than Clark’s, but it is really asking a lot for a layman like me to absorb the pertinent details of this work.
In some ways, the whole area of Free Energy minimization reminds me of Hinton's Back Propagation, the efficient algorithm that makes deep learning neural networks work. So, the book will appear to computational neuroscientists and artificial intelligence practitioners. The book's first part is about the theory, and the second part is about how to implement the ideas. There is even a reference to some computational formula, but it appears to be more oriented towards Matlab than Python, which will likely cause some issues for the hobbyist.
I will re-read it and try to understand in greater detail how the mathematics is mapped to the cortical layered structure of the human brain, which sounded very plausible.
I gave it four stars.
The beginning and ending chapters provide a good overview of Active Inference, Bayesian Brain, Predictive Coding, and related topics.
The middle part of the book is dense and specialist. Thankfully, there is a valuable appendix to help brush up the reader's algebra, and the start of the book gives a good reminder of Bayes Theorem and conditional probability.
I recommend people interested in the topic also read Anil Seth's Being You: A New Science of Consciousness and Andy Clark’s The Experience Machine. Both books give a good overview of mind, brain, and behaviour, but are targeted at a broader readership, so sacrifice the maths.
Seth introduced me to Karl Friston’s work, but even his overview was beyond me, and I had to read it a few times. I think the ideas and implications in Active Inference are appealing, and this book gave so much more insight than Clark’s, but it is really asking a lot for a layman like me to absorb the pertinent details of this work.
In some ways, the whole area of Free Energy minimization reminds me of Hinton's Back Propagation, the efficient algorithm that makes deep learning neural networks work. So, the book will appear to computational neuroscientists and artificial intelligence practitioners. The book's first part is about the theory, and the second part is about how to implement the ideas. There is even a reference to some computational formula, but it appears to be more oriented towards Matlab than Python, which will likely cause some issues for the hobbyist.
I will re-read it and try to understand in greater detail how the mathematics is mapped to the cortical layered structure of the human brain, which sounded very plausible.
I gave it four stars.
April 12, 2022
Dense overview of active inference, primarily focused on the brain. In classic Friston style, lots of maths.
Still waiting for a clearer explanation of Active Inference applied to higher level sociotechnical systems, and for a simple modeling software to really grok it.
Still waiting for a clearer explanation of Active Inference applied to higher level sociotechnical systems, and for a simple modeling software to really grok it.
June 8, 2024
Intruiguing… Refreshing perspectives on free energy and entropy….
January 10, 2025
Me dumb. Book wordy. Theory good.
January 20, 2024
This book easily rates five stars for the effort and care the authors put into its preparation, while the experience it provides in a single reading is closer to three stars. That latter rating is not necessarily a fault of the book but rather a consequence of the density of the material and its usefulness to someone who might choose to read it more out of curiosity than out of an immediate need to put its ideas into practice. This is essentially a textbook that does not lend itself to a quick cover-to-cover reading. In fact, I offer as a caveat for my review that it is informed by a thorough reading of the first half of the book and a variable-speed skimming of the second half.
Active inference and the free-energy principle have been on my radar for some time, and I have read some (but certainly not all) of Karl Friston's original papers on the topic, roughly a gazillion of which are cited in this book. Friston is listed as an author of this volume but half-jokingly admits in the preface that he was not allowed to write most of it, saying that the book's aims call for a "crisp and clear writing style that is beyond me." I admit that this line alone got me excited about the book.
The free-energy principle, despite being very mathematical, is not terribly complicated. It essentially says that your brain is always guessing what it will perceive next and tries to be right as much as possible. It's like playing a guessing game and getting better at guessing by improving your model of the world. But your brain also does things to make its guesses right through its choice of actions. The core math behind the idea has been around for ages and independently plays a major role in Bayesian inference, machine learning, and even thermodynamics. What's new is that Friston's application of this concept to the brain creates a unified view of perception and action, which are typically grouped together conceptually but are often modeled separately. He argues that perception and action work together to minimize our surprise (prediction error) at the world.
The book makes connections to the other areas where the free-energy principle applies and also collects suggestions and evidence for how the principle might be instantiated in the brain. It is partly due to these various cross-disciplinary connections and biological details that things begin to get a bit overwhelming. There is simply too much information to keep straight, so unless you go in with a specific agenda in mind (e.g. "I want to create an active inference model of Process X"), the best you might hope for on a first reading is to get the gist and a mental bookmark to come back to the book should you ever need the details.
There are some good tips and recipes for researchers who wish to code up some active inference models, but since I have no such intention at the moment, I passed over these sections fairly quickly. The appendices also appear helpful and are another indication that the authors hoped to make this book as self-contained as possible. My impression is that this is an excellent reference that I will want to come back to again ... after my brain digests what made for a pretty heavy meal.
Active inference and the free-energy principle have been on my radar for some time, and I have read some (but certainly not all) of Karl Friston's original papers on the topic, roughly a gazillion of which are cited in this book. Friston is listed as an author of this volume but half-jokingly admits in the preface that he was not allowed to write most of it, saying that the book's aims call for a "crisp and clear writing style that is beyond me." I admit that this line alone got me excited about the book.
The free-energy principle, despite being very mathematical, is not terribly complicated. It essentially says that your brain is always guessing what it will perceive next and tries to be right as much as possible. It's like playing a guessing game and getting better at guessing by improving your model of the world. But your brain also does things to make its guesses right through its choice of actions. The core math behind the idea has been around for ages and independently plays a major role in Bayesian inference, machine learning, and even thermodynamics. What's new is that Friston's application of this concept to the brain creates a unified view of perception and action, which are typically grouped together conceptually but are often modeled separately. He argues that perception and action work together to minimize our surprise (prediction error) at the world.
The book makes connections to the other areas where the free-energy principle applies and also collects suggestions and evidence for how the principle might be instantiated in the brain. It is partly due to these various cross-disciplinary connections and biological details that things begin to get a bit overwhelming. There is simply too much information to keep straight, so unless you go in with a specific agenda in mind (e.g. "I want to create an active inference model of Process X"), the best you might hope for on a first reading is to get the gist and a mental bookmark to come back to the book should you ever need the details.
There are some good tips and recipes for researchers who wish to code up some active inference models, but since I have no such intention at the moment, I passed over these sections fairly quickly. The appendices also appear helpful and are another indication that the authors hoped to make this book as self-contained as possible. My impression is that this is an excellent reference that I will want to come back to again ... after my brain digests what made for a pretty heavy meal.
November 23, 2023
Evaluating the writing, not necessarily the ideas. The core thesis is an interesting one: organisms aren't simply passive receivers of perceptual information which they use to form models of understanding. Rather, they engage in active inference, they hold prior beliefs and seek out either confirming/disconfirming information, update their beliefs, or reconcile them in other ways (e.g. by choosing a different environment). The principle behind action is the minimization of Free Energy, which is overly simplified as a measure of surprise. And this can be applied to a whole host of fields, neuroscience, behavioral psychology, AI, with specific and predictive mathematical formulas.
I'm not the audience for this book and I'm not certain who is. It is not clearly written - at least 50% of it is over my head. It's highly technical, both in language and in content, so reads like a difficult college textbook. It's the type of book that has a Matlab appendix. Very difficult to get through.
Even conceptually, I think the explanation is lacking for someone like me who's not already active in the field. There are some big jumps in the equations for free energy, equating several different things that to me feel like leaps but aren't justified in explanation. There is a particular jump that's remarked as being such - intention is represented as a prior. Meaning there is no difference (in the equations) between desire and knowledge. The book explains why this is useful, but it's making greater claims and could have used more explanation as to why it's true.
There are also 2 practical quibbles I had that made the book literally difficult to read - the book uses a tiny font with a huge page margin. Why? And the actual print is faint, as if it was printed in ink-saving mode. Tiny, faint print is not reader-friendly.
My colleague recommended this to me as being brilliant. I'll take their word for it because I could not understand much of it.
I'm not the audience for this book and I'm not certain who is. It is not clearly written - at least 50% of it is over my head. It's highly technical, both in language and in content, so reads like a difficult college textbook. It's the type of book that has a Matlab appendix. Very difficult to get through.
Even conceptually, I think the explanation is lacking for someone like me who's not already active in the field. There are some big jumps in the equations for free energy, equating several different things that to me feel like leaps but aren't justified in explanation. There is a particular jump that's remarked as being such - intention is represented as a prior. Meaning there is no difference (in the equations) between desire and knowledge. The book explains why this is useful, but it's making greater claims and could have used more explanation as to why it's true.
There are also 2 practical quibbles I had that made the book literally difficult to read - the book uses a tiny font with a huge page margin. Why? And the actual print is faint, as if it was printed in ink-saving mode. Tiny, faint print is not reader-friendly.
My colleague recommended this to me as being brilliant. I'll take their word for it because I could not understand much of it.
May 25, 2025
« Surprise has to be interpreted in a technical sense: it measures how much an agent’s current sensory observation differ from its preferred sensory observations − that is, those that preserve its integrity (e.g, for a fish, being in the water). Importantly, minimizing surprise is not something that can be done by passively observing the environment: rather, agents must adaptively control their action-perception loops to solicit desired sensory observations. »
« Homeostatic regulation can be achieved in Active Inference by specifying the viable ranges of physiological parameters as priors over interoceptive perception. »
« On this view, it has been proposed that emotional valence, or the positive or negative character of emotions, can be conceived as the rate of change (first time-derivative) of free energy over time (Joffily and Coricelli 2013). Specifically, when a creature experiences an increase in its free energy over time, it may assign a negative valence to the situation ; whereas when it experience a decrease of its free energy over time, it may assign it a positive valence.
Good explanation of the theory.
Time will tell if sentient being actually works this way.
« Homeostatic regulation can be achieved in Active Inference by specifying the viable ranges of physiological parameters as priors over interoceptive perception. »
« On this view, it has been proposed that emotional valence, or the positive or negative character of emotions, can be conceived as the rate of change (first time-derivative) of free energy over time (Joffily and Coricelli 2013). Specifically, when a creature experiences an increase in its free energy over time, it may assign a negative valence to the situation ; whereas when it experience a decrease of its free energy over time, it may assign it a positive valence.
Good explanation of the theory.
Time will tell if sentient being actually works this way.
February 13, 2024
This is the book that finally got me to sit down and figure out what active inference/the Free Energy Principle is all about. The equations are there (with the appendices filling in the gaps), but there is a huge focus on interpreting the equations and discussing what they imply about perception and behavior. Overall it is very readable, but I got more from some parts than from others. There is a tendency to describe the implications of the theory somewhat vaguely or tautologically, such as by identifying certain words with terms of an equation and then just describing what the equation implies using those words (i.e., without justifying the correspondence or contrasting with alternatives). This isn't always done, but the parts where it is (often with some repetition) feel more like filler. That's what brings it down to 4 stars for me, but again overall immensely helpful for grokking this theory.
September 20, 2025
This book introduces a few key concepts in variational inference, Bayesian thinking, and statistical physics. It’s a good survey of interesting ideas and the framework itself is very compelling theory for the brain. Active inference is about minimizing surprise and free energy principle.
For example, perception minimizes {Bayesian surprise, prediction error, D_KL} and actions minimize {surprise, info gain, entropy}.
However it delves into too much complexity and doesn’t prove from first principles (I feel choosing one field and deriving from axioms is better than multiple paradigms).
For example, perception minimizes {Bayesian surprise, prediction error, D_KL} and actions minimize {surprise, info gain, entropy}.
However it delves into too much complexity and doesn’t prove from first principles (I feel choosing one field and deriving from axioms is better than multiple paradigms).
May 4, 2025
There is exactly one novel idea in this book (expected free energy) and the authors try to smuggle it in unmotivated without drawing the reader’s attention to it. I’m reminded of the sophistical garbage pundits do with accounting identities (C+I+G+NX, MV=PQ)
November 23, 2025
On the positive side, this is a good overview of variational inference and Bayesian statistics as applied to brain science. The principle of this statistical approach, which aims to minimize surprise by both learning and acting , is clearly explained (for those who can interpret mathematical formulas). On the negative side, the graphs and figures are mostly of poor quality and the writing style is sometimes convoluted to the point that it distracts from the primary claims.
July 2, 2025
Finally, an accessible introduction to FEP. Very readable and makes it (nearly) obvious. Need to process my notes.
August 1, 2025
textbook-group.activeinference.institute
March 12, 2023
This book almost qualifies as a textbook, since it is fairly technical, but anyway, the idea of active inference is brilliantly creative. A dichotomy between inference and control disappears, and the exploration/exploitation trade-off is more-or-less resolved. I'm not sure why I've never heard of this approach before.
Displaying 1 - 16 of 16 reviews