The Experience Machine: How Our Minds Predict and Shape Reality

by Andy Clark

A grand new vision of cognitive science that explains how our minds build our worldsFor as long as we've studied the mind, we've believed that information flowing from our senses determines what our mind perceives. But as our understanding has advanced in the last few decades, a hugely powerful new view has flipped this assumption on its head. The brain is not a passive receiver, but an ever-active predictor.At the forefront of this cognitive revolution is widely acclaimed philosopher and cognitive scientist Andy Clark, who has synthesized his ground-breaking work on the predictive brain to explore its fascinating mechanics and implications. Among the most stunning of these is the realization that experience itself, because it is guided by prior expectation, is a kind of controlled hallucination. We don't passively take in the world around us; instead our mind is constantly making and refining predictions about what we expect to see. This even applies to our bodies, as the way we experience pain and other states is shaped by our expectations, and this has broader implications for the understanding and treatment of conditions from PTSD to schizophrenia to medically unexplained symptoms. From the most mundane experiences to the most sublime, it is our predictions that sculpt our experience.A landmark study of cognitive science, The Experience Machine lays out the extraordinary explanatory power of the predictive brain for our lives, mental health and society.

Genres: Psychology, Nonfiction, Philosophy, Science, Neuroscience, Audiobook, Artificial Intelligence

320 pages, Hardcover

First published March 2, 2023

About the author

Librarian Note: There is more than one author in the GoodReads database with this name. See this thread for more information.

Reviews

[Nick Traynor · Rating 3 out of 5]

If you are not familiar with the theory of predictive processing, an idea which goes back to Hermann von Helmholtz in 1860 but which has been revitalised by Karl Friston in the 21st century, then this book will give you a good introduction. It is a compelling framework for understanding human cognition and its central tenet of the perception-action cycle bootstraps the ideas of 1950s cybernetics into modern neuroscience. Andy Clark adds his own perspective to this with the concept of the extended mind, that he developed in the 1990s with philosopher David Chalmers, and which is an application of Richard Dawkins's idea of the extended phenotype to philosophy of mind. There is not a lot of detail here though; for that you are better off reading the books on predictive processing by Anil Seth, Lisa Feldman Barrett or Mark Solms, and the tired references to visual illusions and well-known psychological experiments were a bit boring.

[Nelson Zagalo · Rating 2 out of 5]

There's nothing new here. On the other hand, writing a whole book just about the theory of predictive models of perception is not very interesting, reminding me of another book doing a bit of the same "A Thousand Brains: A New Theory of Intelligence" (2021) by Jeff Hawkins

If you're interested in the theory, I recommend reading the much richer “Being You: A New Science of Consciousness” (2021) by Anil Seth.

[Cav · Rating 4 out of 5]

"What is your relationship to the reality you perceive? In what ways do you shape it, and, by extension, in what ways do you shape yourself, often without even knowing it?"

The Experience Machine was a very interesting examination of the way we all see the world, and a look into a different fundamental understanding of how our brains operate.

Author Andy Clark is a British philosopher who is Professor of Cognitive Philosophy at the University of Sussex. Clark is one of the founding members of the CONTACT collaborative research project whose aim is to investigate the role environment plays in shaping the nature of conscious experience.

Andy Clark:


The topic of this book is incredibly interesting. The thesis the author lays out is one of the brain being a prediction-based machine. This new line of thinking turns traditional scientific orthodoxy on its head. It has previously been assumed that consciousness worked the other way around. That is; the brain gathers sensory information, and then generates a picture based on that info. The writing here forwards the case that this is not so. We see the world based on our expectations of the data coming into our sensory organs.

Clark opens the book with a decent intro. He mentions phantom phone buzzing syndrome, something that we all have likely experienced before.

He's got a good writing style (for the most part) that shouldn't have trouble holding the finicky reader's attention. There is a lot of super-interesting content covered here.
I also found the formatting to be well done. The writing is broken into broad-based chapters, and each chapter, into segmented writing with relevant headers at the top.

Clark expands upon the thesis of the book a bit more in this quote:

"Whereas sensory information was often considered to be the starting point of experience, the emerging science of the predictive brain suggests a rather different role. Now, the current sensory signal is used to refine and correct the process of informed guessing (the attempts at prediction) already taking place. It is now the predictions that do much of the heavy lifting. According to this new picture, experience—of the world, ourselves, and even our own bodies—is never a simple reflection of external or internal facts. Instead, all human experience arises at the meeting point of informed predictions and sensory stimulations.
This is a profound change in our understanding of the mind that fundamentally alters how we should think about perception and the construction of human reality. For much of human history, scientists and philosophers saw perception as a process that worked mostly “from the outside in,” as light, sound, touch, and chemical odors activate receptors in eyes, ears, nose, and skin, progressively being refined into a richer picture of the wider world. Even well into the twenty-first century, leading models in both neuroscience and artificial intelligence retained core elements of that view.
The new science of predictive processing flips that traditional story on its head. Perception is now heavily shaped from the opposite direction, as predictions formed deep in the brain reach down to alter responses all the way down to areas closer to the skin, eyes, nose, and ears—the sensory organs that take in signals from the outside world. Incoming sensory signals help correct errors in prediction, but the predictions are in the driver’s seat now. This means that what we perceive today is deeply rooted in what we experienced yesterday, and all the days before that. Every aspect of our daily experience comes to us filtered by hidden webs of prediction—the brain’s best expectations rooted in our own past histories."

Anecdotally speaking, I have been thinking a lot about the mind-body connection, the power of expectations, and environmental feedback in the last few years. The fields of high performance and excellence have already tapped into this modality, and modern science is only now starting to catch up.

He continues the quote from the start of this review here, outlining what the book will talk about:

"...In this book, I draw on paradigm-shifting research to confront these crucial questions and ask what these insights mean for neuroscience, psychology, psychiatry, medicine, and how we live our lives. We’ll look hard at experiences of the body and self, from chronic pain to psychosis, and see how work on the predictive brain helps explain a wide spectrum of human behaviors and neurodiversity. We’ll reassess our own experiences of the world, from social anxiety and emotional feedback loops to the many forms of bias that can creep into our judgments. We’ll also explore some ways that predictive brains might support “extended minds,” blurring the boundaries between ourselves and our best-fitted tools and environments."

Some more of what is covered here includes:
• Functional neurological disorders
• Schizophrenia
• Hard and soft problems of conciseness
• Dutch microscopist Antonie van Leeuwenhoek and homunculus theory
• The gut-brain connection via the vagus nerve
• Placebo and nocebo effects
• Psychedelic drugs
• Meditation and mindfulness

The author underlines the importance of this new research, and the ramifications of a more widespread acceptance and understanding of this paradigm-shifting view:

"This new understanding of the process of perceiving has real importance for our lives. It alters how we should think about the evidence of our own senses. It impacts how we should think about the way we experience our own bodily states—of pain, hunger, and other experiences such as feeling anxious or depressed. For the way our bodily states feel to us likewise reflects a complex mixture of what our brains predict and what the current bodily signals suggest. This means that we can, at times, change how we feel by changing what we (consciously or unconsciously) predict.
This does not mean we can simply “predict ourselves better,” nor does it mean we can alter our own experiences of pain or hunger in any way we choose. But it does suggest some principled and perhaps unexpected wiggle room—room that, with care and training, we might turn to our advantage.
Handled carefully, a better appreciation of the power of prediction could improve the way we think about our own medical symptoms and suggest new ways of understanding mental health, mental illness, and neurodiversity."

Although most of the content here was super-interesting, there was a good chunk of chapter 5 that had some questionable writing about police bias against black people and shootings. The author chastises police for their supposed racial bias against black men, mentioning the shooting of them by police. It never seems to occur to him that these biases could have possibly originated from their real-world experiences in their line of work. He mentions how a "disproportionate" number of black people get shot versus other races, but doesn't mention (or know) that blacks commit a disproportionate amount of crime (especially violent crime) compared to other races. And, if a group of people commit more violent crime, that group will likely be having more violent interactions with the police.

I can't help but notice the glaring irony in the author of a book that describes the brain as a prediction-based, pattern-recognizing machine being seemingly oblivious to the fact that a lot of these interactions are prediction-based, utilizing real-world pattern recognition.
This kind of mindless nonsense is becoming more and more common in science books, and it typically goes unchallenged out of people's fear of being called racist.

So, we have an out-of-touch bleeding-heart, ivory-towered intellectual that obviously has zero understanding of violence and its escalation, who's never held a gun in his life, passing judgment on, and chastising thousands of people whose job it is to keep our society safe while dealing with violent career criminals. Generalizing the police as no more than blood-thirsty racist murders. What complete nonsense. Stay in your lane, Clark...

Also, I found the latter part of the book dragged a bit. Some of the writing here was a bit more long-winded than it could have been. Now, I am admittedly very particular about how lively and engaging my books are, so my reviews are always heavily weighted toward this criterion.

***********************

Despite my above criticisms, I still enjoyed The Experience Machine. The complex understanding of human conciseness and how the brain and body interact are still in their infancy. I look forward to reading more about this emergent field in the future.
3.5 stars.

[India M. Clamp · Rating 3 out of 5]

3.1.24 RIP

“This book is about those balances and an emerging science that turns much of what we thought we knew about perceiving our worlds upside down.”
—Andy Clark

[Morgan Blackledge · Rating 5 out of 5]

REALLY GOOD BOOK.

Can’t review it right now.

I’ll try to make time to write about it soon.

In the meantime.

5/5

[Hugh · Rating 4 out of 5]

The subject of this book is all new to me, and I come at these concepts with near-total ignorance.

I honestly felt like my mind was expanding with each page flip in the first half of the book. Clark uses simple and surprising examples to support or illustrate concepts, and I legitimately learned a lot.

In the second half it began to feel like a slog. It felt a little repetitive, or perhaps some of the material was more familiar to me so I could guess where he was going with it. Or maybe things just started to feel a little too esoteric.

The last couple chapters contain real-world manipulations and ‘hacks’ to exploit, correct or otherwise disrupt some unwanted tendencies in the brain. It discusses how technology can aid in this, and includes some interesting VR use cases.

Overall, I really liked this. An accessible and mostly engaging discussion of cutting edge science and philosophy.

[shamim]

مگه میشه ادم کپسولی رو بخوره که میدونه دارونماست و بهش بگن خوب میشه و چنین فکری بکنه واقعا خوب بشه!
عجیب غریبه ها.

[Moh. Nasiri · Rating 5 out of 5]

ماشین تجربه - اندی کلارک

[Chris Boutté · Rating 5 out of 5]

As soon as I heard “cognitive philosophy”, I was sold. I listened to an interview with Andy Clark about this book from the awesome Converging Dialogues podcast and knew I had to check the book out. While the book is extremely interesting, it also bored me at some points. That’s nothing to do with the author or his writing, though. There are just parts that dive a little further into neuroscience than I prefer, but overall, I absolutely loved this book.

The book is about how we experience the world and how our brain is a prediction machine. It’s really interesting learning about our perception of reality and reading from a philosopher who thinks about this stuff all day. I really enjoyed the sections about how our predictions shape our biases and what we can do about it, and there’s also some awesome stuff about placebos and similar topics.

[Cole Mrgich · Rating 4 out of 5]

An interesting and persuasive expansion on his earlier cognitive theories, particularly expanding on the “Recruitment Problem” of the Extended Mind Theory. I think what he says makes sense to me.

[Lily’s Haiku Book Review · Rating 3 out of 5]

Truth is subjective
Your brain is making things up
Mind over matter

[Yaaresse]

DNF @ abt 10%.
Not sure how someone can take such a fascinating subject and make it so dull that it literally put me to sleep while reading it.

[Naomi · Rating 3 out of 5]

An interesting book looking at how our brains perceive, pay attention, and often make up bits of our daily lives. It was really interesting in places, but other bits just didn't engage me at all.
I did enjoy the chapter looking at 'rewiring' and the different techniques explored - I could have had much more on that!

[Gabrielle C · Rating 3 out of 5]

Andy Clark’s new book is a poorly done ripoff of Lisa Feldman Barrett’s book, How Emotions Are Made, published 6 years prior to Clark’s drawn out drivel. Great title and cover, Andy, but that’s about it, because your content seems to directly take from Barrett’s content and does so poorly. If you’re interested in the mind-brain-body connection, I highly recommend How Emotions Are Made, because it’s written to educate readers, rather than Clark who wants to bore you with longwinded explanations that take the scenic route, while never leading you to your predicted destination.

[Emily | emilyisoverbooked]

DNF

[Stefanos · Rating 4 out of 5]

Came for the book’s cover and Andy Clark’s fabulous shirts, stayed to see why the title was “The Experience Machine” instead of “The Predictive Machine”.

The book introduces Predictive Processing, a neuroscientific model which promises to unify perception, cognition and action. Traces of predictive processing can be found all the way back to the work of Immanuel Kant e.g., the active application of innate categories and concepts such as Space, Time and Causality. Inspired by Kant, Hermann von Helmholtz developed the concept of unconscious inference, but it wasn’t until people like Karl Friston that Predictive Processing –and Active Inference– was formulated as a comprehensive neuroscientific model.

But what is Predictive Processing? Well, perhaps it’s better to start with what it is not, or, what it aims to supplant: the conventional view of human perception, or the Bottom-Up (or feedforward) framework. In essence, according to the Bottom-Up model, the world is out there while the brain passively receives raw sensory information, processes the input signals and gradually constructs higher-level representations. Let’s take vision for example. In the Bottom-Up view, light hits the eye, stimulating the photoreceptor cells in the retina, which convert the light into neural impulses. These signals travel along the optic nerve to reach the visual cortex which processes the incoming signal, extracts features (e.g., edges, contours, shapes, colors, textures etc.). Gradually, higher-level features (e.g. object parts, configuration, movement, intentionality) are detected, forming a visual representation which can be combined with prior semantic information to recognize and categorize objects (e.g., a cat) along with typical characteristics (e.g., “furry”, “four-legged”), associated behaviors or attributes (e.g., “purring”, “playful”).



Predictive processing flips the script. It conceptualizes the brain as, first and foremost, a prediction machine constantly making predictions about the external world (and internal states) and adjusts them based on feedback (prediction errors) from sensory evidence, which are weighted based on their estimated importance (precision weighting). In essence, comprising the following elements:
- A Generative Model is constantly generating predictions regarding both external events and internal states, which are informed by prior experiences, beliefs, knowledge, and prior states.
- Moment-by-moment predictions allow for real-time adaptation to the evolving environment.
- Prediction Errors: incorrect or incomplete predictions, serving as signals to reconcile predictions with sensory evidence.
- Precision Weighting: estimates the significance of prediction errors within a specific context. Responsible for balancing between predictions and prediction errors.

So let’s revisit an example of visual perception through the lens of Predictive Processing. As you walk through a forest, your brain anticipates what is likely to be present based on prior knowledge, expectations, and contextual cues—trees, bushes, and various vegetation. Suddenly, the brain receives an unexpected input. An unaccounted rustling noise nearby. This triggers the generation of new hypotheses regarding the possible causes of the noise, drawing upon your past experiences and expectations.
- Input: “Rustling noise” + low-resolution visual input + contextual expectations + prior experience
- Prediction: “Is that a snake!?! Prepare to flee!?!”
- Prediction Error: “Oh no, Wait. That's not a snake!”
- Precision weighting: “We must focus on the error from the visual input. We must get this right people!”
- Prediction: “Oh, it was just a branch”.
- Prediction Errors: “No incoming error!”
…

Andy Clark's begins with perception, where a lot of work has been done, aligning with both prior experimental findings and emerging research. He delves into how predictive processing offers a more comprehensive framework for understanding various perceptual phenomena, including illusions such as the convex face illusion, sine-wave speech, Mooney images, and even the “Viral Dress”.

Beyond Perception
But Predictive Processing does not only claim to account for perception, but also action, planning, learning, emotions and perhaps even consciousness. Very crudely:

Action: a way to resolve predictive errors by changing the world. “I am going to be thirsty. I should be drinking that glass of water” -> Prediction error: “i am not drinking that glass of water” -> “move the hand, grab the glass and drink” fulfills the prediction.

Learning: through repeated errors (e.g., playing the wrong note while learning to play the piano) and corrections, the generative model is optimized so as to reduce future errors.

Emotions: Instead of discrete and hard-wired categories of experience that are produced by distinct brain circuits and triggered based on the context, Predictive Processing posits that emotions are well, predicted – constructed and socially learned.
So in the previous example, the brain predicts the existence of a snake as well as the appropriate affective response if a snake was indeed there. This affective response encompasses a range of sensations and physiological changes that can be collectively summarized and categorized as “fear”. (See Lisa Feldman Barrett).

Consciousness: Returning to my initial question, where does subjective experience fit into the picture? Clark admits that he “nervously” stands by his title, considering Predictive Processing to be our best clue so far for explaining subjective experience. In this view, sentience may be explained as the result of turning the predictive machinery inward, predicting future internal and bodily states. And by developing a predictive account of the construction of the self and language and self-awareness we might come close to explaining consciousness while deflating the Hard Problem.

When things go wrong
In the previous example, the predictive brain worked reasonably well, maintaining a reasonable balance between predictions and sensory evidence – or more precisely, the interplay between predictions, prediction errors and precision weighting. However, this balance is not static. There is no single best weighting schema optimal for every scenario. The importance, reliability or confidence of predictions can vary, as the world is highly complex, and different situations demand different responses.

You can imagine how this process can go wrong. If too much emphasis is placed on predictions and not enough on prediction errors, the brain will start to hallucinate. Such imbalances could potentially explain conditions like psychosis.
Here, we can also consider the role of prior experience (e.g., trauma) and expectations (e.g., social bias) in shaping the predictions of the generative model. For instance, if a person has PTSD from being bitten by a venomous snake, their brain might predict every branch as a snake to be on the “safe side”. Or how a racist cop might misperceive a Black person holding a phone as holding a gun based on false prior expectations and biases. (IF we are willing to take their word for it).

Conversely, if too much emphasis is placed on sensory evidence, it becomes challenging to identify subtle patterns in noisy environments. The brain would struggle to determine where to focus its resources at any given time. According to Clark, emerging evidence suggests that individuals on the autism spectrum may experience an over-weighting of incoming sensory evidence, leading to difficulties in filtering and prioritizing information.

Predictive brains may also have a tendency to fall into spurious self-confirming cycles, contributing to conditions such as social anxiety, depression, and chronic pain. For instance, while acute pain signals immediate danger, prompting the subject to stop an activity, with chronic pain, the system can become compromised, associating certain actions with the expectation of pain. This can lead to real pain experiences even in the absence of tissue damage (nociplastic pain). In such cases, the predicted experience of pain is materialized, reinforcing that the prediction was correct, thus creating a vicious cycle that is difficult to break.

Hacking the Predictive Brain
Andy Clark discusses how observations like these, aligns Predictive Processing very well with the emerging field of Computational Psychiatry, promising to bridge the gap between neuroscience and psychiatry. And while the aforementioned examples focused on how predictive brains can go wrong, there is also the flip side. We can improve predictive brains by nudging the predictive machinery to the desired direction – altering priors, expectations or re-calibrating precision weighting. Clark discusses Cognitive Behavioral Therapy, Self-affirmation, Pain Reprocessing Theory, VR therapy, Music therapy etc and how they fit in the Predictive Processing framework.

For instance, consider how past trauma can be seen as over-weighting certain predictions in order to avoid the same traumatic event and how various approaches can help alter this weighting. Or how mindfulness may involve gaining greater control over the predictive machinery of the brain, particularly in terms of precision weighting, aka attention.

Predicting the score of this review
Overall, I found the initial chapters introducing Predictive Processing to be the strongest – with my favorite being the Appendix, which delved into more technical details. Sidenote: If you are already familiar with the work of Anil Seth or Lisa Feldman Barrett, you might not encounter much new information.
Personally, I also appreciated the chapter on the connection between Predictive Processing and Computational Psychiatry even if Clark primarily re-contextualizes existing observations and knowledge from the perspective of Predictive Processing.
Given that Predictive Processing is still in its infancy, it may be premature to expect some radical shift in our understanding of mental health. In any case, the field seems to have a promising future.

On the other hand, I found Chapter 5 and 6 to be the weakest sections of the book. The former largely revisits Clark’s previous work, in collaboration with David Chalmer, on Extended Cognition, while the latter briefly explores the societal and political implications of Predictive Processing. It primarily delves into social biases, racial and gender stereotypes, and touches on the spread of misinformation—undoubtedly critical issues, yet I would have appreciated a broader (or at least deeper) exploration.

Given my background in Computer Science and my current role as a Deep Learning researcher, Predictive Processing makes sense to me — maybe a bit too much sense. As a result, I approached the book with some caution, concerned that it might simply reaffirm my prior experiences with predictive and generative AI models. Consequently, I had hoped to encounter more in-depth discussions and comparisons with alternative models, as well as a greater emphasis on the current debates within the field and limitations of the framework. Unfortunately, these expectations were not met for the most part. Thankfully, I was pleased to encounter a recent review examining the empirical evidence for predictive processing, concluding that they “offer modest support” and it “tends to explain behavioral data reasonably well”. There is also a nice editorial (co-written by Andy Clark) that delves a bit more into the ongoing debates in the field.

[Tracey · Rating 4 out of 5]

This is an interesting read. It flipped some things around in my brain in terms of how human senses work in relation to what we call objective reality. I am still working through a lot of the big ideas in this, but I think there were some immediate take aways related to how our bodies experience stress, physical pain, and emotions that are helpful. I think this is probably a little longer than it needs to be. I wish more of the book was written like the appendix, to be honest. Straightforward in its approach. Overall, an interesting read.

[Tom Walsh · Rating 5 out of 5]

Nice synthesis of the components of Consciousness.

Does the Mind’s Computer invent Reality?! Or does it sit there waiting for our Senses to tell it what’s going on?!

Two different takes on how Consciousness arises out of the three pound machine between our ears. Andy Clark’s Theory provides a fusion of an active Mind constantly seeking Sense Data to manufacture and error check predictions about our internal states and external interactions with The World we constantly encounter.

This highly readable account paints a very satisfying picture of Brain and Senses working together to help us manage our Reality. Plenty of research and examples to back up his claims clear enough for the interested layman. Five Stars. *****

[Denis Romanovsky · Rating 4 out of 5]

This book explains a modern theory of predictive brain. A kind of holistic approach, all components together, including senory information, internal predictions with emotions and actions that impact the prediction process. The only negative aspect about this book is the style of writing - sometimes it was hard to follow and I had to re-read.

[N · Rating 3 out of 5]

Notes
Nociceptive pain can be compared to the warning light in a car when it correctly indicates some kind of mechanical or electrical problem. Neuropathic pain is more like a faulty warning light—a constant intrusive signal caused by damage to the warning light wiring.

“nociplastic pain.” This was defined as pain arising from abnormal processing of pain signals without any clear evidence of either tissue damage or any other recognized systemic pathology. In other words, the warning light is on but there is simply no obvious cause—not even damage to the warning light wiring itself.

religious beliefs could regulate the experience of physical suffering, arguing that a kind of high-level reframing of the sensory signals mediated their actual experience and exerted an analgesic effect.

it proactively predicts the arrival and intensity of pain information and estimates the likely reliability of its own predictions, up- or down-regulating experiential pain accordingly.

to find a needle recently dropped in a bed of hay. According to predictive processing, my brain ups the precision-weighting on specific aspects of the visual information that would indicate a small silvery object, thereby increasing my chances of success. That’s what attention, if these accounts are correct, really is—attention is the brain adjusting its precision-weightings as we go about our daily tasks, using knowledge and sensing to their best effect. By attending correctly, I become better able to spot and respond to whatever matters most for the task I am trying to perform (for more on this, see the Appendix). Precision estimation is thus the heart and soul of flexible, fluid intelligence.

Schizophrenia often involves both hallucinations (apparent perceptions that fail to match the real world) and delusions—strange beliefs such as the belief in an organization of technological experts. Important early research applying ideas about the predictive brain suggested that these two features might be flowing from a common source: waves of falsely generated prediction error signals.

chronic depression involves a resistance to updating our negative expectations when confronted with what ought to be good evidence of positive outcomes. This failure to update in the face of good evidence (Barrett’s “locked-in brain”) most likely involves abnormally high precision on prior negative beliefs, which in turn robs unanticipated positive information of the power to alter the inner model that is delivering negative anticipations. The highly weighted (hidden) belief that outcomes will be negative acts as what has usefully been described as a kind of “cognitive immunization” to the effects of countervailing positive information, causing us to either avoid gathering, ignore, or otherwise downgrade perfectly good positive evidence—such as genuine evidence that we are liked and valued.

“salience detection hypothesis,” these chills occur when we encounter something that our brain identifies as critical new information that resolves important uncertainties. This makes it a kind of physiological echo of the “aha” moment when things suddenly fall into place.

phrase—“probability designs”: artifacts engineered to interact in reliable ways with our own predictive brains. Books, novels, plays, and movies are all probability designs. Attention (precision-weighting) plausibly plays a key role here,

Ideomotor theory: where perception is brain adjusting prediction (precision-weighting) to best explain the sensory evidence; action is brain predicting sensory evidence that will match a desired effect and then adjusting motion that produces the sensory evidence with least errors.

Why visualization works - create a stream of expected sensory evidence that makes more number of microadjustments in motor actions to produce the best shot.

If a robot arm inserts random time-delays in tickling, we can indeed tickle ourselves

Exteroception (external stimuli) vs Interoception (blood sugar, body temp, heart rate) that informs homeostasis (bring body back to equilibrium point, sweating) vs allostasis (change the equilibrium point adaptively - release cortisol to flood blood sugar for action).

Error dynamics: brain evaluates how well it is minimizing prediction errors. In-the-zone, high error success (flow state environment that’s neither too easy, nothing to learn, nor too hard, not able to reduce errors).

In practice, chronic pain seldom falls solely into one of the three categories (nociceptive, neuropathic, or nociplastic). Instead, there is a continuum of cases and a great many—especially cancer pain and spine pain—seem to involve complex mixtures of all three. What matters for present purposes is not to shoehorn anyone’s experience of chronic pain into one of the categories, so much as to recognize that a lot of the disability that comes with chronic pain and injury is linked to that hidden inference—the inference that the felt pain means we ought not to push ourselves any further. In the grip of that inference we shrink our worlds, lending further support to the prediction that we are simply not capable of doing many of the things that would otherwise expand and enrich our lives. Yet in many cases of chronic pain and disability, this inference is mistaken.

[Mehtap exotiquetv · Rating 4 out of 5]

Realität. Sie entsteht in unserem Kopf. Doch wie? Wie schafft es unser Gehirn die Realität zu formen? Und was können wir daraus lernen.

[Atlantisli]

not much to offer if you are already from cognitive science field but if you are not, i think it is a great book to learn about predictive processing.

[Ray Mathew-Santhosham · Rating 5 out of 5]

New favorite!!! Very illuminating about a variety of different conditions/ways of experiencing life like PTSD, depression, autism spectrum disorder, chronic pain, as well as neurotypical minds.

[Andreas Dueck · Rating 4 out of 5]

Interesting insight on how the mind operates and how it builds our day to day experience.

[Lisa Dick · Rating 3 out of 5]

The first part was pretty good, but then it felt very repetitive and that it dragged on. I found Jeff Hawkins 'A Thousand Brains' much better.

[Alina · Rating 5 out of 5]

Clark is a wonderful writer with compelling, unexpected ideas. My only complaint with this book is that it is repetitive. Clark’s key ideas could be spelled out in probably one chapter’s length, and the rest of the book involves applications of these ideas across various contexts, like explaining everyday illusions and psychopathological symptoms. He also spends a good amount of time on his more speculative “extended mind” thesis as a statement about the metaphysics of mind, which I’m not as interested in, and don’t find as convincing. This repetitive quality, however, is natural, given its genre as a book-length project pitched at a general audience.

For readers who want to get directly to the heart of Clark’s ideas, I’d recommend two papers of his, which represent different philosophical projects of his. First, there’s “Dreaming the Whole Cat: Generative Models, Predictive Processing, and the Enactivist Conception of Perceptual Experience” (2012), which presents his ideas on predictive processing. The ideas in this paper are repeated in this book. The basic idea is that a compelling picture of mental processing is that there are many hierarchical models, like layers, where the higher ones try to predict the patterns in the lower ones. We are driven to act in ways which alter sensory input so that these predictions can be done more quickly and successfully.

For example, when we see a tomato, we see the whole tomato, even though the visual data coming in reflects just one side of it. This seeing the whole is made possible by the fact that higher levels of our mental model of the world are trained to yield that the best prediction of what is going on in the world, in light of this sensory input, is that there’s a whole tomato. So we see the whole. In other words, we perceive and experience what our brains predict is most likely, rather than what’s literally coming in.

Moreover, unbeknownst to us, our actions and expectations are always for the sake of minimizing error. If a prediction is erroneous, the brain will come up with what it takes to be the most likely hypothesis to explain why the error happened, and our experience will altered accordingly, if this hypothesis were correct. For example, if I bite into the tomato, and it tastes too sugary to be a tomato, I quickly experience it as some sweet fruit I’ve mistaken, and later a friend confirms that it was a persimmon. If usually whenever I see my friend we are high energy and greet each other enthusiastically, and this time around he is low energy, I quickly experience him as sad or tired. If I were in a socially anxious mood, I might experience him as having a grudge against me, or some other explanation of the prediction error that seems most likely, in light of this mood.

The second paper I’d recommend, which gets to the heart of Clark’s ideas repeated much throughout this book, is “Word, niche and super-niche: how language makes minds matter more” (2005). Here, Clark explores a speculative but I find compelling idea: our language use has added a “dimension” to reality, which systematically changes the possibilities of perception and thought. Particularly, language, being abstract, offers labels/categories, which particulars get tagged with, so that we can deal with large swathes of particulars at a single blow, a single perception or thought. Reality becomes “symbolic,” where single things come to be registered as many possible things, and dealing with the former is a proxy for dealing with the latter. This tagging also allows for enhanced powers of selective attention. When much is going on, we can focus on a few things, those we’ve chosen by implicitly using language about it. This ties in with our enhanced agency; we can control our environments and lives by selective attention. We need not literally use language to achieve these effects; explicitly using language “scaffolds” our cognitive and perceptual skills, so that without explicitly using it, we can have perceptions and thoughts akin to those that’d occur if we did.

I have a side-thought about a possible tension between these two philosophical projects of Clark’s. His predictive processing model of cognition suggests that all mental activities (those traditionally called perception vs imagination vs reasoning, for example) can be sufficiently understood by one and the same vocabulary of “perceptual experience” and one and the same model of cognition. But it seems that there are crucial differences between these types of mental activity. Clark appreciates this in his work on the role of language use in the development of human cognition; using language isn’t the same as passively perceiving the world, and it changes the possibilities of later passive experiences.

I’d like to see work done on trying to fit such points about agency and selective attention, as distinctive powers of language use, under the predictive processing framework. I’d fancy these most neatly fit under the concept built into the framework regarding the hierarchical levels of prediction; the higher order levels correspond to our personal agency and experience, and our deliberately challenging or making sense of our experiences via language use could be understood as predictive activities going on at these higher levels.

The issue with this, however, is that there is reason to doubt that “prediction” can amount to the same sort of mental activity between the deliberate, language-based sort, on the one hand, and the automated, information-processing sort, on the other hand. When I consciously or deliberately wonder why my friend seems blue, this is self-conscious, and so anything I think of will “spark off” in my consciousness objections or alternatives, or additional associations which vindicate or supplement what I’m coming up with. I can always challenge whatever is sparked off in my mind. In contrast, the prediction that goes down at the lower levels of cognition does not involve this possibility of sensitive monitoring, of challenging or elaborating upon itself (unless we want to posit a homunculus picture of the mind, where there is a little person who makes up our brain and has its own autonomy and mental will, independent from our own; and this kicks off an undesirable regress).

Maybe there’s something to that such predictions that use language are going on at lower levels, and there’s no need to posit a homunculus; the difference between the mental activity of “prediction” at those levels, and that of the higher level of our own volition, is a matter of our turning our attention to the happenings at those levels, like a spotlight illuminating that which was previously dark. But this picture leaves unexplained what this volitional attention consists in; if we rely upon Clark’s predictive processing model, it would need to be explained in terms of happenings at the lower levels once more. There seems to be no entry point in his model for our personal agency to come in, to be appreciated as significantly distinct from whatever automated information-processing that goes on in the brain.

Clark appreciates our agency, which comes out in his work on language use. The fact that his predictive processing work doesn’t neatly accommodate his ideas about agency amounts to a tension which whose resolution I’d like to see.

One more complaint about collapsing distinct types of mental activity under the language of perception and under one processing model: I worry that information we consciously retrieve or create, at those higher levels, has different “affordances.” The affordances might be distinct on the basis of whether such information we consciously manipulate is episodically remembered, or creatively imagined, or created in a self-disciplined manner according to certain norms, either ethical or epistemological—not even yet to speak of the difference between these affordances, as a whole, and the affordances of information that is subpersonally manipulated.

For example, consider the differences in “affordances” of information we consciously deal with. If I’m remembering a past experience with my friend, I sense this as tied to reality, and it will shape my reasoning as if it were true. In contrast, if I fancifully imagining my friend (as smoking too much again as a possible explanation to his blue mood, when I have no evidence at all that he has done this), this imagining will not shape my reasoning in this way. In contrast, whatever information low levels of cognitive processing deal with have neither affordance; in order for them to have either, we must conscious attend to it (or “let it rise” to conscious awareness through some other means), and by that point, it is no longer the same kind of information (e.g., sensory data v an understood, intelligible state of affairs).

Anyways, Clark’s ideas are compelling and fun to engage with (as reflected in my rambling above perhaps). Read this book if you want to see these ideas introduced in a non-technical way and worked out under the context of a range of everyday experiences.

[Manu · Rating 4 out of 5]

The subtitle of the book is "How our minds predict and shape reality", and that's what the book is about. The conventional notion of cognition, at least to me, is that it begins with sense organs perceiving and providing inputs from what we experience, and the brain quickly piecing it all together to present me a coherent picture of what is, and what I should do next. But if we go by the "predictive brain" thesis, the brain doesn’t just passively interpret the world but is constantly predicting, shaping, and refining our reality based on sensory inputs.

Back in the 19th-century, the scientist Hermann von Helmholtz posited that there must be some underlying (unconscious) process of logical reasoning that is inherent in optical and auditory perception. Now, cognitive science agrees - a large part of our experience is guided by the predictive brain, which constructs a simulation, thus already shaping our understanding of reality, but then sometimes iterates its own understanding based on inputs from the world, and sometimes makes us act in ways to reduce the prediction error, depending on error dynamics. This means that nothing in our experience arrives unfiltered - from basic pain to ego, everything is a construct. "We are what predictive brains build".

The predictive brain has four primary elements - a 'generative' model that has some existing data and knowledge about our world at large, the constant predictions that come from it, the 'prediction errors' that arise when incomplete or incorrect predictions meet sensory evidence and account for it, and the precision weighting (estimates) that define the impact of sensory stimulations and predictions. It is amazing how the underlying process of perception and action are the same, and only the direction differs. In fact, our actions are also self-fulfilling predictions, and in both short and long term, there are ways in which we can make it work in our favour. There is some scientific backing for 'manifestation', it would seem that the brain orchestrates a series of actions to fulfil its predictions. No, you can't wish things/experiences into existence, you still have to work for it! :)

This predictive processing helps explain not only ordinary perception but also phenomena like chronic pain, mental health disorders, and even illusions - these experiences may be due to our brain’s predictions going slightly awry. A great example is chronic pain, which can be seen as a misfiring of this predictive system, where the brain expects pain, thus “creating” it even when no external cause remains. This insight means that potentially, the treatment method could be about recalibrating these predictive mechanisms rather than solely addressing physical symptoms.

What this thesis also does is question the duality of things like self and world, mind and body, and so on. Clark quotes Lisa Feldman Barrett - "every thought, memory, emotion, or perception that you construct in your life includes something about the state of your body. Your interoceptive network, which regulates your body budget, is launching these cascades." A mix of inward-looking, outward-looking, and action-guiding to construct and control a controlled hallucination. Fascinating!

Another interesting part was how nature has played it such that we also have openness and exploration to inform the predictive brain. Error dynamics estimations track how well the brain is doing at minimising prediction error. Being 'in the zone' is thus a mix of reduced prediction error and handling any error fluently. Those sensitive to their own error dynamics will seek out learning environments, so in the long game, the prediction errors can be further reduced.

Cognitive philosophy seems to have become a theme this year for me, and Andy Clark references my favourite read on the subject this year - Anil Seth's Being You, as well as another from four years ago - Lisa Feldman Barrett's How Emotions are Made. This book is a great addition to this 'series' - philosophy, cognitive science, and practical implications. It has given me me more perspective as I watch my brain!

[Jess · Rating 4 out of 5]

The title captures the thesis - this is an exploration of how our minds predict and shape reality. I enjoyed this. It’s interesting to consider how much of our experience is shaped through pure sensory input vs how much we project on to any given situation.

[Chris Chu · Rating 5 out of 5]

Really good read. I thoroughly enjoyed the book.
The main point of the book is predictive processing and extended mind. The author argues that our whole experience is shaped by the predictions of our brain, and the precision-weighted interplay between inward bodily information, outward sensory information and the prediction of our mind create a unifying picture of human experience. What we do every moment is trying to minimize the prediction error of our mind. And there are some very compelling arguments and evidences across many dimensions that suggest this is really the case. This is my first time learning this theory and the writing makes it really easy to understand, and the pace and structure of the book makes the author's ideas clear and accessible. I really appreciate that all the researches and arguments presented in the book are listed at the note section, so I can look those up and see myself if I agree with the interpretation of the author. This is such a fascinating topic to me and I think this book does a very good job in introducing the theory of predictive brain. The only thing I don't really agree with the theory is how we formulate actions, as for me the explanation sounds a bit sketchy and not very straightforward like the other parts of the thinking process.
Overall, this is such a nicely written science book that I would recommend to anyone who wants to learn how our mind creates the world we live in, and now I have a brand new way to think about how I see the world and how everything I know shapes my reality.

[Craig Martin · Rating 3 out of 5]

Clark is an influential thinker in cognitive neuroscience and several associated fields. He was a friend of the recently deceased Daniel Dennett, and also wrote a famous paper on the ‘extended mind’ with David Chalmers, who coined the ‘Hard problem of consciousness’, so he is used to controversy.

His main expertise and thoughts are in the field of ‘predictive processing’ of the human mind. In the ‘Experience Machine’ Clark has written an engaging and thought-provoking introduction to this subject . Clark's writing is clear and accessible, making complex ideas approachable for a wide audience. While I appreciated the broad strokes of his arguments and the philosophical insights he offered, I felt that the book was somewhat light on detail, and empirical evidence for his ideas.

Predictive processing or ‘predictive coding’ is essentially a mechanistic ‘error minimising’ concept, and one familiar to control engineers, data scientists and AI enthusiasts. Model a prediction, compare it with the real world, and adjust it until it fits (think of regression analysis in statistics, gradient descent used in the back-propagation algorithms of supervised learning models, and more recently token prediction in generative AI models, such as Chat-GPT). Clark suggests that predictive coding is the way in which the brain makes sense of the world.

Many neuroscientists present a feed-forward approach for the brain’s cognitive processsing: part of the eye’s retina receives light photons reflected from a ‘tiger’, these trigger an amplified neurochemical/electrical tsunami that is passed on through several parts of specialised neural circuitry tuned to lines, orientations and patterns in the visual cortex. The brain’s neural wiring (axons and dendrites) connect up with memory circuits. Long-term memory drops ‘Tiger’ into short term memory, and various neurotransmitters in older parts of the brain driver motor neurons, and you flee, fight or freeze, but generally against a Tiger its game over.

Clark says it goes the other way round. He says there is another direction of influence—one running backward, from deep within the brain down toward the eyes and other sensory organs. He states that the number of neuronal connections carrying signals backward in this way exceeds the number of connections carrying signals forward by a very substantial margin, in some places by as much as four to one. A quick summary of this idea is as follows:-

Prediction Generation: The brain continuously generates models or predictions about the world based on prior knowledge and experiences.

Error Minimization: When sensory input is received, it is compared against these predictions. Any discrepancy between the prediction and the actual input is termed a "prediction error."

Updating Models: The brain updates its models to minimize prediction errors. This updating process allows the brain to refine its understanding of the world and improve future predictions.

Hierarchical Processing:Predictive coding operates hierarchically, with higher-level brain areas generating more abstract predictions that guide lower-level sensory processing.

Perception as Inference: Perception is viewed as an inferential process where the brain infers the most likely causes of sensory inputs, effectively perceiving the world by continuously testing and refining its predictions.

This suggests that in addition to the neural highway of information flow along axons from the sensory organs to various sub processing units in the brain, there is an equivalent contra-flow system taking the predictions and weighing them against what the senses say. In many ways, he suggests, quoting Anil Seth:

“We do not see things as they are, we see them, as we are”

From a technical perspective this sounds similar to Karl Friston’s ‘Free energy’ principle (which Anil Seth introduced in Being You: A New Science of Consciousness). In the context of "The Experience Machine," Clark extends the idea of predictive coding to explain how our experiences and perceptions are shaped by the brain's predictive processes. The Experience Machine metaphor suggests that our sense of reality is constructed by the brain's predictions, making our experiences fundamentally subjective and constructed. This has many implications:

Active Perception: Our perception of reality is not a passive reception of sensory data but an active process of hypothesis testing and prediction.

Flexibility and Adaptability: The brain's predictive coding mechanism allows for a high degree of flexibility and adaptability in responding to new and unexpected sensory inputs.

Illusions and Expectations:This theory explains why illusions and expectations can strongly influence what we perceive, as the brain's predictions can sometimes override actual sensory input.

“Emotion, mood and even planning are all based in prediction too. Depression, anxiety and fatigue all reflect alterations to the hidden prediction that shape our experience. Alter those predictions (for example, by ‘reframing’ a situation using different words) and our experience alters.”

I found myself comparing Clark's approach to that of his University of Sussex colleague, Anil Seth. Seth's work often delves more deeply into the neuroscientific evidence, offering a richer, more data-driven perspective on similar themes. Additionally, the contributions of other leading figures in the field, such as Christof Koch and Philip Goff, provide contrasting viewpoints that are both rigorous and illuminating. Koch’s extensive research on the neural correlates of consciousness and Goff’s exploration of panpsychism offer valuable context and alternative explanations to Clark's arguments.

Overall, Experience Machine is a stimulating read that successfully introduces important concepts in the philosophy of mind and cognitive science. Yet, for those seeking a more detailed and evidence-based examination of these ideas, supplementing this book with works by Seth, Koch, and Goff might provide a more rounded understanding.

I gave the book three stars.