The Experience Machine: How Our Minds Predict and Shape Reality

by Andy Clark

here’s the twist—these experimentally biased experiences induced expectations themselves. Seeming to experience higher or lower levels of pain in the ways cued by the geometric shapes created a feedback loop in which participants’ own experiences appeared to confirm their (false) cue-based expectations. This “false confirmation” cemented their misguided belief in the predictive power of the different cues. You might have expected subjects to learn, over time, that the cues were not reliable, but this did not occur. But then again, how could it? Each time the heat was applied, the different geometric cues caused them to experience its intensity in line with the false expectations they had learned. So every exposure seemed (subjectively speaking) to simply confirm the predictive power of the cues!

It may be worth pausing for a note on terminology. I will use “structural disorder” and “structural damage” to refer to any cases where there is a standard neurological condition, bodily injury, or disease process present: one whose action already accounts for the experienced pain, disability, or sensory change.

At the core of that picture is the idea that predictions about our own sensory capacities or physical abilities are playing a key role, causing genuinely experienced symptoms to fall into line with those hidden expectations.

Precision variation is what attention (a useful but somewhat nebulous concept) 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

Predictions of pain or impairment become highly overweighted, and those predictions overwhelm the actual sensory evidence, forcing experience to conform to our own hidden but misplaced expectations.

simply distracting the sufferer by making them direct their attention elsewhere often makes functional (but not structural) tremors vanish. Patients with these tremors also spend much longer looking at them than those whose tremors have standard causes and they greatly overestimate how often their tremor occurs.

This creates a version of the famous “refrigerator light illusion.” You might infer that your fridge light is constantly on just because the light is on every time you look inside. But actually, it is the act of looking (opening the door) that turns on the light. Similarly, you might believe you have a near-constant tremor because the

tremor is always there when you pay attention to it. But if the tremor is actually in whole or in part the result of the process of “predicting and attending” itself, that assumption may be wildly wrong.

“Hoover’s sign.” Named after the American physician Charles Franklin Hoover, this is present when a patient with unexplained weakness in one leg proves able, when their attention is directed elsewhere, to exert normal amounts of pressure with that leg.

Strongly anticipating pain, numbness, weakness, or other symptoms alters patterns of attention (precision-weightings) in ways that can either

amplify or entirely generate the experience—which then seems to confirm those very expectations. This is simply our old friend the “White Christmas” effect all over again, but here affecting the way we experience our own body rather than the sounds we hear.

This is rather like the case of the performer with stage fright whose true abilities are masked by their own mounting expectations of failure. The circularity is daunting. Every new instance of stage fright confirms the expectations of failure, and those expectations ensure that the instances of failure accumulate. Recognizing this circularity is, however, often the first step in breaking the cycle, as we’ll see later when looking at ways to “hack” our own predictive brains.

after several bouts of back pain, people start to process the world differently…their pain becomes embedded [among] the things they associate with themselves. If they are shown an image of a staircase, for instance, their first thought is, “I can’t climb it.” After a while, you see and feel things coated with pain. You no longer need the injury to feel pain. And you might experience more intense pain, purely because you’re expecting it.

attention and expectation are key players in the construction of all our experiences of health and illness, and this is true even when standard structural causes (damage or disease) are present.

What we see, hear, and feel—even when everything is working exactly as it should—is never a direct reflection of the state of our own body or the wider world. Instead, the world and body we experience is always part construct: a product of our own conscious and nonconscious predictions.

WE HAVE seen that expectations and predictions deeply influence what we see, hear, and feel.

this is not a facile nod toward “positive thinking.” Rather, it is a detailed proposal about how our brains control our bodies. The upshot is that successful action involves a kind of self-fulfilling prophecy. Predicting the detailed sensory effects of a movement is what causes that very movement to unfold.

The core idea was that actions come about because we mentally represent the completed effects of the action. In other words, the idea of the completed action is what brings the actual action about. This is sometimes said to reverse a commonsense notion of causality, since instead of the action causing

the effect, it is the representation of the effect (the completed action) that causes the action itself to unfold. It’s not really that the effect precedes the cause, but rather that the cause turns out to be a kind of mental image of the effect. This became known as the “ideomotor theory of action,” since the idea (or mental image) of the completed motor action is what brings the actual movements about.

shows how the “idea” of a successful action can be the very thing that brings that action about.

By launching a cascade of sensory predictions, and then rendering them true by means of action (thus eliminating the resulting prediction errors), the brain creates the desired movements. In other words, I strongly predict looking out the window and that prediction acts as a kind of self-fulfilling prophecy.

the whole process is one in which abstract predictions cause ever-more-concrete predictions, enabling my top-level representation of a desired consequence (seeing the gulls) to cause the bodily actions that bring it about.

There are two different, but equally effective, ways to minimize prediction errors during our encounters with the world. The first is by using prediction errors to help us discover the best guess about how things are out there in the world. But the second is to act so as to make the world fit some of our predictions. Instead of finding the prediction that best fits the sensory evidence (perception), you now

find or create the sensory evidence that best fits the prediction. This is the predictive processing route to action. Actions are simply the brain’s way of making its own proprioceptive predictions come true.

In each case, the brain is seeking to achieve a fit between what is predicted and what the sensory evidence suggests. But in the case of action, the fit

is achieved by altering the sensory evidence to bring it into line with the prediction.

That means actively disattending to the present state of the arm, hence dampening down that sensory information. This kind of careful attending and disattending is also familiar from sports coaching, where skilled players are taught to imagine the desired outcome (the spot where the tennis ball will go, the arc of the golf ball to the green).

But for this to work, we need to attend away from the sensory information about our current bodily state and attend toward (increase the precision-weighting upon) the sensory information that would be expected if the action were successfully performed. This also helps explain why it is that (in the case of already skilled players) simply imagining yourself playing the shots or making the movements can actually improve subsequent performance. Even though no physical actions are involved (this is pure imaginative rehearsal) we are training ourselves to generate the mental representations that, on the day, will actually serve to bring the shots or movements about.

For many processes, waiting for feedback cues (such as wilting plants) is a bad idea, as what is really needed is ongoing preemptive action.

We do not wait until we are actually out of fuel (sugar, water) before taking remedial action. Instead, we model ourselves well enough to step in in good time. The same situation arises in nuclear reactors, aviation, and many other areas. In the case of a nuclear reactor, it is crystal-clear that waiting too long for feedback before initiating corrective action is ill-advised. Systems that instead predict the

future from their current state, current actions, and a model of how those actions will affect that state, are always one step ahead of the game.

Bodily motions are selected to bring sensory inputs into line with precise top-level predictions about just where the car should now be heading, at what speed, and so on. Those actions, in turn, reflect the overarching policy (drive fast but safely, and don’t get pulled over) that looks most likely to get us to the airport on time.

as all novices know, just wishing that our car or racquet would respond a certain way is not sufficient to make it so. This is because fluent performance of that kind requires a highly trained inner model. Such a model puts perception and action together at every level, from the top (seeing the right trajectory) to the bottom—pressing on the brake just so, while turning the steering wheel exactly the right amount. At every stage, success follows when the inner model delivers predictions that act as motor commands bringing about the predicted sensory states. And as noted earlier, these sensory predictions already factor in all the biomechanics, synergies, and shortcuts that evolution and training have installed.

As sports personalities have been known to remark following especially spectacular performances, “the more I practice the luckier I get.” The great Muhammad Ali once said, “If my mind can conceive it and my heart can believe it, then I can achieve it.” But there is no substitute—as Ali himself knew very well indeed—for hard work and training. The hidden task of all that training, we can now appreciate, is to enable our brains to predict (via a cascade that often starts with a very high-level goal or aim) the many subtle sensory consequences of an unfolding successful action. This is very different to thinking we need to learn how to “make the action.” Instead, we learn how things will look and feel if we are getting the action right—we learn its sensory consequences, highlighting those most necessary for success. Get all that right and predicting those looks and feels automatically controls the body in the ways that most reliably ensure success. The good news is that difficult control problems can, in this way, be solved simply by learning to predict what we would see and feel if we were getting things right.

We have begun to glimpse what may be one of the deepest implications of predictive processing. It is that goal-directed behavior involves using predicted outcomes to help structure the actions that will best serve to make those outcomes real.

What about even longer-term goals and projects? These are not so very different, except that we must now minimize prediction errors in a yet more abstract and temporally extended space. If I plan to become a better surfer, my brain needs to make the “realistic-yet-optimistic” prediction that I will indeed later be such a surfer. With that goal (long-term prediction) active, I can use what I know about how things work in the world to identify important steps along the way, generating a policy that might—according to my current skill set and personal circumstances—include moving to the coast, taking classes, or vacationing in Tarifa.

We must at some level strongly predict that we will occupy the states that we can plausibly attain and that best realize our goals. We will then act in ways designed to eliminate errors calculated relative to the optimistic-yet-realistic prediction that those goals are achieved. Realistic optimism is thus the order of the day.

In other words, are things going better or worse than expected? Things are going better than expected if we are fluently eliminating more (and more important) errors than anticipated. Things are going worse than expected if we are encountering more errors, or dealing with them less fluently, than anticipated. The feeling of “being in the zone” in sports reflects unexpectedly good error dynamics of this kind.

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. Every prediction you make, and every categorization your brain completes, is always in relation to the activity of your heart and lungs, your metabolism, your immune function, and the other systems that contribute to your body budget. The whole of our mental lives, Barrett argues, reflects nothing so much as the brain’s busy and deeply anticipatory body-budgeting activity.

An intriguing suggestion, again from Professor Lisa Feldman Barrett and colleagues, is that depression is often best seen as a “disorder of allostasis.” This would mean that depression

involves mistaken forms of bodily prediction involving energy regulation.

For example, we expect not to go out and explore new opportunities, leading us to stay home, and then find that new opportunities (as predicted) keep on failing to present themselves.

One important and consistent finding in this area is that 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.

When we perceive the world, we weave together information about what’s out there with information about our own inner physiological states and our own tendencies to action.

His visual experiences here seem to bear out his strong prior beliefs. But looking back on this today, we may well suspect that it was really the other way around—that Leeuwenhoek’s visual experience reflected nothing so much as those strong prior beliefs.

“feeling significantly distressed or threatened can predictively contribute to perceiving the world as more

stressful or threatening in a very literal sense.”

One place we can clearly intervene is by altering the environments that train our own prediction machinery in the first place.