What Is It Like to Be a Prediction Machine?
Imagine that you are a system in a universe that evolves relentlessly through time. You want, above all else, to persist.
Although the universe in which you exist is always changing, it has regularities that make some prediction of the future possible. If it had no regularities, if it were completely stochastic, evolving randomly, you would not exist at all. Stable structures only arise in worlds that are at least partly predictable, and even then they persist only temporarily. Left to itself, structure erodes. What allowed you to persist yesterday may fail tomorrow.
Your future existence depends on what happens next.
In such a world, prediction is not optional. It is the only way to act in time rather than too late. It is the only way to allocate limited resources wisely. It is the only way to respond to change before change overwhelms you.
From this point of view, perception itself is already predictive. Action is predictive. Memory is predictive because it shapes expectations about what is likely to happen again. A system that does not anticipate the future cannot survive for long in a world that unfolds through time, continually presenting new challenges to persistence.
This is the situation every living system finds itself in. We are entities with incomplete information, embedded in a universe that changes faster than we can ever fully observe. Intelligence, in this light, is the ability to make useful predictions under constraint.
Most of this work happens quietly. Much of our interaction with the world is handled by specialized processes that operate over limited domains and short timescales. They guide familiar actions, track recurring patterns, and anticipate immediate outcomes without consulting the rest of the system. Walking across a room, catching a falling object, reading a sentence, or maintaining posture rarely requires deliberation.
These local prediction mechanisms handle familiar situations efficiently and without effort. I call them local because they recruit limited areas of the brain, and many can operate simultaneously in parallel. They track patterns, anticipate outcomes, and guide action without drawing attention to themselves. We do not experience them as thinking. They simply work.
If these local mechanisms were always sufficient, nothing more would be needed. But the world does not always cooperate.
Change can be abrupt and sudden. Novelty can quickly become overwhelming, and habitual behaviors may no longer be up to the task. Predictions can conflict with each other or fail to provide useful guidance in the current moment. When this happens, these more local mechanisms are no longer enough. The system needs to escalate. It must coordinate information across broader regions, over longer timescales, and across competing possibilities for what might happen next.
Psychologists sometimes describe this contrast using the language of fast and slow systems, or automatic and deliberative processes. What matters here is not the existence of two distinct systems, but a difference in scale. Most prediction is local, handled by relatively small regions of the brain. Conscious experience emerges when prediction becomes more global.
When local prediction becomes insufficient, the system must conscript a wider range of resources. However, this broader coordination cannot be sustained indefinitely. It is metabolically and computationally demanding, which is why it is rare, brief, and invoked only when necessary.
Global prediction is therefore episodic rather than continuous. This kind of large-scale coordination does not run constantly in the background, but gathers itself in bursts when circumstances demand it. Attention waxes and wanes. Engagement intensifies in the face of uncertainty, then relaxes as predictions become more accurate. Tasks that once required deliberate effort often become automatic with practice, drawing on less and less conscious coordination.
And yet, experience feels continuous.
This apparent continuity is not given directly from experience. It is constructed.
Memory plays a central role here. Discrete episodes of global coordination, which we experience as consciousness, are compressed, stitched together, and smoothed into a coherent account of what happened. Gaps are omitted. What is remembered is not the raw experience, but a narrative that binds separate conscious and semiconscious moments into a single, flowing experience.
This is why it is possible to be conscious without later recalling the experience. Global coordination can occur without recruiting memory mechanisms strongly enough to produce a lasting record. When that happens, experience is vivid in the moment but leaves little behind.
It is also why consciousness is not an on or off state, but comes in gradations. Some episodes draw memory deeply into the coordination, producing rich narrative continuity. Others are brief and fleeting, leaving little trace once they pass.
Examples of constructed experience are everywhere once we know how to look for them.
Consider balance. Our sense of balance feels immediate, unified, and transparent. Yet it is not delivered by any single sensory channel. It emerges from the coordination of vestibular signals from the inner ear, visual cues such as the horizon, and proprioceptive feedback from the body, especially from the legs when standing upright.
When these sources agree, balance feels effortless. When they conflict, as when vertigo disrupts vestibular signals, experience becomes unstable. Attention is drawn inward. Conscious effort is required to regain equilibrium.
There is no privileged signal that defines uprightness. There is only integration, or its failure.
Optical illusions provide an even clearer demonstration. In many cases, what appears to be a uniformly colored surface is assembled from sparse and individually insufficient cues. The visual system extracts edges and contrasts first, then propagates color and brightness across space and time, producing a stable surface percept that was never present in the retinal input at any single instant. We do not experience direct images of the world, but the brain’s best hypothesis about its structure.
The same constructive process operates across time as well. As the body moves, the visual field shifts continuously on the retina. And yet the world is experienced as stable rather than sliding or jittering with each step or turn of the head. Perception adjusts for self-motion, integrating changes in the visual field with signals about movement to maintain the experience of a stationary world.
Certain visual illusions, such as neon color spreading, demonstrate that much of what we see is constructed. The uniform redness of a red surface is mostly inferred. Once this is acknowledged, a further question becomes difficult to avoid. If the uniformity of redness is perceptually constructed, why assume that the experience of redness itself is a primitive given? Why stop the story of construction halfway?
Perceptual qualities are real as experiences. But they are real as outcomes of predictive processes, stabilized by context, learning, and coordination. They are not intrinsic mental atoms delivered directly to awareness. There are no colors in the brain in the way they are experienced; there are only signals that participate in constructing those experiences.
This perspective helps explain why familiar philosophical puzzles about so-called “qualia,” the felt qualities of experience, and inverted color thought experiments are so difficult to resolve. In these thought experiments, two people are imagined to behave identically and use color words in exactly the same way, even though the colors they experience internally are supposed to be swapped.
These puzzles assume that experiences are detachable objects that could, in principle, be compared from the outside or exchanged while everything else remains the same. But perception does not work that way. There is no free-floating redness to invert, only roles played within a predictive system shaped by its history and its constraints.
On this view, consciousness is what global predictive coordination feels like when it occurs. Such coordination draws together resources that would otherwise be operating independently and in parallel. Because this reorganization is metabolically and computationally demanding, it cannot be sustained continuously.
The character of conscious experience depends on which resources are recruited. When memory, imagery, emotion, or self-modeling are drawn into the coordination, experience gains depth, continuity, and significance. When fewer resources are involved, experience may be vivid but brief, leaving little behind.
Nothing in this account denies the reality of experience. What it denies is the idea that experience must be something over and above the activity of a system coping with change. What it denies is the demand that first-person existence be made available from a third-person point of view.
Prediction, coordination, and construction can be studied from the outside. What it is like to undergo that coordination cannot be exported because it is perspectival. It is what it is like to actually be the system doing the predicting.
Seen this way, the deepest questions about consciousness lose their air of paradox. We are persistence machines in a changing world, doing the only thing such machines can do. We anticipate. We adapt. And when local predictions fail, we feel what it is like to bring more of ourselves to bear on an uncertain future.
Conscious experience is not a primitive layer beneath cognition, but a higher-order stabilization that emerges when predictive processes must coordinate across levels to maintain coherence.