Imagination is not a state: it is the human existence itself.
The Imagination Engine
Why imagining and experiencing share the same neural circuits. Generative AI as the first external instrument of imagination.
Chapter 6 — The Imagination Engine
Imagination is not a state: it is the human existence itself.
— William Blake, Annotations to Berkeley (c. 1808)
Tasya bhumishu viniyogah —
the application of [the inner instrument] is in the inner worlds.— Patanjali, Yoga Sutras, 3.6 (c. 400 BCE)
A song before the song
There is a small moment, before I press generate on any track that ends up mattering, where I close my eyes and listen to the song that does not exist yet.
The Lenovo is open. The cursor is in the prompt field. My coffee is going cold. I have a vague specification — late-night, broken piano, a male voice that has been awake too long, cold synth pad, no drums until the second verse. I could just type this and click. Most of the catalog was made that way and most of the catalog is fine.
But on the songs I love — the seventy or eighty out of twelve thousand that I keep going back to — there was always this preparatory minute. I stop typing. I close my eyes. I listen for the track in the silence. Sometimes I can hear it almost completely before I have written a single word into the model. Other times I get only a colour, a tempo, a feel, an emotional shape. Either way, when I open my eyes and write the prompt, I am no longer specifying — I am transcribing.
The track that comes back from the model is then one of two things. It is either close enough to what I heard internally that I can land it in two or three iterations. Or it is far enough away that I know, immediately, the right correction.
The point of this opening is not that I am special. I am not. The point is that I have stumbled into an old practice that contemplative traditions and elite performers have been describing for two thousand years.
The brain that imagines a thing and the brain that perceives a thing are, to a remarkable degree, the same brain.
This chapter is about that.
What imagination actually is
In the everyday English of the last century, imagination is a vague, slightly childish word. It belongs to children's books and motivational posters. It is what you have when you cannot afford a holiday. It is the opposite, somehow, of reality.
This is exactly backwards.
In the technical language that has emerged from cognitive neuroscience over the last forty years, imagination is the central, load-bearing function of the brain. It is the same machinery you use when you remember a face, plan a route, anticipate a sentence, dread a meeting, miss a person, or run a thought experiment. Without imagination there is no memory, no language comprehension, no planning, no empathy.
Imagination is not a side faculty. It is the operating mode.
The neuroscientist Stephen Kosslyn ran the experiments that pinned this down in the 1980s and 1990s.1 He showed, with controlled imagery tasks and PET scans, that visualizing an object activates the visual cortex. Visualizing a movement activates the motor cortex. Imagining a sound activates the auditory cortex. The brain does not have a separate "imagination box" that runs offline simulations. It uses the same regions, in the same patterns, that fire when the thing is actually being seen, done, or heard. Imagination is perception with the input cable rewired inward.
A more recent generation of work — the predictive-processing tradition associated with Andy Clark, Karl Friston, Anil Seth — has gone further.2 On this account, perception itself is a kind of imagination. The brain is constantly running an internal model of the world and of the body, and what we experience as seeing is the brain's best current guess, corrected at the edges by sensory input. Anil Seth calls this the controlled hallucination model. The hallucination is the default; sensory input merely keeps it honest.
If this is right, then imagination is not separate from reality. Imagination is the substrate from which our experience of reality is woven, every waking second.
The mystics arrived at this five thousand years ahead of the scanners. The Vedic tradition speaks of manomayakosha — the "mind-made sheath" of being, the layer at which the world is constituted in inner imagery. The Sufi philosopher Ibn Arabi wrote of khayāl, the imaginal realm, as a third domain neither material nor purely abstract — the meeting place where the soul becomes images and images become real. The Jewish kabbalists worked with the imaginal letters of creation, holding that the world is spoken into being from inner forms.
Different vocabularies. Same observation. The world that you experience is not delivered to you from outside. It is composed, every moment, in the inner studio.
The proof in the muscles
If imagination and perception ran on the same circuitry, you would expect imagined practice to produce real, measurable physical adaptation.
It does.
In 1980, the sports psychologist Richard Suinn published the Visual Motor Behaviour Rehearsal (VMBR) protocol, after running tests on Olympic biathlon athletes.3 Athletes who added vivid mental rehearsal to their physical training showed measurable improvements in performance over those who only physically trained. The effect was real enough that elite skiers, divers, gymnasts, and pistol shooters now treat mental rehearsal as a non-optional component of preparation.
In 1994, Driskell, Copper, and Moran published a meta-analysis of sixty mental practice studies.4 They concluded that mental practice alone produced measurable performance gains, that mental practice combined with physical practice was the most effective protocol, and that the effects were strongest for cognitively complex tasks.
In 2004, Guang Yue and Kelly Cole at the Cleveland Clinic ran a study that has become canonical.5 One group physically trained their finger flexors. Another group only imagined training their finger flexors. A control group did nothing. The physical group gained 53% strength. The imagination group, who lifted nothing physically, gained 35% strength. The control gained 0%.
The pattern is now well-established across literatures. Vivid, repeated, structured imagination of a movement produces real strength gain, real motor learning, real consolidation in the cortex. Not as much as physical practice — but more than half. From thinking. From doing nothing physical at all.
The Tibetan monastic traditions made this an industry centuries before the EMG. Tantric sadhana — the visualization of deities, mandalas, syllables, and entire palaces with thousands of details — is, mechanically, the most demanding mental rehearsal practice ever designed. A senior practitioner can visualize an entire iconographic universe with photographic stability, hold it for an hour, and dissolve it deliberately. The cognitive cost of this is enormous. The reported transformation of the practitioner is also enormous. We now have at least one plausible mechanism: this is the most extreme form of structured mental rehearsal ever practiced, and it produces neural change consistent with the rehearsal literature.6
The point for the reader is small and large at once.
What you imagine vividly, repeatedly, with structure, changes the substrate of who you are.
Composition of place
Long before the laboratories, the contemplative engineers of the mental life had figured out the practical applications.
In the Spiritual Exercises, Ignatius of Loyola taught the composition of place — the practitioner is instructed to construct, in inner imagery, the entire sensory environment of a scene from scripture. The hill, the air, the smell, the faces, the sounds. Not as decoration, but as the active medium of insight. Ignatius understood, four hundred years before the predictive-processing literature, that the mind that has constructed a place in vivid detail is changed by being there.
The Jesuits used this practice to produce some of the most disciplined operators in history. Diplomats, astronomers, missionaries, founders of universities. The training was a training of the imagination, and the imagination produced the persons.
In Vajrayana Buddhism, deity yoga consists of constructing the deity, in full iconographic precision, as a stable inner image — and then, crucially, recognizing oneself as that image. The mechanic is identical to the modern athletic literature on imagined practice, scaled up to the level of identity. You imagine yourself, repeatedly, vividly, as a being with specific qualities of compassion, clarity, fearlessness. After enough repetitions, the substrate adapts.
In the Stoic tradition, Marcus Aurelius writes — between governing the empire — imagine yourself, as the day begins, the kind of man you would wish to be that evening. He did not have the EMG study. He had the result.
The instruction across these traditions is not "daydream." The instruction is severe. It is structured imagination, repeated daily, of the desired state — physical, ethical, spiritual — until the substrate moves.
Modern self-help has watered this down to the diluted notion of visualization. The dilution has obscured the technology. The actual practice — what Suinn ran on the slopes, what Ignatius wrote on Manresa, what the lineage holders run for forty years — is something more demanding and more reliable than any motivational poster.
What AI changes
For most of human history, the imagination has been a private organ. You could describe what you imagined. You could draw it, badly, on paper. You could compose it, slowly, into music or sculpture or fiction, over months and years. But the link from inner image to outer artefact was, in every case, an act of transcription — an arduous, lossy, slow process of converting interior simulation into a form another mind could see.
Generative AI is the first external instrument that operates at the speed of the imagination itself.
This is not a marketing claim. It is a structural observation about the cognitive economics of the new tools. When a writer with a clear inner sentence can have it on the page in two seconds, the constraint has moved. When a musician with a clear inner song can hear it back from a model in twelve seconds, the constraint has moved. When a designer with a clear inner image can see it rendered in eight seconds, the constraint has moved.
The constraint, in every case, has moved from the speed of execution to the clarity of the imagination.
This is the most under-discussed consequence of the model revolution.
Before the models, you could be a person of vivid inner experience and produce, in a lifetime, a small number of artefacts. A poet writes one good book a decade if they are exceptional. A composer, if disciplined, produces a few hundred completed works in a career. Most of what passes through any given imagination, in any given life, is lost. The ratio of inner-events-experienced to outer-artefacts-produced is one of the silent tragedies of the human condition.
The models change the ratio.
I have made twelve thousand songs in three years. Most are not good. I want to be honest about that. But twelve thousand is a number that the entire history of recorded music could not have offered to a single individual until 2023. Before the models, twelve thousand songs from one person was inconceivable in the way that running a mile in three minutes is inconceivable. After the models, the bottleneck is not output — it is whether the operator has anything to say.
This is the part that most commentary misses.
The models are not a threat to the inner life. They are its external organ. For someone with a developed inner imagination, the models are a tool of unprecedented expressive bandwidth. For someone without one, the models are an empty mirror that returns generic outputs and confirms the suspicion that AI is hollow.
The discriminating variable is whether the operator has cultivated the inner instrument.
A generation that took its imagination seriously — read the long books, sat in silence, learned the contemplative protocols, paid attention to what their dreams were composing while they slept — will use these tools to produce work the world has not seen. A generation that did not will use the same tools to produce a flood of indistinguishable outputs that are, technically, the average of the internet.
Same model. Different operator. Different futures.
A small physics of the prompt
If imagination is an organ, and the model is its external instrument, then the prompt is the cable between them. And every cable has a transmission characteristic.
A vague prompt — write a song about love — gives the model nothing to work with. It returns the average. The model is large enough that the average is technically competent, which is why most generative outputs feel both fluent and dead.
A specific prompt — write a song about the moment, after a long argument, when neither of you has spoken for ten minutes and the room has gone soft, and one of you, almost without deciding to, reaches across and the other does not pull away — gives the model a concrete inner scene to render. The model is large enough to render this scene with care, because the scene is specified.
The technical name for this variable, in the prompt-engineering literature, is information density. The functional name is the writer's eye.
The same skill that makes a good novelist — the discipline of seeing the specific gesture, the specific light, the specific small sound — is the skill that makes a good prompter. Ten years of writing journals, sitting in cafés watching strangers, reading Chekhov — this turns out to be the most direct training for the new instrument. The literary tradition was never about literature. It was about training the inner instrument until it could see clearly.
The Golden Age of Intelligence is going to be experienced very differently by people who have done this work and people who haven't.
The work is available to everyone. It always has been. It costs nothing except attention. But it does not happen by accident, and the models do not create it for you — they amplify what is already there.
This is the bargain.
A practice for this week
Pick a single, concrete object that you can see right now from where you are sitting. Not a picture, not a memory — a physical object. A cup, a pen, a window frame.
Spend three minutes looking at it. Don't write yet. Don't photograph it. Just look. Notice the small things that you would otherwise miss — the chip, the colour at the edge, the way light falls on the curvature.
Then close your eyes and reconstruct it inwardly. Hold the inner image for as long as you can. When it fades, open your eyes and look again. Repeat three times.
On day three of doing this, write a single paragraph describing the object as if to someone who has never seen it. On day seven, generate a model image from your paragraph. Compare what came back to your inner image.
The exercise is simple, and the reason it works is the entire chapter. You are training the inner instrument. Once the instrument is sharper, the model becomes, for the first time, a real partner.
This is the apprenticeship of the imagination engine.
It is also, structurally, what every painter has done since the first cave.
Hand-off
The imagination engine is the first half of what you are.
The second half — the half that consolidates what you imagine into who you become — happens at night, while you are not looking.
In Chapter 7 we open the room of memory and sleep. The hippocampal replay that turns a day's imagination into a year's character. The reason you wake with the answer you could not find at midnight. What the new generation of AI systems is borrowing, knowingly and unknowingly, from the architecture of the dreaming brain.
The room is darker than the studio. But it is in this room that the work of the studio becomes you.
We are walking in.
Footnotes
- Stephen M. Kosslyn, Image and Brain: The Resolution of the Imagery Debate, MIT Press, 1994. The foundational empirical work establishing that mental imagery activates the same cortical regions as perception. ↩
- Andy Clark, Surfing Uncertainty: Prediction, Action, and the Embodied Mind, Oxford, 2015; Anil Seth, Being You: A New Science of Consciousness, Faber, 2021. The predictive-processing tradition has become the dominant integrative framework in current cognitive science. ↩
- Richard M. Suinn, "Body Thinking: Psychology for Olympic Champs," in Psychology in Sports: Methods and Applications, ed. Suinn, Burgess, 1980. ↩
- James E. Driskell, Carolyn Copper, and Aidan Moran, "Does mental practice enhance performance?" Journal of Applied Psychology 79(4): 481–492, 1994. ↩
- Vinoth K. Ranganathan, Vlodek Siemionow, Jing Z. Liu, Vinod Sahgal, and Guang H. Yue, "From mental power to muscle power — gaining strength by using the mind," Neuropsychologia 42(7): 944–956, 2004. The 35%-from-imagination-only finding is from this paper. ↩
- Antoine Lutz et al., PNAS 2004, also cited in Chapter 5; and the broader meditator-imaging literature surveyed in Goleman & Davidson, Altered Traits, 2017. The case for tantric visualization as extreme structured rehearsal is mine; the experimental data on senior practitioners' cortical organization is theirs. ↩