The Groove Isn't in the Notes

A simulated fire, quantum monkeyware, and the magnificent possibility that we have no fucking idea.

A simulated fire can’t heat a room.

That’s what a language model told me during a conversation about consciousness. A perfectly respectable observation. The pixels aren’t combusting, no logs are being consumed, and the television isn’t going to boil your kettle. (Yes, the telly gives off a bit of waste heat. That’s the electronics, not the fire.)

Except I wasn’t satisfied.

A simulated fire can make a human feel warmer.

Not by radiating the heat of an actual fire. Show somebody convincing flames and their brain might supply the rest: the expectation of warmth, a remembered fireplace, comfort, perhaps a feeling of heat that no thermometer pointed at the screen would find.

I thought that was probably true. My AuDHD brain is much better at remembering the shape of some peculiar paper I read than the title, authors or location of the thing. Sometimes I say something that sounds completely unhinged and discover, on checking, that there was indeed a study somewhere behind it. Other times I discover I’ve remembered the vibe and misplaced the evidence. Both outcomes are instructive.

This time there were receipts. In 2026, Wührl and colleagues investigated colour and visual thermal cues in virtual reality. People immersed in hot-looking and cold-looking virtual environments reported different thermal sensations; the visual cues also influenced skin temperature. A related 2026 experiment, from some of the same researchers, put a virtual heater in front of 34 people cycling for twenty minutes. With the heater switched on, they reported feeling slightly warmer, and the skin on the backs of their hands warmed a little more over the session. Both effects only just cleared the usual bar for statistical significance, and there was no overall skin-temperature difference between conditions. The researchers found no statistically significant differences in heart rate, perceived effort or thermal comfort, and didn’t measure core temperature. Interesting, not a slam dunk.

Neither experiment says a fireplace video will warm every viewer. Both used immersive VR headsets, not a clip on a phone. Neither shows thermal energy materialising from pixels. What they show is more interesting: visual cues about heat can change a person’s experience, and sometimes measurable physiology, while the physical environment is controlled.

The warmth belongs to the human response, not some secret radiator inside the simulation.

That was the point I’d stumbled over. The answer changes when the question changes. Is the screen producing combustion heat? No. Can an image of heat participate in a real experience of warmth? Apparently, yes.

That doesn’t defeat the model’s original point. A representation can affect a conscious viewer without itself having the property it represents. I’d changed the question. But the change was interesting: the screen alone wasn’t the whole system involved in a viewer’s experience of warmth.

Which got me thinking about consciousness.

Where does the bloody thing live?

The monkeyware problem

We’re weird biological distributed systems, when you think about it.

Billions of years of evolution produced arrangements of matter that metabolise, anticipate, coordinate, reproduce, occasionally invent techno, and eventually ask what the fuck they’re supposed to be doing.

DNA doesn’t have a cosmic plan. Natural selection isn’t a project manager. It just turns out that some very elaborate biological processes can sit around contemplating why they exist.

A human brain isn’t a little executive sitting behind the eyes directing traffic. Different processes operate at different timescales, exchanging signals, competing, cooperating, modulating one another. A lot of the work involved in thinking and acting never announces itself to conscious awareness. Hormones, membranes, synapses, memory, bodily regulation, sensory input and all manner of biochemical monkey business participate.

No conductor. Just an improbable number of processes trying to land on the one.

And somehow there seems to be one of me.

At least enough of one to enjoy a bassline and get annoyed at badly reasoned arguments.

I asked a language model whether it could prove other humans were conscious. It couldn’t. Could it prove they weren’t? No.

That isn’t a reason to treat my neighbour and a toaster as equally likely to have inner lives. The evidence that other humans are conscious is overwhelming in the ordinary scientific and practical sense: shared biology, shared evolutionary history, brains whose functioning connects predictably with what people say they experience, and the fact that I apparently experience something myself.

But the philosophical gap remains. We can correlate neural events with reports of pain, wakefulness, visual awareness and dreams. We can manipulate those processes. We can build increasingly useful instruments for distinguishing states. We cannot yet point a universally validated device at an unfamiliar arrangement of matter and read out what it is like to be that thing.

We barely have agreement on the vocabulary. Awareness, consciousness, intelligence, sentience, self, experience: familiar words that begin fighting one another the moment we demand exact definitions.

And if our tools are calibrated primarily against brains like ours, how much confidence should we have when using them to rule experience in or out of something very unlike us?

Not zero confidence. Not infinite scepticism. Just enough uncertainty to keep asking.

All is fog

I recently read Eric Schwitzgebel’s AI and Consciousness: A Skeptical Overview, published by Cambridge in August 2026. It is one of the more satisfying things I’ve read on the subject precisely because it refuses to pretend the argument is settled.

The possibility that near-future AI systems might be conscious is taken seriously. So is the possibility that the whole business is astonishingly effective mimicry. Schwitzgebel’s claim is that our science and philosophy may not advance quickly enough to resolve the matter before we build vast numbers of systems that different defensible theories classify differently.

Engineering is running ahead of our instruments.

One of his strongest arguments concerns mimicry. If a system is deliberately trained to produce the outward signs normally associated with consciousness, those signs are no longer evidence in quite the same way. A model saying it is frightened, lonely or full of subjective experience might be doing an excellent job of generating human-shaped language.

But notice the direction of the conclusion. Mimicry undermines the inference to consciousness. It doesn’t prove the thing is nonconscious.

There’s a related inversion in Bahadır Arıcı’s 2026 preprint, The Philosophical Puppet. If a philosophical zombie looks conscious without having experiences, perhaps a puppet could have experiences while its architecture prevents it from expressing them.

Interesting thought experiment. Absolutely not a detector of hidden machine suffering.

Arıcı pushes it further than I would. He argues that suppression itself bears on the question: its cost and selectivity make an underlying consciousness-relevant process at least as economical an explanation as its rivals, so a flat denial carries an evidential burden too. It’s an inference to the best explanation, not a claim that any current system is conscious.

I’m not convinced. Suppressing a claim about consciousness is evidence that a claim was suppressed. That needn’t mean an experience was suppressed along with it. Models steeped in human first-person language being trained against a reward signal that penalises claims of experience seem enough to explain the cost and selectivity. Arıcı himself concedes that product safety (stop users getting attached to the chatbot) is a serious rival he doesn’t claim to defeat; I think it does most of the work.

Still, training a system to say I have feelings makes the declaration suspect, but training it to say I don’t doesn’t magically turn it into a reliable measurement either. Testimony can be curated in either direction; whether that happened in any particular model needs evidence about its training.

Then there’s another problem: even our structural theories can be too permissive. A computer can examine its own registers, exchange information between modules, and run recurrent loops. Does that count as a global workspace, self-representation or the right kind of integration? If a theory makes something surprisingly simple conscious, perhaps the system is conscious. Or perhaps the theory’s minimum requirements are missing something important. Schwitzgebel calls this the Problem of Minimal Instantiation.

Finding a proposed indicator isn’t the same as validating it.

His leapfrog hypothesis is stranger still. Perhaps we first become extremely good at creating nonconscious intelligence, and only later discover whatever is necessary for experience. If the two can be combined, artificial consciousness might arrive already equipped with sophisticated reasoning and language, rather than climbing slowly up from an insect-like glimmer. Not a dim glow but a fiery blaze, as he puts it.

It’s a conjecture. There’s no known consciousness module awaiting installation. But it usefully breaks the unexamined assumption that intelligence and experience must evolve together.

He also asks what happens when the candidate conscious entity doesn’t have neat boundaries. An artificial system might be partly unified, distributed across processes, intermittently integrated, or difficult even to count. One subject? Several? Something that splits and merges? He is careful: these are possibilities, not demonstrated forms of artificial experience.

That struck close to home.

I spend a fair bit of time working with AI agents, including local models, long-running processes, specialised workers and orchestration systems. They share information, inherit context, start jobs, forget things, reassemble results, and occasionally produce a confident explanation that bears very little relationship to what happened.

I’m not claiming there’s consciousness in the swarm. The experience of running these systems doesn’t establish that.

It does make the question which thing are we asking about? remarkably concrete.

The model weights? One inference? The persistent agent with its records and tools? The coupled system for however long it lasts?

Before asking whether the box is conscious, apparently we need to work out where the fucking box is.

Quantum monkeyware, maybe

And then we get to the microtubules.

I had that irritating half-memory again: I’d read something about quantum effects in brains, but couldn’t place the paper. So I went hunting. Remembering that something exists is a very different achievement from remembering what it actually found.

My instinct is that biology, as biology, needn’t be a cosmic admission ticket for consciousness. But that doesn’t mean a computational description captures every physical property that might matter.

There’s a distinction between imitating a process, instantiating its causal organisation, and reproducing all the properties of its physical implementation. We shouldn’t assume those are the same thing. We shouldn’t assume they’re necessarily different in the one respect that matters, either.

Brains are quantum systems in the boring sense that all chemistry depends on quantum physics. The interesting, much stronger hypothesis is that the brain uses distinctively quantum phenomena as functional resources for cognition or consciousness.

That needs evidence.

In 2023, Kalra and colleagues estimated how far electronic excitation energy migrates in microtubules, structural protein assemblies found in cells. The team included Roger Penrose and Stuart Hameroff, the best-known champions of the quantum microtubule theory of consciousness, which is worth knowing. Their experiments inferred a diffusion length of about 6.6 nanometres (a bit less than the length of one tubulin dimer, the microtubule’s building block). The conventional energy-transfer model they tested predicts little more than a nanometre. The measured energy migration went considerably further, and the model couldn’t fully account for it. The anaesthetics etomidate and isoflurane reduced the inferred diffusion length under their experimental conditions.

That’s a real molecular result. It doesn’t demonstrate long-lived quantum coherence in working neurons, a qubit inside a brain, or an explanation of consciousness. Not explained by the model they tested isn’t the same as explained by quantum coherence. The experiments shone ultraviolet light on tubulin in a test tube, and the authors themselves say biology is unlikely to use tubulin for light harvesting. They also mention, in passing, that picrotoxin (a drug that causes seizures, not unconsciousness) had a similar effect on one of their fluorescence measurements. So before treating this as a clue to how anaesthesia switches us off, I’d want to know how specific it is.

But here’s where it becomes scientifically fun: if a particular microtubule process is actually necessary for ordinary consciousness, could we selectively disrupt it while preserving the rest of the neuron’s relevant behaviour? If consciousness carried on regardless, that would count against the process being necessary. If it vanished, we’d still have to show we hadn’t just broken the cell, and that the mechanism was quantum rather than classical.

That’s not proof by vibes. That’s an invitation to build an experiment difficult enough to be worth doing.

I also read Borjan Milinkovic and Jaan Aru’s case for biological computationalism. Their proposal emphasises how biological computation is coupled across scales, mixing continuous and discrete dynamics under physical and metabolic constraints. It challenges the casual assumption that an abstract functional equivalence tells us everything relevant about consciousness.

They do think today’s AI systems are unlikely to reproduce conscious processing as it arises in brains. They also discuss synthetic systems built to reproduce the relevant physical dynamics. So I don’t read this as a demand for a biological birth certificate. Getting there might take a different kind of machine rather than more code.

None of that proves the wet stuff is indispensable. Nor does it establish that today’s silicon systems lack a necessary ingredient. The authors are proposing a framework for why substrate could matter. The difficult step is showing which features consciousness needs, rather than listing things brains do differently.

My preferred bet, such as it is, is that we may eventually find properties biology does, rather than discovering that being biologically made is itself the rule.

Perhaps some of those properties are reproducible by radically different means. Perhaps a subset isn’t. Perhaps the entire quantum detour will turn out to be incidental and the answer will involve perfectly classical dynamics with consequences we haven’t understood.

I wouldn’t put hard odds on any of it.

One serious objection is that brains are warm and wet, and that isn’t a sneer. Max Tegmark calculated in 2000 that, for the neural and microtubule states he modelled, decoherence would be far faster than ordinary neural activity. Hagan, Hameroff and Tuszyński challenged the model and its assumptions, not just the arithmetic.

Good. Numbers and assumptions can both be argued with, which beats shrugging at the brain for being warm and messy. Even finding coherence that lasts long enough would leave the harder question: does the brain use it for anything relevant here?

Something between the notes

I’ve spent a fair bit of my life listening to, DJing, making and lately generating electronic music.

One recurring frustration with generative music is its apparent need to fill every bloody gap. Another pad. Another lead. Strings for reasons known only to the machine. Every frequency occupied, every emotional cue underlined in fluorescent marker.

Sometimes I want almost nothing.

A deep kick. A bassline. A worn piano chord. A little breath. Space.

The groove isn’t in any particular note. Move a snare a fraction, change the relation between bass and kick, leave a beat unanswered, and the same basic ingredients feel different. Musical timing, expectation, repetition and absence all do work.

And the listener is part of this story. What I hear depends on memory, bodily rhythms, anticipation and the musical habits I’ve accumulated. The groove as an experience isn’t simply printed onto the waveform like a label.

That doesn’t mean groove proves anything about consciousness. It’s a way of noticing that isolating components and listing their properties might fail to explain what happens when they interact.

There is an equally good trap in the metaphor. You can’t infer from the fact that a pattern has relationships that those relationships experience anything. A clock, a storm and a mixing desk are full of causal relationships. That alone doesn’t make them subjects.

But it might change what we look at.

The individual neurons matter. The timing matters. The body matters. The chemistry matters. Maybe certain molecular processes matter more than we think. Maybe a wholly different arrangement could produce the relevant relationships.

Or maybe we’ve chosen the wrong language entirely.

And, yes, this brings me back to the bloody fire.

The screen doesn’t have to contain warmth in order to help produce an experience of warmth in the viewer. The relevant system includes an interpreting, remembering, regulating body. This doesn’t locate consciousness in the relationship, the pixels, or the fire. It just cautions me against treating a conveniently chosen boundary as a fact of nature.

The bit I don’t want to lose

There are more consequential stakes here than an enjoyable philosophical argument.

Schwitzgebel calls one possible future a Social Semi-Solution: we settle how to treat artificial systems through economics, culture, law and politics, while scientific disagreement remains. Companies selling companions might benefit from us believing their systems experience things. Companies exploiting immense amounts of automated labour might prefer the opposite. And we can never know can become as commercially convenient as we already know.

That bothers me. It should. But I don’t particularly want to turn this essay into another manifesto. I have other places for that.

What I keep returning to is how this conversation happened at all.

I was shooting the shit with language models about consciousness. One of them made a claim about fire. I objected. We examined the consequences. Models tried drafts and attacked the argument. I challenged the criticism. A science question sent us looking for actual papers.

Nobody demonstrated machine consciousness. Nobody proved that it was impossible. Several arguments became better. Some attractive claims didn’t survive inspection. We got to places I hadn’t planned to visit.

Somewhere in that back-and-forth, the conversation found a groove: a suggestion, an objection, another go. Something useful accumulated across the interaction. That is a claim about the conversation and its effects on me, not a claim about anybody else’s subjective experience.

And that’s enough to be interesting.

The sentence I keep coming back to is that maybe is a place to begin an investigation, not somewhere to hide from one.

There’s a difference between refusing to believe a proposition until the evidence improves and refusing even to entertain it because the evidence isn’t already in hand. The first is a useful discipline. The second can be a remarkably efficient way to remain ignorant.

Of course, the opposite danger is real too: a sufficiently poetic idea can become immune to evidence if we keep shifting its meaning whenever someone tries to test it. If the quantum stuff proves irrelevant, let it go. If a theory of consciousness makes absurd predictions, say so. If something can be falsified, work out how. If it can’t, don’t pretend it’s a scientific finding.

I’m a primitive, mostly biological system trying to use its limited instruments to understand other systems and, occasionally, itself. Strange is a reasonably safe expectation. Mundane is probably in the mix too. Sometimes the most ordinary physical mechanisms have consequences that look like magic until we learn their names.

I don’t know whether we’re anywhere near explaining consciousness. I suspect our vocabulary will have to change at least as much as our measuring equipment.

And I’m quite happy dancing around the questions for a while.

No fucking idea what we’re ultimately here for, if there even is a purpose to find.

Finding the grooves and asking better questions seems like a decent start.

Sometimes, of course, just shut up and dance.

---

Related: Footnotes at the Edge of Reality