At the age of four, I steered a push cart into a ditch and nearly drowned. My brother, only a few years older and doing all of the pushing, ran for help, leaving me flailing and gulping water, grasping at the tenuous clump of grass at the water's edge that saved my skin.
I didn't reach for the clump of grass. I didn't even see it. My body did what bodies do when panic takes over: it flailed and grasped indiscriminately. Flailing is behavioural variability pressed into service as a last resort, and it is far older than any of us. When conditions worsen, our least favourite stomach bug E. coli stops swimming in straight lines and starts tumbling at random, not that we should feel sorry for its misfortune. A pursued hare zigzags. An animal whose learned routine suddenly stops working begins to vary its behaviour before it gives up. Random behaviour of this sort works because it tests the surroundings quickly and widely, and somewhere in the scatter is sometimes a lifeline, a clump of grass, a foothold, another's hand. Enough of our ancestors survived this way that the disposition to flail was passed on. It is neither a skill nor a mistake, since there is no aim for it to miss. It is what life does when the threat or predicament is overwhelming.
I've been thinking about that watery ditch since attending a philosophy seminar on illusionism and artificial consciousness, and I'll come back to the seminar. But first, a drawing class.
Error in the form of a line
The day before the seminar, two colleagues and I were teaching drawing to a large group of second-year Fine Art students. What strikes me repeatedly is how rarely students consider the quality of the lines they make. Their criterion of success is accuracy, usually meaning something close to what a camera would record. I can't claim to have been any different as a student. The weight, cadence, sensitivity and refinement of a line, and its fitness for what it describes: all of that barely figures when accuracy is pursued so exclusively. We have to keep insisting on the quality of the line, and one worry often surfaces: the wavering line, whose shaky imprecision students tend to treat as a weakness. I sometimes find myself saying: you're not cameras, you're not robots, you're human, and your imperfections are part of what makes your drawings yours. Accuracy is often important, but a good drawing makes clear that its wavering is a feature, not a bug: an opportunity for distinctiveness and disciplined control. In many ways cameras free artists from the burden of record-making and open a space for other qualities of drawing to be explored and valued. Odd, then perhaps, that so many students still measure their drawings against photographs.
Drawing from observation balances accuracy on one side against something much closer to handwriting on the other. The tremors, hesitations and small corrections of a living hand are shaped by skill, never entirely removed. A good line carries a record of its own making: a journey of adjustment and accommodation, open to the vicissitudes of chance. The tradition has a name for the most honest version of this. Pentimenti are the corrections left visible, like the searching lines in Giacometti or the ghost limbs in Rembrandt and Degas, kept because the artist saw that they belonged, or at least that it did no harm to leave them.
Artists tend to distrust the word "style," perhaps for the same reasons as Charles Rennie Mackintosh, who famously penned the above warning: "There is hope in honest error, none in the icy perfections of the mere stylist." Yet we all recognise a signature style in the work of others, and we often value it. What matters is where it comes from. A signature grows out of a particular body's variability, disciplined over years of practice but also inescapably unique and steadfast (in a way that isn't always welcome). It turns up in the work whether you put your name to it or not. It is partly beyond us, and that is what makes its honesty real. The mere stylist applies the look of a signature from outside and can take it off again, like someone who switches accents to suit the company, unaware of how suspicious that can appear. Software that adds a hand-drawn wobble to a vector line is the modern example: the appearance of tremor with no tremor behind it. Icy error.
Kinds of error
Mackintosh's word "honest" perhaps implies a genealogy of error.
Flailing comes before error altogether: inherited variability under threat, with no aim to miss.
Honest error is a skilled attempt that goes wrong in a way that can be caught and learned from. Joseph Priestley isolated oxygen while calling it "dephlogisticated air." He made an important discovery inside a mistaken theory, and the experiments that theory prompted eventually exposed it. Students make honest errors every day, and most of what they learn comes from how they negotiate these challenges.
Harder to see is what I'd call captive error, or perhaps misfounded error, since the fault lies in the foundations and not in the work built on them. It is the error of skilled, sincere people working inside a technique they have come to take for nature. Theories of perception have long treated lens projection, whether in the camera obscura, the retina or the photograph, as settled evidence of what we see. From there it was a short step to the prejudice, repeated by Piaget and countless teachers since, that children and the members of ancient or isolated cultures draw "what they know, not what they see." We don't see in two dimensions. Translating what we see onto a flat surface is one of the hardest things an artist ever learns, and the painters who developed perspective knew this better than anyone.
Finally there is icy error, Mackintosh's target: error polished until nothing can catch on it.
Science and art
The analogy with science and philosophy is less straightforward. Roughly, science seeks to show what holds in spite of us, whereas art seeks to show us, in all our complexity. Philosophy sits awkwardly between the two. It concerns itself with science, but also with ethics, art and how to live, where we ourselves are very much the subject. And some philosophers, J. L. Austin and Gilbert Ryle among them, have a voice you would recognise anywhere, with a dry wit that reminds me of the television comedy of my childhood, Ronnie Barker's especially. That isn't mere stylish decoration. Finding the right form for a thought is part of thinking it, and a well-placed joke about ordinary words can expose a confusion faster than a page of argument. So I don't think philosophers should write like machines. But whatever the subject, their first duty is to clarity (Austin and Ryle would agree I'm sure), and where they deal with science that duty is at its most exacting, because what is at stake is whether our concepts fit what holds in spite of us. Lyricism can illuminate an ethical question. In a theory of the brain it can only mislead. Once a matter is settled, loose talk does little harm. We still speak of the sun rising, and nobody is misled. But where the foundations are the very thing in question, loose talk is the hazard, the watery ditch. Using a word like "representation" as promiscuously as the field does means building on exactly the ground that most needs careful surveying.
None of this means that scientists and philosophers should never speculate naively, flail around for a foothold or grasp blindly in the darkness. In new territory, searching and floundering are exactly right until some clumps of grass can be grasped, just as the first marks of a drawing are properly a tracery of tentative lines. But once you've learned at least to tread water, you're better off trying to swim, even if falling back on wild flailing sometimes comes up with the goods in a pinch. The risk of drowning is too high, especially when you have a genuinely better way not only to survive but perhaps to thrive.
The seminar
This brings me back to the seminar, where I heard Joe Cruz give a thoughtful and open-hearted defence of illusionism: the view that consciousness is a kind of illusion produced by our inability to represent our own fine-grained processing. It came with the suggestion that machines might one day be conscious in just the same way. I disagreed with nearly all of it, but the error, if I'm right, is captive or misfounded error, not dishonest error, and that is what interests me here.
Mainstream cognitive science and philosophy of mind work inside the picture of representation. They take it for granted that brains encode, infer, predict and represent. Neurons carry information. These are metaphors borrowed from public practices of picturing, writing and signalling, and read literally. In a sense, this is lyricism on the scientific line, style intruding where precision would be more appropriate and is most needed. Behaviourism at least attributed thinking and feeling to whole organisms, not to their parts, their organs, synapses and cortices. What replaced it moved the mind back inside and installed it in the parts. The ghost in the machine wasn't exorcised. It was downsized and multiplied, until the brain became a crowd of little ghosts: neurons that signal, modules that infer, systems that represent. To philosophers and thinkers of my bent, that is a profound and unwarranted mistake.
Physicalism came up several times in the talk and afterwards in the discussion. Contemporary philosophers are clearly anxious to keep woo out of their theories. I wish they could see that representationalism is the new woo. Those little ghosts need a better account than representationalism can give them. Nobody thinks that imagining dancing requires an inner dance, so there is no reason to suppose that representing requires inner representations. The absurdity doesn't vanish because the components are modelled as ghostly little abstractions. Had I counted how often the word "representation" was used during the talk, the tally would have looked less like a technical term than a tic, and obsession is the tell of severe error in any domain.
To be fair, the talk invited us to be doubtful about its conclusions, and I didn't hold back in communicating the view that representation is an activity that only has a place in the negotiation between perceivers, not in their inner workings. When I asked what Peter Hacker would make of the Hard Problem of consciousness, Cruz candidly admitted that it had been a long time since he'd read any Hacker. The candour does him credit, but it left the view badly exposed. Hacker's work is the most sustained critique of exactly the presuppositions the talk relied on, and no amount of declared doubt makes up for passing over your strongest critics.
I also pointed out that philosophy has no consensus on what representations are, and Cruz agreed. That is a remarkable admission from a view that finds representations everywhere, in cortices, in babies and even in cows. Yes, that's moos with woos. Science doesn't proceed by ignoring prior insight, yet contemporary philosophy regularly treats hard-won insights from within its own domain as though they were theories, as interchangeable as Lego bricks. That's a genuine problem when the question of what representations are gets treated as either too obvious (they're just stand-ins), too basic (we have bigger fish to fry) or too pernickety to deserve attention, with all that fuss about transparency, illusion, resemblance, likeness, symbolisation, normativity, ofness and aboutness. When he tried to say what he meant, Cruz reached for "aboutness" and "reference." But these are the categories of words and symbols, and they fit pictures badly. A photograph doesn't depend on reference to be of anything, and what it is of is distinct from what it is about. We can argue until the moos come home about what a photograph is about, but what it is of usually needs no interpretation at all. An account of representation built from the vocabulary of language can't get the photograph right, let alone the cow.
I know I must seem arrogant in rooms like that. The philosophers there freely admit that their theories are theories, and I suspect they take me to be in the grip of a theory of my own. What I have is closer to an observation, something I can show and practice every day: someone making or using one thing to stand in for another, for someone else to see or sense. Donald Brook's account sets this out with far more care than I can here, but his three modes of matching, simulating and symbolising provide a far more firm starting point than reference and aboutness. Wittgenstein thought that philosophy should offer reminders of what we already know, not theories, so that if it ever advanced theses, they would be beyond debate because everyone would agree with them (PI §128). That is the kind of certainty I mean, the ordinary certainty with which any of us can say what a map, a portrait or a pointing finger is. If physicalism is to stand in good stead, philosophy first needs to agree on what representations are, and what they are not.
It was telling that one of the philosophers present (FM) remarked in the discussion that the talk had dented her low opinion of illusionism a little, mostly because it was so nicely presented. Style persuaded where the concepts didn't.
And the AI machines? Whatever else they do, they neither flail nor waver. A system with nothing at stake and no inheritance of desperation has no survival instinct to override its programming when it's thrown in the ditch. It goes on producing fluent, well-formed output, and its errors arrive with the same polish as its successes. It may even stumble on a happy accident, but it can't recognise one as a gift, because it has no purposes of its own for the accident to serve. Seeing the mistake worth keeping is our instinct and where creativity earns its keep.
Cruz's suggestion was that AI perhaps need only be embodied to be conscious. That's just one example of how much weight representation is made to bear. My view is far more demanding and I'm sure I'm not alone in this. The flailing in the ditch was an inheritance, handed down by countless ancestors for whom it sometimes made the difference, and consciousness belongs to creatures with that kind of history: members of a species, with something at stake. Programming artificial instincts into an embodied AI, while no doubt easily done, would be icy perfection, not raw embodiment, nor the most urgent expression of what it is to be a living creature. Tools are refined by their makers; they don't evolve on their own or their species' behalf. A robot body would give AI somewhere to be, but no lineage to be heir to and no ecological adaptability honed by eons of the most uncompromising Darwinian selection.
Wittgenstein wrote in the Tractatus (6.52): "We feel that even when all possible scientific questions have been answered, the problems of life remain completely untouched. Of course there are then no questions left, and this itself is the answer." The so-called "hard problem of consciousness" reads this backwards. It treats what is left untouched as a further scientific question awaiting an answer. What is untouched belongs to the drawing, the clump of grass, and the living line, and there is hope in the honest errors we make there.
The seminar referred to above: Joe Cruz (Williams College, Philosophy and Cognitive Science), "Illusionism and Artificial Consciousness," Philosophy, Psychology, Neuroscience research seminar, Reid Seminar Room, Philosophy, University of Glasgow, 25 September 2026.
