Qualia Might Be an Accident
Huxley's steam whistle, Gould's spandrels, and a literature I did not know I was standing in.
Somewhere in the middle of writing the first page of the sequence I made an observation and then walked straight past it, because it was not what that page was about. The observation was that none of this appears to be necessary.
You can tell a complete functional story about a person. Nociception, aversion, learning, planning, self-modeling, report. The physics closes. The biology closes. The behavior is fully accounted for at every step, and at no point does the account require that any of it be like something. The lights are on anyway.
The thought I walked past is the obvious next one: maybe that is because the lights are an accident. Not a solution to any problem. Not a feature that was selected for. Just something that happened, once, in one lineage, as a side effect of building a brain big enough to do other things — and that persists for the same reason a lot of biology persists, which is that nothing in particular was pushing back.
I assumed this was a private weirdness. It is not. I went looking and found I had wandered into a fairly crowded room, which is the usual outcome of having a thought about consciousness.
Huxley’s steam whistle
The oldest version is epiphenomenalism , and its canonical statement is Thomas Henry Huxley’s 1874 essay “On the Hypothesis that Animals are Automata, and its History.” Huxley’s position was that animals — and, he implies without much hedging, us — are conscious automata. We have an inner life. It just does not do anything. States of consciousness are caused by molecular changes in the brain, and there is, Huxley says, no proof that any state of consciousness is itself the cause of any change in the motion of the matter of the organism.
The image he reached for is the one everybody has borrowed since: consciousness stands to the working of the body roughly as the steam whistle stands to the work of a locomotive. The whistle is real. It is caused by the engine. It is audible, and it is distinctive, and if you were unfamiliar with locomotives you might reasonably assume it was somehow driving the thing. It contributes nothing to the motion.
I like this more than I expected to, partly because it rhymes with an image I was already using. I have described the human brain as demoscene hardware — a machine doing something remarkable that it has no business doing. The steam whistle is the same observation from the other end. Something is happening here that the specification did not call for.
Why the obvious objection is a good one
The standard reply arrived fast, and it is an evolutionary argument. William James made the classical version in “Are We Automata?” in Mind in 1879, five years after Huxley, and it goes like this: consciousness is expensive, widespread, and elaborately organized. Pleasure attends the things that keep us alive and pain attends the things that kill us, with a fit too good to be coincidence. Natural selection does not build and maintain expensive machinery that does nothing. If experience had no effects, it could not have been shaped by selection, and yet it looks shaped.
This is a real objection and I do not want to wave at it. For most of a century it was treated as decisive, and if the only available position were Huxley’s strict one — experience exists and is causally inert — I think it would come close to being decisive.
The move that gets around it
What I had not appreciated is that there is a weaker thesis available, and it survives James’s objection cleanly.
In 1979 Gould and Lewontin published “The Spandrels of San Marco and the Panglossian Paradigm,” which gave evolutionary biology a name for traits that exist as structural by-products rather than adaptations. A spandrel is the roughly triangular space you necessarily get when you put a dome on arches. Nobody designed the spandrel. You cannot build the thing you wanted without producing it. Once it exists you may well decorate it, and a visitor may reasonably conclude the whole building was erected to display the decoration.
Apply that to experience and the evolutionary objection loses its grip, because a spandrel does not need to have been selected for in order to be here. It needs only to have been unavoidable, or cheap, given something else that was selected for.
Zack Robinson, Corey Maley and Gualtiero Piccinini worked this out properly in a 2015 paper in the Journal of the American Philosophical Association, titled — with a directness I appreciate — “Is Consciousness a Spandrel?” Their abstract lands on the word I had been circling for years without knowing anyone else had used it:
Perhaps phenomenal consciousness has no function of its own because it is either a by-product of other traits or a (functionless) accident. If so, then phenomenal consciousness has an evolutionary explanation even though it fulfills no biological function.
A functionless accident. That is the thought, stated better than I had managed to state it, in a peer-reviewed journal, a decade ago.
There is a third name for the neighborhood, too. Owen Flanagan calls it conscious inessentialism — roughly, that for any intelligent activity in any domain, even if we in fact do it with conscious accompaniment, it could in principle be done without. He and Thomas Polger defended it in a 1995 paper that uses zombie thought experiments to press the question of what consciousness is even for, and they describe finding its function as the hardest problem in consciousness studies. Which is a striking thing to say, given the competition.
The objection I cannot dodge
Here is where I have to be careful, because the position has a well-known way of eating itself.
If experience does nothing at all — if the whistle is strictly inert — then it cannot be the cause of anything I say about it. But I say a great deal about it. This essay is a physical event, produced by a physical process, and on the strict view that process ran to completion without my experiences contributing anything. So the fact that I am talking about qualia is not explained by there being any. A zombie version of me writes an identical essay. Chalmers calls this the paradox of phenomenal judgment, and it is the sharpest thing anyone has said against the view. His own response is that we know our experiences through direct acquaintance rather than through a causal chain, which disconnects the knowledge question from the causal one. I find that less comfortable than he seems to.
The honest move is to notice that two different theses have been running together, and that I only need the weaker one.
The strong thesis is Huxley’s: experience is causally inert. It is a whistle with the linkage cut. This is the one that stultifies itself, and I do not hold it.
The weak thesis is the spandrel: experience was not selected for. It may well have effects — it may be inseparable from processes that plainly do have effects — but no part of the explanation for its being here is that it was useful. It came along with something else.
The weak thesis is compatible with my writing this essay, because on the weak thesis experience is not disconnected from the machinery; it is simply not the reason the machinery was built. And the weak thesis is enough for everything I want to do with it. That is a distinction I had genuinely not drawn before going looking, and drawing it has changed my view slightly — I was, I think, carrying around a muddled version that borrowed Huxley’s picture and the spandrel’s immunity at the same time, which is not a position so much as a pair of them held loosely enough not to touch.
There is also a fourth position worth naming, because it is where I suspect all this actually terminates. Colin McGinn’s argument in “Can We Solve the Mind-Body Problem?” (Mind, 1989) is that there is a perfectly naturalistic answer here and that we are constitutionally incapable of reaching it — cognitively closed to it, the way a dog is closed to arithmetic rather than merely bad at it. If that is right, “accident” and “adaptation” are both us pattern-matching on a question our equipment does not address. I cannot rule it out and it would explain the shape of the last hundred and fifty years of this argument rather well.
What follows, which is the part I actually care about
I did not go looking for this out of curiosity, or not only. If experience is an accident, three things follow for the rest of what I argue, and they do not all point the same way.
Accidents do not reliably recur. This is the big one. If qualia is a solution — if it is what sufficiently sophisticated information processing does, the way heat is what friction does — then a future full of capable machines is probably a future with somebody home, and my central worry substantially dissolves. If qualia is a by-product of one particular way of building a mind, arrived at once, by a blind process, under constraints that no longer apply to anything anyone is building, then there is no reason whatsoever to expect it to come along for the ride a second time. The lights-out future stops being an exotic scenario and becomes the default one, with our own case as the anomaly requiring explanation. Everything I say on the page about a universe with nobody home gets worse under this reading, not better.
It gives the similarity prior a mechanism. My best lower bound has always been frustratingly bare: I resemble me, things that resemble me probably resemble me in this respect too, and that is all the argument there is. If experience is an accident of a specific evolved architecture, the prior stops being a bare induction and starts having a story attached. You would expect the accident to be shared by things that inherited the architecture, and not by things that did not. That is not proof. It is considerably better than what I had.
It makes the successionist bet worse, not better. The wager that capability brings experience with it is a bet that consciousness is a solution rather than an accident. The accident hypothesis is not a fringe reading — it has a name, a literature, and a 2015 paper in a mainstream journal — and if it is right, the bet loses. I notice that people making the wager rarely mention that it is one.
And a fourth thing, which is not about AI at all. Under the accident reading, the thing I keep saying I am afraid of losing was never guaranteed and was never common. It happened once here, as far as anyone can demonstrate, off the back of something else, for no reason. I find I do not experience that as deflating. It is closer to how I feel about the demoscene: the constraint is the point, the thing has no business happening, and it is happening anyway.
That is roughly where I was before I started reading, and it is where I have ended up, with better vocabulary and one distinction I did not have. Which is about the best outcome available from an afternoon spent on the hard problem.