Consciousness and blindsight

(For those interested in large language models: I did try holding a conversation with one about my idea of writing this page, and it excitedly proclaimed I'm right and should do it. However, I believe these models are deliberately biased, via RLHF and similar techniques applied after initial training, toward affirming the user, so their agreement is at best only weak evidence. I did not ask for writing help: all wording is my own.)

Dennett seems to be a behaviourist---his book reads a bit like the more recent Consciousness and the Social Brain by Michael Graziano (2013) suggesting that the experience of being conscious is a delusion, no more than a vivid mental model we have for controlling our signal-processing priorities which we call "attention", and Consciousness and the Brain by Stanislas Dehaene (2014) who was afraid that too much discussion of "qualia" can lead to dualism and the belief seen in most (but not all) religions that man has a "soul" that's somehow separate from the brain it's in. While I'm not a fan of dualism, I don't think fear of dualism should stall any and all discussion of qualia, any more than I think fear of young-earth creationism should stall any and all discussion of the current limitations of evolutionary theory---that's basically letting the "other side" push you into stopping your scientific progress, which is just what I thought you didn't want. So can we try not to be afraid of the big bad dualists and talk about qualia anyway?

At this point I should perhaps disclose that I am particularly sensitive to people who think they know about brains telling me that I have a delusion, because I had to go through misdiagnosis. Before the doctors had access to enough technology to prove beyond reasonable doubt that my brain really was faulting in its visual cortex, I had therapists trying in vain to figure out if some "repressed memory" was causing me to be deluded into thinking I had visual problems and needed to invent coping behaviours. When we finally got concrete evidence from EEG and ERG testing (the best part of which was that I didn't have to speak, so there was no longer any question of was I making it all up), it was such a relief to know I wasn't just deluding myself into the condition. So when the behaviourists come along and say "ah but your whole conscious experience is a delusion" I have to try very hard not to imagine them working at the places I saw in the bad old days before I was properly diagnosed. It might be better if people didn't take a word normally associated with pathological cases and overload it to refer to an aspect of normal functioning.

Dennett's book used the phenomenon of 'blindsight' in his argument that the "qualia" of consciousness is just an illusion, and it is here that I believe I can most directly undermine his argument (nothing personal---he was doing his best---it just happens to be an argument I can't agree with). Unless I've misunderstood the book, he basically said that if a person blind due to brain damage can be prompted to "guess" what they would see if their brain were working better (and have a better than expected chance of guessing correctly), and could then learn to "guess when to guess" and keep improving until it's a lot like "real" seeing minus the "qualia", then that shows we're only deluding ourselves into thinking that the "qualia" of consciousness means anything.

I'm sorry---I may be merely a run-of-the-mill computer scientist, not a cognitive-science director---but I do happen to know more than Professor Dennett did about blindsight. At least, I've had a whole lifetime of working with the versions of it presented by my own damaged visual cortex and I'm afraid his book didn't give me the impression that he knew what he was talking about---so let me try to explain what it's really like in here.

I'm a partially-sighted CVI case. It's difficult to explain what that means to people who want a quick summary. They ask questions like "how far can you see" and I have to resist the temptation to reply "93 million miles at sunset but that's nothing to your 2.5 million light-years if you've seen M31". And if I miss lots of big and obvious things only to notice a small detail they hadn't seen themselves, that doesn't make sense to them at all. To give them an impression of how unpredictable it can be, I sometimes describe it as "someone telling you what they see but over a bad phone line so you don't catch all the words" or (for Trekkies) the bridge of the Enterprise with a muddled crewman who keeps forgetting how to read the console but can unexpectedly manage sometimes---which can mean I'll notice more if I'm able to look for long enough, but I can't count on that. These descriptions are only approximations. Perhaps a more accurate one is, I have experienced sight at different "levels" of consciousness---and more than one level in the same person means I can actually compare them.

My experiences have ranged from:

• no awareness at all,
• to a vague hunch I know something but without having any idea how I know (the traditional "blindsight" phenomenon),
• to a feeling that something is moving or changing somewhere but without being sure what (possibly due to my V5 area managing to get a signal through without the V4 working with it; this can get annoying, confusing or overwhelming if persistent---flashing festive lights are the worst, especially when the manufacturers of such things add embedded code to show off by varying the period, and I have been known to unplug College Christmas trees when desperate although I always inform them if I had to do that),
• to the impression of having seen something for just a brief flash of time followed by vague memories of it, like waking from a dream that seems to have had a lot more context than its actual length accounted for but you can't quite remember all the details and sometimes aren't even sure if you slept or not,
• to a partial perception of a written word or symbol along with the sense of recalling a verbal memory of the whole word which might or might not be correct (my usual heuristic is "if it doesn't make logical sense run it through again" although I have to be even more careful if it's important),
• to an image that is somewhat more stable but not completely interpretable, and yet not "blurred" or anything like that (for example, I can't normally tell people's faces apart when my PIT cortex is being wobbly, but I still see "a face" it's just that it looks to me like most people have the same face),
• to varying higher levels of interpretability according to what the system's doing (but usually within only a very small angular field that's not always as controllable as I'd like),

I believe the level of "qualia" is proportional to the system's confidence that it has a meaningful read of the signal. I can act on a half-conscious "hunch" and be correct more often than chance, but I wouldn't want to risk doing so in any context in which getting it wrong has major or even dangerous implications: I'd need to check things first (by non-visual means if necessary: if you've ever seen me use my cane to check an obstacle in a way that seemed like I already suspected where it was, that'll be why), or at least add in a more conscious level of Bayesian inference---a rough calculation that the probability of something being there in the current context, given that I seem to be detecting it, is the probability of both of these things together divided by the probability of the detector going off anyway in the current context---and the correct threshold to act on will depend on the severity of consequences of false positives versus false negatives.

As I've been assisting Cambridge's large Chinese population, I've been trying to train my blindsight system to function as a rudimentary "probable CJK-user detector" but with mixed results including false positives (if I've ever asked you if you speak Chinese when you clearly don't look Chinese, that'll be why, sorry). And due to the probable nature of the damage to the posterior inferotemporal cortex I still can't get it to recognise faces or expressions: I can sometimes learn to recognise people via non-facial aspects of their appearance, but only up to a certain confidence level, and not if they modify the aspects of their appearance that I'm picking up on---and I don't always consciously know what these are.

It's interesting that people have decided to call fictional information generated by large language models "hallucinations"---I also can draw wrong conclusions only too easily if I take too seriously those sensations which I shouldn't be sure about, although I'm not sure exactly what level of consciousness of the false information would be required to apply the term "hallucination" to it in the human sense. For someone who doesn't believe large language models are conscious (pre-trained means it's only "crystallised" intelligence in Cattell's model and it's literally as if they're answering you in their sleep), I can still feel a certain kind of sympathy toward them, because they don't know how they're coming up with stuff---and can confabulate if asked---and neither do I know how my blindsight is working, and can feel under pressure to guess if asked. In both cases we may actually be picking up on some totally unexpected thing that happens to correlate with the result we want most of the time but not always. (At least I'm aware of unconscious bias.) I'm not sure how to get "explainable AI" when I know I can't even have explainable eyesight.

And yet, despite all the confusion, I do have varying levels of consciousness, and I'm not sure if Dennett's "multiple drafts" model really explains how the system updates its guesses as fragments of new data somehow make it through the faulty cortex---what he calls the "Orwellian" version of "Cartesian theatre" feels closer but still not quite right. Incidentally, George Orwell's "four fingers" torture scene would be unlikely to work on me because I'd too readily concede the uncertainty, skipping straight to the "fingers like moving trees" part even without the pain---and it wouldn't take too much pushing for me to do that with other senses too, although I wouldn't be made to agree with a dystopia in that way as I have certain convictions more "stubborn" than my sensory perception. I may admit I'm not qualified to condemn but I still believe some actions are wrong. Sorry to disappoint you O'Brien. (I try to at least be nice to my interrogators: it might not be their fault if they can't see things differently. Same goes for mistaken psychiatrists and qualia-denying behaviourists.)

Incidentally I'm not too keen on Professor David Berman's idea either, that Dennett was committing the "typical mind fallacy"---drawing on the "mental imagery" ideas of 19^th-century eugenicist Francis Galton, Berman basically said Dennett couldn't do qualia himself and for this reason thought nobody else could either. I don't think we have enough evidence to make such a claim when a simpler explanation is that Dennett was so wrapped up in behaviourist philosophy that he could easily dismiss his own sensations of qualia as illusory.

Cairns-Smith on the other hand suggested consciousness may be related to ill-understood physics---he mentioned quantum mechanics but that seems quite fragile (we'd want something that wouldn't be messed up by an MRI scan for a start)---he suggested it wouldn't form without an evolutionary purpose (which is why refrigerator thermostats and, by extension, large language models, can't really be conscious if their qualia cannot affect the physical process) and it's basically some unknown aspect of how the brain works (no dualism here). I'm not sure about everything he said but the "we still don't know" part sounded good.

Consciousness does seem closer to a "wave" (a process) than a "particle" (a thing)---due to the body's renewal mechanisms, you are neither physically nor psychologically the same person as you were years ago---Dawkins said in a 2005 talk "you weren't there" when thinking of your childhood events because your atoms have changed, although I'm not sure if he was completely correct to say "not a single atom" would remain---he has overstated things before---but the principle holds. Some of our thoughts and memories now may be strongly correlated to how they were a few years ago, but even that's unlikely to be exactly the same because we are changing processes, like the paradox of Theseus' ship---although that at least had an unbroken world-line but consciousness doesn't: you probably view yourself as at least very nearly the same individual you were yesterday, but if you slept last night, your awareness has not been continuous in physical spacetime (there's also a possibility it might not be continuous even during the day, but it definitely breaks at sleep). And I don't think I'd really know any different if my entire brain state were somehow transferred to a completely different physical brain before I woke up (although obviously I don't want anyone trying this as you don't know how to do it without breaking something)---and if multiple diverging copies were possible, they'd all be versions of myself and it wouldn't make sense to ask which one is "more real" (but still I wouldn't be sure the existence of backup versions would give me a right to risk my life on a larger category of problems than I would now). Julian Barbour and others' take on "timeless physics" probably needs more work but does suggest a framework in which consciousnesses can jump in spacetime and yet be "the same" without requiring dualism, if identity is perpetuated by similarity in configuration spaces.

But none of this necessarily makes "qualia" not real by any definition of "real" that can apply to things the brain does. We might not yet know what it is but I'd argue it's at least as real as the sunset: we know that's caused by the planet's rotation placing the sun behind Earth from our perspective, not the sun descending into the ground, but the fact that some people might once have misinterpreted what it is doesn't stop it from being a real phenomenon. And if you do want to argue that qualia is not real, you might want to come up with something more convincing than Dennett's example of blindsight because, while different cases can be different, I'm rather afraid my particular life has shown that particular example doesn't necessarily work (but nice try).

I was privileged to have had a good discussion about the brain and visual perception with Professor John Daugman (1954-2024, the iris-recognition inventor)---he showed me illusions and abstract art so I could say how I was interpreting them (sometimes reading more into them than a "normal" person would). He also agreed that current theories of evolution are insufficient to explain the human brain although he was keen to add that he didn't want his thought on that to be misinterpreted as a belief in any form of creationism and he felt it was a pity that scientists are reluctant to discuss flaws in evolutionary theory for fear of giving ammunition to creationists: he felt we'll have an improved theory in future.