Tag Archives: illusions

Flat thinking

Posted on by
14

Nick Chater says The Mind Is Flat in his recent book of that name. It’s an interesting read which quotes a good deal of challenging research (although quite a lot is stuff that I imagine would be familiar to most regular readers here). But I don’t think he establishes his conclusion very convincingly. Part of the problem is a slight vagueness about what ‘flatness’ really means – it seems to mean a few different things and at times he happily accepts things that seem to me to concede some degree of depth. More seriously, the arguments range from pretty good through dubious to some places where he seems to be shooting himself in the foot.

What is flatness? According to Chater the mind is more or less just winging it all the time, supplying quick interpretations of the sensory data coming in at that moment (which by the way is very restricted) but having no consistent inner core; no subconscious, no self, and no consistent views. We think we have thoughts and feelings, but that’s a delusion; the thoughts are just the chatter of the interpreter, and the feelings are just our interpretation of our own physiological symptoms.

Of course there is a great deal of evidence that the brain confabulates a great deal more than we realise, making up reasons for our behaviour after the fact. Chater quotes a number of striking experiments, including the famous ones on split-brain patients, and tells surprising stories about the power of inattentional blindness. But it’s a big leap from acknowledging that we sometimes extemporise dramatically to the conclusion that there is never a consistent underlying score for the tune we are humming. Chater says that if Anna Karenina were real, her political views would probably be no more settled than those of the fictional character, about whom there are no facts beyond what her author imagined. I sort of doubt that; many people seem to me to have relatively consistent views, and even where the views flip around they’re not nugatory. Chater quotes remarkable experiments on this, including one in which subjects asked political questions via a screen with a US flag displayed in the corner gave significantly more Republican answers than those who answered the same questions without the flag – and moreover voted more Republican eight months later. Chater acknowledges that it seems implausible that this one experiment could have conditioned views in a way that none of the subjects’ later experiences could do (though he doesn’t seem to notice that being able to hold to the same conditioning for eight months rather contradicts his main thesis about consistency); but in the end he sort of thinks it did. These days we are more cautious about the interpretation of psychological experiments than we used to be, and the most parsimonious explanation might be something wrong with the experiment. An hypothesis Chater doesn’t consider is that subjects are very prone to guessing the experimenter’s preferences and trying to provide what’s wanted. It could plausibly be the case that subjects whose questions were accompanied by patriotic symbols inferred that more right-wing answers would be palatable here, and thought the same when asked about their vote much later by the same experimenter (irrespective of how they actually voted – something we can’t in fact know, given the secrecy of the ballot).

Chater presents a lot of good evidence that our visual system uses only a tiny trickle of evidence; as little as one shape and one colour at a time, it seems. He thinks this shows that we’re not having a rich experience of the world; we can’t be, because the data isn’t there. But isn’t this a perverse interpretation? He accepts that we have the impression of a rich experience; given the paucity of data, which he evidences well, this impression can surely only come from internal processing and internal models – which looks like mental depth after all. Chater, I think, would argue that unconscious processing does take place but doesn’t count as depth; but it’s hard to see why not. In a similar way he accepts that we remember our old interpretations and feed them into current interpretations, even extrapolating new views, but this process, which looks a bit like reflection and internal debate, does not count as depth either. Here, it begins to look as if Chater’s real views are more commonsensical than he will allow.

But not everywhere. On feelings, Chater is in a tradition stretching back to William James, who held that our hair doesn’t stand on end because we’re feeling fear; rather, we feel fear because we’re experiencing hair-rise (along with feelings in the gut, goose bumps, and other physiological symptoms). The conscious experience comes after the bodily reaction, not before. Similar views were held by the behaviourists, of course; these reductive ideas are congenial because they mean emotions can be studied objectively from outside, without resort to introspection. But they are not very plausible. Again, we can accept that it sometimes happens that way. If stomach ache makes me angry, I may well go on to find other things to be angry about. If I go out at night and feel myself tremble, I may well decide it is because I am frightened. But if someone tells a ghost story, it is not credible that the fear doesn’t come from my conscious grasp of the narrative.

I think Chater’s argument for the non-existence of the self is perhaps his most alarming. It rests on a principle (or a dogma; he seems to take it as more or less self evident) that there is nothing in consciousness but the interpretation of sensory inputs. He qualifies this at once by allowing dreams and imagination, a qualification which would seem to give back almost everything the principle took away, if we took it seriously; but Chater really does mean to restrict us to thinking about entities we can see, touch or otherwise sense. He says we have no conscious knowledge of abstractions, not even such everyday ones as the number five. The best we can do is proxies such as an image of five dots, or of the numeral ‘5’. But I don’t think that will wash. A collection of images is not the same as the number five; in fact, without understanding what five is, we wouldn’t be able to pick out which groups of dots belonged to the collection. Chater says we rely on precedent, not principle, but precedents are useless without the interpretive principles that tell us when they apply. I don’t know how Chater, on his own account, is even aware of such things as the number five; he refuses to address the metaphysical issues beyond his own assertions.

I think Chater’s principle rules out arithmetic, let alone higher maths, and a good deal besides, but he presumably thinks we can get by somehow with the dots. Later, however, he invokes the principle again to dismiss the self. Because we have no sensory impressions of the self, it must be incoherent nonsense. But there are proxies for the self – my face in the mirror, the sound of my voice, my signature on a document – that seem just as good as the dots and numerals we’ve got for maths. Consistency surely requires that Chater either accepts the self or dumps mathematics.

As a side comment, it’s interesting that Chater introduces his principle early and only applies it to the self much later, when he might hope we have forgotten the qualifications he entered, and the issues over numbers. I do not suggest these are deliberate presentational tactics, rather they seem good evidence of how we often choose the most telling way of giving an argument unconsciously, something that is of course impossible in his system.

I’m much happier with Chater’s view of AI. Early on, he gives a brief account of the failure of the naive physics project, which attempted to formalise our ‘folk’ understanding of science. He seems to conflate this with the much wider project of artificial general intelligence, but he is right about the limitations it points to. He thinks computers lack the ‘elasticity’ of human thought, and are unlikely to acquire it any time soon.

A bit of a curate’s egg, then. A lot of decent, interesting science, with some alarming stuff that seems philosophically naive (a charge I hesitate to make because it is always possible to adduce sophisticated justifications for philosophical positions that seem daft on the face of it; indeed, that’s something philosophers particularly enjoy doing).

Illusions for robots

Posted on by
33

robot illusionsNeural networks really seem to be going places recently. Last time I mentioned their use in sophisticated translation software, but they’re also steaming ahead with new successes in recognition of visual images. Recently there was a claim from MIT that the latest systems were catching up with primate brains at last. Also from MIT (also via MLU) though, has come an intriguing study into what we could call optical illusions for robots, which cause the systems to make mistakes which are incomprehensible to us primates. The graphics in the grid on the right apparently look like a selection of digits between one and six in the eyes of these recognition systems. Nobody really knows why, because of course neural networks are trained, not programmed, and develop their own inscrutable methods.

How then, if we don’t understand, could we ever create such illusions? Optical illusions for human beings exploit known methods of visual analysis used by the brain, but if we don’t know what method a neural network is using, we seem to be stymied. What the research team did is use one of their systems in reverse, getting it to create images instead of analysing them. These were then evaluated by a similar system and refined through several iterations until they were accepted with a very high level of certainty.

This seems quite peculiar and the first impression is that it rather seriously undermines our faith in the reliability of neural network systems. However, there’s one important caveat to take into account: the networks in question are ‘used to’ dealing with images in which the crucial part to be identified is small in relation to the whole. They are happy ignoring almost all of the image. So to achieve a fair comparison with human recognition we should perhaps think of the question being not ‘do these look like numbers to you?’ and more like ‘can you find one of the digits from one to six hidden somewhere in this image?’. On that basis the results seem easier to understand.

There still seem to be some interesting implications, though. The first is that, as with language, AI systems are achieving success with methods that do not much resemble those used by the human brain. There’s an irony in this happening with neural networks, because in the old dispute between GOFAI and networks it was the network people who were trying to follow a biological design, at least in outline.  The opposition wanted to treat cognition as a pure engineering problem; define what we need, identify the best way to deliver it, and don’t worry about copying the brain. This is the school of thought that likes to point out that we didn’t achieve flight be making machines with flapping, feathery wings. Early network theory, going right back to McCulloch and Pitts, held that we were better off designing something that looked at least broadly like the neurons in the brain. In fact, of course, the resemblance has never been that close, and the focus has generally been more on results than on replicating the structures and systems of biological brains; you could argue that modern neural networks are no more like the brain than fixed-wing aircraft are to birds (or bats).  At any rate, the prospect of equalling human performance without doing it the human way raises the same nightmare scenario I was talking about last time; robots that are not people but get treated as if they were (and perhaps people being treated like machines as a consequence.

A second issue is whether the deception which these systems fall into points to a general weakness. Could it be that these systems work very well when dealing with ‘ordinary’ images but continue go wildly off the rails when faced with certain kinds of unusual ones – even when being pout to practical use? It’s perhaps not very likely that  system is going to run into the kind of truly bizarre image we seem to be dealing with, but a more realistic concern might be the potential scope for sabotage or subversion on the part of some malefactor.  One safeguard against this possibility is that the images in question were designed by, as it were, sister systems, ones that worked pretty much the same way and presumably shared the same quirks. Without owning one of these systems yourself it might be difficult to devise illusions that worked – unless perhaps there are general illusions that all network systems are more or less equally likely to be fooled by? That doesn’t seem very likely, but it might be an interesting research project.  The other safeguard is that these systems are not likely to be used without some additional safeguards, perhaps even more contextual processing of broadly the kind that the human mind obviously brings to the task.

The third question is – what is it like to be an AI deceived by an illusion? There’s no reason to think that these machines have subjective experience – unless you’re one of those who is prepared to grant a dim glow of awareness to quite simple machines – but what if some cyborg with a human brain, or a future conscious robot, had systems like these as part of its processing apparatus rather than the ones provided by the human brain?  It’s not implausible that the immense plasticity of the human brain would allow the inputs to be translated into normal visual experience, or something like it.  On the whole I think this is the most likely result, although there might be quirks or deficits (or hey, enhancements, why not) in the visual experience.  The second possibility is that the experience would be completely weird and inexpressible and although the cyborg/robot would be able to negotiate the world just fine, its experience would be like nothing we’ve ever had, perhaps like nothing we can imagine.

The third possibility is that it would be like nothing. There would be no experience as such; the data and the knowledge about the surroundings would appear in the cyborg/human’s brain but there would be nothing it was like for that to happen.  This is the answer qualophile scpetice would expect for a pure robot brain, but the cyborg is more worrying. Human beings are supposed to experience qualia, but when do they arise? Is it only after all the visual processing has been done – when the data ariive in the ‘Cartesian Theatre’ which Dennett has often told us does not exist? Is it, instead, in the visual processing modules or at the visual processing stage? If so, then we were wrong to doubt that MIT’s systems are not having experiences. Perhaps the cyborg gets flawed or partial qualia – but what would that even mean..?