Posts tagged ‘artificial intelligence’

Picture: heraldic whale. … it’s AGI now. I was interested to hear via Robots.net that Artificial General Intelligence had enjoyed a successful second conference recently.

In recent years there seems to have been a general trend in AI research towards more narrow and perhaps more realistic sets of goals; towards achieving particular skills and designing particular modules tied to specific tasks rather than confronting the grand problem of consciousness itself. The proponents of AGI feel that this has gone so far that the terms ‘artificial intelligence’ and AI no longer really designate the topic they’re interested in, the topic of real thinking machines.  ‘An AI’ these days is more likely to refer to the bits of code which direct the hostile goons in a first-person shooter game than to anything with aspirations to real awareness, or even real intelligence.

The mention of  ‘real intelligence’ of course, reminds us that plenty of other terms have been knocked out of shape over the years in this field. It is an old complaint from AI sceptics that roboteers keep grabbing items of psychological vocabulary and redefining them as something simpler and more computable. The claim that machines can learn, for example, remains controversial to some, who would insist that real learning involves understanding, while others don’t see how else you would describe the behaviour of a machine that gathers data and modifies its own behaviour as a result.

I think there is a kind of continuum here, from claims it seems hard to reject to those it seems bonkers to accept, rather like this…

Claim: machines… Objection
add numbers. Really the ‘numbers’ are a human interpretation of meaningless switching operations.
control factory machines. Control implies foresight and intentions whereas machines just follow a set of instructions.
play chess. Playing a game involves expectations and social interaction, which machines don’t really have.
hold conversations Chat-bots merely reshuffle set phrases to give the impression of understanding.
react emotionally There may be machines that display smiley faces or even operate in different ’emotional’ modes, but none of that touches the real business of emotions.

Readers will probably find it easy to improve on this list, but you get the gist. Although there’s something in even the first objection, it seems pointless to me to deny that machines can do addition – and equally pointless to claim that any existing machine experiences emotions – although I don’t rule even that idea out of consideration forever.

I think the most natural reaction is to conclude that in all such cases, but especially in the middling ones, there are two different senses – there’s playing chess and really playing chess. What annoys the sceptics is their perception that AIers have often stolen terms for the easy computable sense when the normal reading is the difficult one laden with understanding, intentionality and affect.

But is this phenomenon not simply an example of the redefinition of terms which science has always introduced? We no longer call whales fish, because biologists decided it made sense to make fish and mammals exclusive categories – although people had been calling whales fish on and off for a long time before that. Aren’t the sceptics on this like diehard whalefishers? Hey, they say, you claimed to be elucidating the nature of fish, but all you’ve done is make it easy for yourself by making the word apply just to piscine fish, the easy ones to deal with. The difficult problem of elucidating the deeper fishiness remains untouched!

The analogy is debatable, but it could be claimed that redefinitions of  ‘intelligence’ and ‘learning’ have actually helped to clarify important distinctions in broadly the way that excluding the whales helped with biological taxonomy. However, I think it’s hard to deny that there has also at times been a certain dilution going on. This kind of thing is not unique to consciousness – look what happened to ‘virtual reality’, which started out as quite a demanding concept, and was soon being used as a marketing term for any program with slight pretensions to 3D graphics.

Anyway, given all that background it would be understandable if the sceptical camp took some pleasure in the idea that the AI people have finally been hoist with their own petard, and that just as the sceptics, over the years, have been forced to talk about ‘real intelligence’ and ‘human-level awareness’, the robot builders now have to talk about ‘artificial general intelligence’.

But you can’t help warming to people who want to take on the big challenge. It was the bold advent of the original AI project which really brought consciousness back on to the agenda of all the other disciplines, and the challenge of computer thought which injected a new burst of creative energy into the philosophy of mind, to take just one example. I think even the sceptics might tacitly feel that things would be a little quiet without the ‘rude mechanicals’: if AGI means they’re back and spoiling for a fight, who could forbear to cheer?

Picture: ducks. The New Scientist, under the title ‘Tests that show machines closing in on human abilities’ has a short review of some different ways in which robotic or computer performance is being tested against human achievement. The piece is not really reporting fresh progress in the way its title suggests  – the success of Weizenbaum’s early chat-bot Eliza is not exactly breaking news,  for example – but I think the overall point is a good one. In the last half of the last century, it was often assumed that progress towards AI would see all the barriers come down at once: as soon as we had a robot that could meet Turing’s target of a few minutes’ successful small-talk, fully-functioning androids would rapidly follow.

In practice, Turing’s test has not been passed as he expected, although some stalwart souls continue to work on it. But we have seen the overall problem of human mentality unpicked into a series of substantial, but lesser challenges. Human levels of competence remain the target, but competence in different narrow fields, with no expectation that solving the problem in one area solves it in all of them.

THe piece ends with a quote from Stevan Harnad which suggests he clings to the old way of looking at this:

“If a machine can prove indistinguishable from a human, we should award it the respect we would to a human”

That may be true, in fact, but the question is, indistinguishable in which respects? People often quote a  particular saying in this respect: if it walks like a duck and quacks like a duck, it’s a duck.  Actually, even in the case of ducks this isn’t as straightforward as it might seem since, other wildfowl may be ducklike in some respects. Given a particular bird, how many of us could say with any certainty whether it were a Velvet Scoter, a White-winged Coot – or just a large sea-duck? But it’s worse than that. The Duck Law, as we may call it, works fairly well in real life; but that’s because as a matter of fact there aren’t all that many anatine entities in the world other than outright ducks. If there were a cunning artificer who was bent on turning out false, mechanical ducks like the legendary one made by Vaucanson, which did not merely walk and quack like a duck, but ate and pooped like one, we should need a more searching principle.  When it comes to the Turing Test, that is pretty much the position we find ourselves in.

There is, of course, a more rigorous version of Duck Law which is intellectually irreproachable, namely Leibniz’s Law. Loosely, this says that if two object share the same properties, they are the same. The problem is, that in order to work, Leibniz’s law has to be applied in the most rigorous fashion.  It requires that all properties must be the same. To be indistinguishable from a human being in this sense means literally indistinguishable, ie having human guts, a human mother and so on.

So, in which respects must a robot resemble a human being in order to be awarded the same respect as a human? It now seems unlikely that a machine will pass the original Turing Test soon; but even if it did, would that really be enough?  Just looking like a human, even in flexible animation which reproduces the pores of the skin and typical human movements, is clearly not enough. Nor is being able to improvise jazz tunes seamlessly with a human player. But these things are all significant achievements nevertheless. Perhaps this is the way we make progress.

Or possibly, at some stage in the future, someone will notice that if he were to bolt together a hundred different software modules and some appropriate bits of hardware, all by then readily available, he could theoretically produce a machine able to do everything a human can do; but he won’t by then see any point in actually doing it.

Picture: Turing.

It’s just 50 years since Alan Turing’s tragic death. The anniversary was marked in Manchester and elsewhere, but little seems to have appeared on the Internet – perhaps surprisingly, given his importance in the development of the computer..

Turing has a number of tremendous achievements to his credit. His war-time code-breaking may be the most famous; but perhaps the most important was the idea of the Turing machine, the theoretical apparatus which defined computation and computers. It had two distinct consequences: on the one hand, it dealt with the Entscheidungsproblem, one of the key issues of 20th century mathematics; on the other, it gave rise, via Turing’s famous (1950) paper, to the period of intense optimism about artificial intelligence which I referred to earlier as the ‘Turing era’ . The curious thing is that these two consequences of the Turing machine point in opposite, almost antithetical directions.

How so? The Entscheidungsproblem, posed by Hilbert, asks whether there is any mechanical procedure for determining whether a mathematical problem is solvable. The universal Turing machine embodies and clarifies the idea of ‘mechanical’ calculation. It is a simple apparatus which prints or erases characters on a paper tape according to the rules it has been given. In spite of this extreme simplicity it can in principle carry out any mechanical computation. In theory, in fact, it can run an appropriate version of any computer program, including the ones being used to display this page. In many respects it appears to be an entirely realistic machine which could easily be put together, but it has certain other qualities which make it an impossible abstraction. For one thing, it has to have an infinite paper tape: for another, it has to be immune to malfunction, no matter how long it runs; and most fundamental of all, it has to operate with discrete states – it must switch from one physical configuration to another without any intervening half-way stages. These characteristics mean that it is actually more like a complex function than a real machine. Nevertheless, all real-world computers owe their computerhood to their resemblance to it.

The clear conception of computation which the Turing machine provided allowed Turing to show that the Entscheidungsproblem had to be answered in the negative – there is no general procedure which can deal with all mathematical problems, even in principle. In fact, Turing was slightly too late to claim full credit for this result, which had already been established by Alonzo Church using a different approach,

The thing is, this result goes naturally with Gödel’s proof of the incompleteness of arithmetic in the sense that both establish limitations of formal algorithmic calculation. Both, therefore, suggest that the kind of computation performed by machines can never fully equal the thought processes of human beings (however those may work), which do not seem to suffer the same limitations. Gödel seems to have interpreted his own work this way. In fact there is some reason to think that Turing initially took a similar view. Andrew Hodges has pointed out that after completing his work on the Entscheidungsproblem, Turing attempted to produce a formal logic based on ordinals. It seems to have been the idea that this new, ordinal-based work would provide the basis for the kind of ‘intuitive’ reasoning which Turing machines couldn’t deliver – the kind human beings used to see the truth of Gödel statements. Only when these efforts failed, it seems, did Turing look for reasons to think that machine-style computation might be good enough to deliver a real mind after all.

Looked at again in this light, the 1950 paper seems more evasive and equivocal. It is a curious paper in many ways, with its playful tone and respectful mentions of ESP and Ada, Countess of Lovelace, but it also skirts the issue. Can machines think? Well, it says, let’s consider instead whether they can pass the Turing test . If they can, well, perhaps the original question is too meaningless to worry about.

But it surely isn’t meaningless: it’s partly because we believe that people really can think that our attitude to death is so different from our attitude to switching off the computer, for example.

It seems possible, anyway, that Turing’s desire to believe that a mechanical mind was possible led him to seek ways around the negative implications of his own work. The logic of ordinals was one possibility: when that failed, the Turing Test was basically another, justifiying further work with Turing-machine style computers.

Had he lived, of course, he might eventually have changed his mind about his own Test, or found better ways of dealing with ‘intuition’. We’ll never know quite how much we lost when, punished for his homosexuality with oestrogen injections and expelled from further participation in Government work, he killed himself with a poisoned apple.

But it is a poignant thought that in the natural course of things he could still have been alive today.

Picture: Turing. Picture: Blandula. Can machines think? That was the question with which Alan Turing opened his famous paper of 1950, ‘Computing machinery and intelligence’. The question was not exactly new, but the answer he gave opened up a new era in our thinking about minds. It had been more or less agreed up to that time that consciousness required a special and particularly difficult kind of explanation. If it didn’t require spiritual intervention, or outright magic, it still needed some special power which no mere machine could possibly reproduce. Turing boldly predicted that by the end of the century we should have machines which everyone habitually treated as conscious entities, and his paper inspired a new optimism about our ability to solve the problems. But that was 1950. I’m afraid that fifty years of work since then have effectively shown that the answer is no – machines can’t think

Picture: Bitbucket. A little premature, I think. You have to remember that until 1950 there was very little discussion of consciousness. Textbooks on psychology never mentioned the subject. Any scientist who tried to discuss it seriously risked being taken for a loony by his colleagues. It was effectively taboo. Turing changed all that, partly by making the notion of a computer a clear and useful mathematical concept, but also through the ingenious suggestion of the Turing Test . It transformed the debate and during the second half of the century it made consciousness the hot topic of the day, the one all the most ambitious scientists wanted to crack: a subject eminent academics would take up after their knighthood or Nobel. The programme got under way, and although we have yet to achieve anything like a full human consciousness, it’s already clear that there is no insurmountable barrier after all. I’d argue, in fact, that some simple forms of artificial consciousness have already been achieved.

Picture: Blandula. But Turing’s deadline, the year 2000, is past. We know now that his prediction, and others made since, were just wrong. Granted, some progress has been made: no-one now would claim that computers can’t play chess. But they haven’t done that well, even against Turing’s own test, which in some ways is quite undemanding. It’s not that computers failed it; they never got good enough even to put up a serious candidate. You say that consciousness used to be a taboo subject, but perhaps it was just that earlier generations of scientists knew how to shut up when they had nothing worth saying…

Picture: Bitbucket. Of course, people got a bit over-optimistic during the last half of the last century. People always quote the story about Marvin Minsky giving one of his graduate students the job of sorting out vision over the course of the summer (I have a feeling that if that ever happened it was a joke in the first place). Of course it’s embarassing that some of the wilder predictions have not come true. But you’re misrepresenting Turing. The way I read him, he wasn’t saying it would all be over by 2000, he was saying, look, let’s put the philosophy aside until we’ve got a computer that can at least hold some kind of conversation.

But really I’m wasting my breath – you’ve just got a closed mind on the subject. Let’s face it, even if I presented you with a perfectly human robot (even if I suddenly revealed that I myself had been a robot all along), you still wouldn’t accept that it proved anything, would you?

Picture: Blandula. Your version of Turing sounds relatively sensible, but I just don’t think his paper bears that interpretation. As for your ‘perfectly human’ robot, I look forward to seeing it, but no, you’re right, I probably wouldn’t think it proved anything much. Imitating a person, however brilliantly, and being a person are two different things. I’d need to know what was going on inside the robot, and have a convincing theory of why it added up to real consciousness.

Picture: Bitbucket. No theory is going to be convincing if you won’t give it fair consideration. I think you must sometimes have serious doubts about the so-called problem of other minds. Do you actually feel sure that all your fellow human beings are really fully conscious entities?

Picture: Blandula. Well…