Monthly Archives: April 2018

Getting Emotional

Posted on by
17

Are emotions an essential part of cognition? Luiz Pessoa thinks so (or perhaps he feels it is the case). He argues that while emotions have often been regarded as something quite separate from rational thought, a kind of optional extra which can be bolted on, but has no essential role, they are actually essential.

I don’t find his examples very convincing. He says robots on a Mars mission might have to plan alternative routes and choose between priorities, but I don’t really see why that requires them to get emotional. He says that if your car breaks down in the desert, you may need a quick fix. A sense of urgency will impel you to look for that, while a calm AI might waste time planning and implementing a proper repair. Well, I don’t know. On the one hand, it seems perfectly possible to appreciate the urgency of the situation in a calm rational way. On the other, it’s easy to imagine that panic and many other emotional responses might be very unhelpful, blocking the search for solutions or interfering with the ability to focus on implementation.

Yet there must be something in what he says, mustn’t there? Otherwise, why would we have emotions? I suppose in principle it could be that they really have no role; that they are epiphenomenal, a kind of side effect of no real importance. But they seem to influence behaviour in ways that make that implausible.

Perhaps they add motivation? In the final analysis, pure reason gives us no reason to do anything. It can say, if you want A, then the best way to get it is through X, Y, and Z. But if you ask, should I want A, pure reason merely shrugs, or at best it says, you should if you want B.

However, it doesn’t take much to provide a basic set of motivations. If we just assume that we want to survive, the need to obtain secure sources of food, shelter, and so on soon generate a whole web of subordinate motivations. Throw in a few very simple built-in drives – avoidance of pain, seeking sex, maintenance of good social relations – and we’re pretty much there in terms of human motivation. Do we need complex and distracting emotions?

Some argue that emotions add more ‘oomph’, that they intensify action or responses to stimuli. I’ve never quite understood the actual causal process there, but granting the possibility, it seems emotions must either harmonise with rational problem solving, or conflict with it. Rational problem solving is surely always best, so they must either be irrelevant or harmful?

One fairly traditional view is that emotions are a legacy of evolution, a system that developed before rational problem solving was available. So different emotional states affect the speed and intensity of certain sets of responses. If you get angry, you become more ready to fight, which may be helpful. Now, we would be better off deciding rationally on our responses, but we’re lumbered with the system our ancestors evolved. Moreover, some of the preparatory stuff, like a more rapid heartbeat, has never come under rational control so without emotions it wouldn’t be accessible. It can be argued that emotions are really little more than certain systems getting into certain potentially useful ready states like this.

That might work for anger, but I still don’t understand how grief, say, is a useful state to be in. There’s one more possible role for emotions, which is social co-ordination. Just as laughter or yawning tends to spread around the group, it can be argued that emotional displays help get everyone into a similar state of heightened or depressed responsiveness. But if that is truly useful, couldn’t it be accomplished more easily and in less disabling/distracting ways? For human beings, talking seems to be the tool for the job?

It remains a fascinating question, but I’ve never heard of a practical problem in AI that really seemed to need an emotional response.

Robot Memory

Posted on by
4

A new model for robot memory raises some interesting issues. It’s based on three networks, for identification , localization, and working memory. I have a lot of quibbles, but the overall direction looks promising.

The authors (Christian Balkenius, Trond A. Tjøstheim, Birger Johansson and Peter Gärdenfors)begin by boldly proposing four kinds of content for consciousness; emotions, sensations, perceptions (ie, interpreted sensations), and imaginations. They think that may be the order in which each kind of content appeared during evolution. Of course this could be challenged in various ways. The borderline between sensations and perceptions is fuzzy (I can imagine some arguing that there are no uninterpreted sensations  in consciousness, and the degree of interpretation certainly varies greatly), and imagination here covers every kind of content which is about objects not present to the senses, especially the kind of foresight which enables planning. That’s a lot of things to pack into one category. However, the structure is essentially very reasonable.

Imaginations and perceptions together make up an ‘inner world’. The authors say this is ess3ntial for consciousness, though they seem to have also said that pure emotion is an early content of consciousness. They propose two tests often used on infants as indicators of such an inner world; tests of the sense of object permanence and ‘A-not-B’. Both essentially test whether infants (or other cognitive entities) have an ongoing mental model of things which goes beyond what they can directly see. This requires a kind of memory to keep track of the objects that are no longer directly visible, and of their location. The aim of the article is to propose a system for robots that establishes this kind of memory-based inner world.

Imitating the human approach is an interesting and sensible strategy. One pitfall for those trying to build robot consciousness is the temptation to use the power of computers in non-human ways. We need our robot to do arithmetic: no problem! Computers can already do arithmetic much faster and more accurately than mere humans, so we just slap in a calculator module. But that isn’t at all the way the human brain tackles explicit arithmetic, and by not following the human model you risk big problems  later.

Much the same is true of memory. Computers can record data in huge quantities with great accuracy and very rapid recall; they are not prone to confabulation, false memories, or vagueness. Why not take advantage of that? But human memory is much less clear-cut; in fact ‘memory’ may be almost as much of a ‘mongrel’ term as consciousness, covering all sorts of abilities to repeat behaviour or summon different contents. I used to work in an office whose doors required a four-digit code. After a week or so we all tapped out each code without thinking, and if we had to tell someone what the digits were we would be obliged to mime the action in mid-air and work out which numbers on the keypad our fingers would have been hitting. In effect, we were using ‘muscle memory’ to recall a four-digit number.

The authors of the article want to produce the same kind of ‘inner world’ used in human thought to support foresight and novel combinations. (They seem to subscribe to an old theory that says new ideas can only be recombinations of things that got into the mind through the senses. We can imagine a gryphon that combines lion and eagle, but not a new beast whose parts resemble nothing  we have ever seen. This is another point I would quibble over, but let it pass.)

In fact, the three networks proposed by the authors correspond plausibly with three brain regions; the ventral, dorsal, and prefrontal areas of the cortex. They go on to sketch how the three networks play their role and report tests that show appropriate responses in respect of object permanence and other features of conscious cognition. Interestingly, they suggest that daydreaming arises naturally within their model and can be seen as a function that just arises unavoidably out of the way the architecture works, rather than being something selected for by evolution.

I’m sometimes sceptical about the role of explicit modelling in conscious processes, as I think it is easily overstated. But I’m comfortable with what’s being suggested here. There is more to be said about how processes like these, which in the first instance deal with concrete objects in the environment, can develop to handle more abstract entities; but you can’t deal with everything in detail in a brief article, and I’m happy that there are very believable development paths that lead naturally to high levels of abstraction.

At the end of the day, we have to ask: is this really consciousness? Yes and no, I’m afraid. Early on in the piece we find:

On the first level, consciousness contains sensations. Our subjective world of experiences is full of them: tastes, smells, colors, itches, pains, sensations of cold, sounds, and so on. This is what philosophers of mind call qualia.

Well, maybe not quite. Sensations, as usually understood, are objective parts of the physical world (though they may be ones with a unique subjective aspect), processes or events which are open to third-person investigation. Qualia are not. It is possible to simply identify qualia with sensations, but that is a reductive, sceptical view. Zombie twin, as I understand him, has sensations, but he does not have qualia.

So what we have here is not a discussion of ‘Hard Problem’ consciousness, and it doesn’t help us in that regard. That’s not a problem; if the sceptics are right, there’s no Hard stuff to account for anyway; and even if the qualophiles are right, an account of the objective physical side of cognition is still a major achievement. As we’ve noted before, the ‘Easy Problem’ ain’t easy…

Forgotten Crimes

Posted on by
5

Should people be punished for crimes they don’t remember committing? Helen Beebee asks this meaty question.

Why shouldn’t they? I take the argument to have two main points. First, because they don’t remember, it can be argued that they are no longer the same person as the one who committed the crime. Second, if you’re not the person who committed the crime, you can’t be responsible and therefore should not be punished. Both of these points can be challenged.

The idea that not remembering makes you a different person takes memory to be the key criterion of personal identity, a view associated with John Locke among others. But memory is in practice a very poor criterion. If I remember later, do I then become responsible for the crime? We remember unconsciously things we cannot recall explicitly; does unconscious memory count, and if so, how would we know? If I remember only unconsciously, is my unconscious self the same while my conscious one is not, so that perhaps I ought to suffer punishment I’m only aware of unconsciously? If I do not remember details, but have that sick sense of certainty that, yes, I did it alright, am I near enough the same person? What if I have a false, confabulated memory of the crime, but one that happens to be veridical, to tell the essential truth, if inaccurately? Am I responsible? If so, and if false memories will therefore do, then ought I to be held responsible even if in fact I did not commit the crime, so long as I ‘remember’ doing it?

Moreover, aren’t the practical consequences unacceptable? If forgetting the crime exculpates me, I can commit a murder and immediately take mnestic drugs that will make me forget it. If that tactic is itself punishable, I can take a few more drugs and forget even coming up with the idea. Surely few people think it really works as selectively as that. In order to be free of blame, you really need to be a different person, and that implies losing much more than a single memory. Perhaps it requires the loss of most memories, or more plausibly a loss of mentally retained things that go a lot wider than mere factual or experiential memory; my habits of thought, or the continuity of my personality. I think it’s possible that Locke would say something like this if he were still around. So perhaps the case ought to be that if you do not remember the crime, and other features of your self have suffered an unusual discontinuity, such that you would no longer commit a similar crime in similar circumstances, then you are off the hook. How we could establish such a thing forensically is quite another matter, of course.

What about the second point, though? Does the fact that I am now a different, and also a better person, one who doesn’t remember the crime, mean I shouldn’t be punished? Not necessarily. Legally, for example, we might look to the doctrines of joint enterprise and strict liability to show that I can sometimes be held responsible in some degree for crimes I did not directly commit, and even ones which I was powerless to prevent, if I am nevertheless associated with the crime in the required ways.

It partly depends on why we think people should be punished at all. Deterrence is a popular justification, but it does not require that I am really responsible.  Being punished for a crime may well deter me and others from attempting similar crimes in future, even if I didn’t do it at all, never mind cases where my responsibility is merely attenuated by loss of memory. The prevention of revenge is another justification that doesn’t necessarily require me to have been fully guilty. Or there might be doctrines of simple justice that hold to the idea of crime being followed by punishment, not because of any consequences that might follow, but just as a primary ethical principle. Under such a justification, it may not matter whether I am responsible in any strong sense. Oedipus did not know he had killed his father, and so could not be held responsible for patricide, at lest on most modern understandings; but he still put out his own eyes.

Much more could be said about all that, but for me the foregoing arguments are enough to suggest that memory is not really the point, either for responsibility or for personal identity. Beebee presents an argument about Bruce Banner and the Hulk; she feels Banner cannot directly be held responsible for the mayhem caused by the Hulk. Perhaps not, but surely the issue there is control, not memory. It’s irrelevant whether Banner remembers what the Hulk did, all that matters is whether he could have prevented it. Beebee makes the case for a limited version of responsibility which applies if Banner can prevent the metamorphosis into Hulk in the first place, but I think we have already moved beyond memory, so the fact that this special responsibility does not apply in the real life case she mentions is not decisive.

One point which I think should be added to the account, though it too is not decisive, is that the loss of responsibility may entail loss of personhood in a wider sense. If we hold that you are no longer the person who committed the crime, you are not entitled to their belongings or rights either. You are not married to their spouse, nor the heir to their parents. Moreover, if we think you are liable to turn into someone else again at some stage, and we know that criminals are, as it were, in your repertoire of personalities, we may feel justified in locking you up anyway; not as a punishment, but as prudent prevention. To avoid consequences like these and retain our integrity as agents, we may feel it is worth claiming our responsibility for certain past crimes, even if we no longer recall them.