Jason returns

By Harry van Vliet on 30 March 2011

Last week I read two PhD proposals on how ‘smart’ agents and microsystems can do their work more ‘intelligently’ if they become ‘aware’ of their surroundings. This is not an uncommon way of putting things: we have ‘smart’ phones, ‘intelligent’ video recorders, and security systems that are ‘aware’ of a possible breaking and entering. But there is a difference between using these words in everyday life as a shortcut to express ourselves and using these words in a scientific context. In the latter you commit yourself to explain what exactly you mean by it. Mostly this boils down to references to ‘schemata’: knowledge structures that are used to infer what the situation is and how to behave in it. The ‘restaurant’-schemata of Schank & Abelson is one of the most famous.

Now let me be clear about my arguments: I am not going to argue against the ‘existence’ (whatever that may be) of organized knowledge structures. What I will do is introduce doubts about the explanatory value of concepts such as frames, conventions, scripts and so on. Maybe you share some of these doubts. Maybe you will begin to share them as I recite answers to some of the questions one might come up with:
How do I activate a schema? It is done automatically!
How do I know which schema I must activate? You just know!
When do I activate a ‘literary schema’? When you are reading literature!
What if I don’t activate a literary schema? You are not reading literature!

The main problem as I see it concerns the aspect of activation. Somehow, knowledge from the back of our minds becomes manifest and is used to comprehend. Somehow, we know what to activate and how much of it we must activate, and we know what is irrelevant and can be ignored for the moment. Talking about scripts, schemata and frames does not provide any real answers to this problem, since they all fall short of explaining how activation is achieved, and therefore what it means to comprehend – or to think, for that matter. To elaborate somewhat on the problem let us take some examples and arguments from what has become known as the discussion
on the frame problem.

The frame problem is made up of two interdependent problems: a problem of quantity and a problem of quality. The problem of quantity concerns how we know that enough knowledge has been activated to comprehend the text or situation at hand. It is the problem of when to stop thinking. Jerry Fodor refers to this as Hamlet’s problem. As an example (taken from Daniel Dennett), let me introduce Jason, a robot. Suppose you constructed this robot, giving it just one task: to fend for itself. One day you arrange for it to learn that its spare battery is locked in a room with a time bomb set to go off soon. Jason formulates a plan to rescue his battery. He is told there is a wagon in the room, and that the battery is on the wagon. Jason hypothesizes that a certain action, which is called PULLOUT(WAGON, ROOM) will result in the battery being removed from the room. Straightaway he acts and does succeed in getting the battery out of the room before the bomb goes off. Unfortunately, the bomb is on the wagon: that is, Jason has missed the obvious implication of his planned act, that pulling the wagon would bring the bomb out along with the battery. So much for Jason it seems. Not quite. The solution is obvious, you say to yourself, Jason-2 must be made to recognize not just the intended implications of his acts, but also the implications of their side effects. When placed in the same predicament as the original Jason, Jason-2 begins, as designed, to consider the implications of his course of action. As Jason-2 has just finished deducing that pulling the wagon out of the room would not change the color of the room’s walls, and was embarking on a proof of the further implication that pulling the wagon out would cause its wheels to turn more revolutions that there are wheels on its – when the bomb exploded. Back to the drawing board. Jason-3 must be taught the difference between relevant implications and irrelevant implications and to ignore the irrelevant ones. Jason-3 sets out on his task, then stops, to your surprise, outside the room with the ticking bomb. ‘Do something’, you yell. ‘I am’, Jason retorts, ‘I am busy ignoring some thousands of implications that I have determined to be irrelevant. Just as soon as I find an irrelevant implication, I put it on the list of those I must ignore and …’ – the bomb goes off. One could go on designing Jasons but somehow he is never bound to come out of it alive. Of course, one could programme Jason to take the battery without pulling the wagon, but that is missing the point completely. You set out to let Jason think. If you tell Jason he must not pull out the wagon, you have solved the problem and Jason is just following orders. That is, you did the thinking part and not Jason. What does this tell us? It tells us that our normal mode of being should be being buried in thought all the time, checking all of the possible implications of our plans. The explanation of why this isn’t so is the same as explaining why we stop thinking at a certain point and start acting. We cannot expect an answer from our scripts, schemata and frame-followers. Remember that they say that comprehending consists of the activation of relevant schemata. But since you designed Jason-3 you know that the aspect of relevancy is superfluous in coming up with an answer. As Minsky stated, “by adding axioms about relevancy, we produce all the unwarranted theorems, plus annoying statements about their irrelevancy”. It did not save Jason, so why should it save us?

(this text is largely taken from ‘Frames of Mind’, published in Bright Lights, Blind Spots, there you can read the remainder of the argument…)

tagged with: artificial intelligence, frame problem, schemata, smart

Comments