For example, every infant is born with such a system for maintaining ‘normal’ body temperature: when too hot, the baby may sweat, pant, stretch out, and/or vasodilate; when too cold, it will curl up, shiver, vasoconstrict and/or raise its metabolic rate.
At first we may be unaware of such processes, because such instinctive reactions begin at very low cognitive levels. For example, when you become too hot, you automatically start to sweat. However, when perspiration drips, you may notice this, and deliberate: “I must find some way to escape from this heat.” Then your acquired knowledge may suggest other actions to take, such as moving to an air-conditioned place. If you feel too cold, you might put on a sweater, turn on a stove, or begin to exercise (which can make you produce ten times as much heat).
Now we can interpret “having a goal” to mean that a Difference-Engine is actively working to remove those differences.
Student: To have a goal, does one really need a representation of the desired situation? Would it not be sufficient just to have a list of desired properties?
This is a matter of degree, because one could never specify every aspect of a situation. We could represent a ‘desired situation’ as a simple, rough sketch of a future scene, as a list of a few of its properties, or as just some single property (for example, that it is causing some pain).
Student: It seems to me that we should distinguish between ‘having a goal’ and actively ‘wanting’ it. I would say that your difference-engine is a “wanting machine’ and that the goal itself is only the part that you called its ‘aim’—its current description of some future situation.’
An imagined description becomes an active goal when one is running a process that keeps changes conditions until they fit that description. Our everyday language does not serve well for making the kinds of distinctions we need and, to remedy that, each technical field must develop its own specialized language or ‘jargon.’ However, it will be hard to do this for psychology until we can agree on which more detailed models of minds to use.
Romanticist: This Difference-Engine idea could account for some of what “having a goal” might mean—but it doesn’t explain the joy of success, or the distress that besets us when we fail to achieve what we’ve hoped for.
I agree that no single meaning of goal can explain all of those cascades of feelings, because wanting is such a large suitcase of concepts that no single idea can embrace them all. Besides, many things that people do come from processes with no goals at all, or goals of which they are unaware. Nevertheless, the Difference-Engine’s characteristics capture more of our everyday concept of ‘goal’ than any other description I’ve seen.
Student: What happens when that difference-engine finds several differences at once? Can it work on them all simultaneously, or must it deal with them one-by-one?
When several differences are evident, one might try to reduce several at once, perhaps by using different parts of the brain. However, Newell and Simon concluded that it is usually best to first try to remove the one that seems most significant, because this is likely to change quite a few of the others. So the General Problem Solver included a way to assign a different priority to each kind of difference that it could detect.
Student: Isn’t that a flaw in that? What if Carol places a block in a place that prevents her from building the rest of her arch? Sometimes reducing one difference might make all the other differences worse.
That turned out to be a severe limitation, because a Difference-Engine, by itself, has no way to plan several steps ahead—for example, by the methods suggested in §5-5—so it cannot sustain a short-term loss for the purpose of later, larger gains. So, although their system could solve many problems, this limitation seems to have led Newell and Simon to move in other directions.[111] In my opinion, they should have persisted, because this project had so many good ideas that I find it strange that it was not further developed in later years. In any case, we can’t expect any one method to solve every problem—and our forthcoming project will try to embody the concepts that Newell and Simon abandoned.
In retrospect, one could argue that the system got stuck because it was not equipped with ways to reflect on its own performance—the way that people can ‘stop to think’ about the methods that they have been using. However, in a great but rarely recognized essay, Newell and Simon did indeed suggest a very ingenious way to make such a system reflect on itself.[112]
On the positive side, the General Problem Solver was equipped with several ways to reduce each kind of difference, and it even included a place for ways to introduce new kinds of representations.
What if one fails to solve a problem, even after using reflection and planning? Then one may start to consider that this goal may not be worth the effort it needs—and this kind of frustration then can lead one to ‘self-consciously’ think about which goals one ‘really’ wants to achieve. Of course, if one elevates that level of thought too much, then one might start to ask questions like, “Why should I have any goals at all,” or, “What purpose does having a purpose serve”—the troublesome kinds of questions that our so-called “existentialists” could never found plausible answers to.
However, the obvious answer is that this is not a matter of personal choice: we have goals because that’s how our brains evolved: the people without goals became extinct because they simply could not compete.
Goals and Subgoals
Aristotle: We deliberate not about ends, but about means. … We assume the end and think about by what means we can attain it. If it can be produced by several means, we consider which one of them would be best …[and then] we consider by which means that one can be achieved, until we come to the first cause (which we will discover last).[113]
Section §2-2 considered some questions about how we connect our subgoals to goals—but did not stop to investigate how those subgoals might originate. However, a Difference-Engine does this by itself because, every difference it needs to reduce becomes another subgoal for it! For example, if Joan is in Boston today, but wants to present a proposal in New York tomorrow, then she will have to reduce these differences:
The meeting is 200 miles away.
Her presentation is not yet complete.
She must pay for transportation, etc.
Walking would be impractical because that distance is too large, but Joan could drive, take a train, or an airplane. She knows this ‘script’ for an airplane trip:
111
See Allen Newell and Herbert Simon (1972),
112
See A. Newell. J. C. Shaw, and H. A. Simon, “A variety of intelligent learning in a general problem solver,” in
113
In Nicomachean Ethics (Book III. 3, 1112b). This appears to be a description of what today we call ‘top-down search.”