artificial intelligence
respond
responses
|
[posted to a discussion list this morning] A couple of days ago I suggested a strategy for coming up with a definition of intelligence. I started with a list of beings that I consider intelligent and asked you all to contribute to it. So far on our list are humans dolphins ants ant-colonies bacteria plants cats rats sphex wasps Colin :) Next step would be to ask what it is that all those entities have in common. Instead, some of y'all are continuing to focus on who is smarter than who, who is not as smart as YOU, and what sorts of behaviors you find dumb. That's a fine project, but it's tangential to mine. You've gotten yourselves into an infinite loop; given that this is a philosophical discussion where philosophical rules of procedure are not being followed, it's no surprise, happens all the time. That's ok, if you need to get this out of your systems. I will go along with your tangent for a (rather long) moment. Hopefully I can demonstrate why it's a better plan to go back to the list of so-called intelligent entities and ask what they have in common. Don't let me stop you from engaging in your project, but I invite other people to continue with my project simultaneously. Some examples came to mind while reading about how to avoid looping behaviors. At the end of my list of examples I discuss why I don't see these as problems but rather as more data about intelligent beings and how to create them. 1. Person loses something. Person knows item was used last night, in the house, after coming in for the last time; so it must be in the house. Person searches house for item. When all possible locations for thing have been checked, they are all searched again. And again. And again. With increasing feelings of frustration. Feelings of stubborn determination increase with the feeling that, since so much searching has already occurred, person must now be very close to goal. Other members of the household are interrogated and enlisted in the search; members question person about all the places person just searched, just to make sure. This can go on for hours, or days, or WEEKS, even when it might be more economical to just buy a replacement item (like a pen, or a sock). 2. Person wishes to purchase a particular piece of clothing: a leather jacket, say. Person has standards about style and fit that are at odds with current fashion. Several pretty good jackets are observed, but none are the Ultimate Perfect Jacket, and so are passed up with a mental note that if no better is found, person will return to buy this one, EVEN IN CASES where the jacket is on closeout and is likely to be gone before return. Every area mall is searched, store by store, even where clothing is "organized" by label rather than by function (e.g., Nieman Marcus and Nordstrom, where they might stock 50 different leather jackets, but they are distributed over approximately 10,000 square feet and hidden amongst the ties, underwear, shorts, jeans, slacks, hats, suits, etc. Stopping for lunch or dinner would be unthinkable: hunger is irrelevant when the Perfect Jacket could be just around the next corner, purchase to be followed by celebratory dinner. Search on foot only ceases when malls close. Next trip includes long drives around city looking for leather stores, motorcycle stores, fashion boutiques, etc. Upon encountering leather store, every likely jacket is tried on, twice. During hours when store shopping is impossible, the internet is scoured. This search continues for two months. Just a little further, just one more jacket, just one more mall in one more suburb... 3. Boy invites girl on "camping trip", a three-week touring vacation to be taken on the way to a conference. Anticipating, girl chats happily with boy about cooking over and warming selves next to open fire, listening to night creatures and chatting about philosophy, then waking up and taking long wilderness hikes and swimming in mountain lakes, nerdily identifying birds with bird book. Boy and girl both pack novels, expecting to read lazily in scenic spots. Boy and girl jointly decide that route will proceed from Indiana south to Virginia, wind north along coast until destination in Toronto is reached. The first day, girl is not allowed to eat or go to the bathroom for 12 hours, because boy insists on driving as far as possible before camping for the night; tent is pitched in the dark, nuts and crackers are eaten for dinner. Next morning boy leaps out of sleeping bag at dawn and begins tearing down camp: 15 hours more of highway driving, one rest-stop is grudgingly visited. Girl complains of hunger, explains female need for frequent bladder relief, and reminds boy of desires to come in contact with wilderness. Boy listens, nods, then gets on highway and drives 15 hours, to get as far as possible before eating nuts and camping for the night. Following this pattern, conference destination is within reach by the third day of the three-week tour. Now touring can begin, says boy. Begin daily 12-hour drive back and forth across northeastern states, to get as much driving in as possible before camping for the night--every day, despite louder and more frantic complaints from girl, for the remainder of the three-week "vacation". Loop ends only when conference begin-date arrives. 4. Person is raised by alcoholic parent, observes irrational, risky, abusive behavior, swears off alcohol entirely, and vows never to consort with alcoholics in the future. Meanwhile person gets very good at putting up with, excusing, disguising, sympathizing with, and caring for alcoholics, a fact readily observed by alcoholics. Surprise! First lover is alcoholic. Person endures behavior similar to parent's, eventually ends relationship, puzzled by failure to recognize and predict problems with lover, renews vows to never drink and never go anywhere near alcoholics in future. Friends note that second lover is "just like" first lover, person resents friends interference, second lover indeed turns out to be alcoholic, person breaks up with lover and apologizes to friends... Repeat. Repeat. For years, perhaps for decades. See extensive literature on this phenomenon (or look around at friends :). 5. Person is rude, mean, and unfriendly. Person can't get laid. Person imagines that other people are deterred by person's frightening intelligence. Good friends try to help, tell person that it is not intelligence, but overall ickiness, that repel opposite sex. Person briefly considers shaping up in order to get laid, decides it would be "dishonest": This is who I am, I'll find someone who LIKES rude, mean, unfriendly people!" Remains bitterly single for long stretches. Eventually finds lover similar to self, ends relationship because person dislikes rudeness, meanness, and unfriendliness in other people. Tries being (dishonestly!) nice, gets laid, eventually drives lover away by reverting to repulsive behavior. Loop continues for entire lifetime. I have never met a single human being who is free of this kind of looping in *some* area of his or her life. I ask myself what that means, what it indicates about the way the human mind works. Y'all seem to be looking at "lower" animals for examples of irrationally repeated behaviors, but if you try to be a little more objective, you'll see it all around you; if you're a lot more objective, you will see it in yourself, too. I have no problem with the project of making a being that is "better" than existing human beings. But my guess is that, before we can do that, we need to at least be able to talk about how we would make one that is *as good as* human beings--or even as good as a sphex wasp. When I see behaviors like the examples I listed above, or like the sphex wasp being tortured by irrationally repetitive entymologists, or tigers pacing back and forth in tiny cages, my first thought is, What is the underlying explanation for this looping? Why do I see it in every instance of every category of creature that I think is intelligent? My hypothesis is that the strategy is sound. It works well the vast majority of the time, however the given creature defines 'works well' for itself. And then in a few weird cases, it doesn't work well at all by anyone's definition of 'works well'. Rather than say, "Look, this strategy sometimes fails! Creatures who fail this way aren't intelligent--or at least aren't as brilliant as ME! Let's make a creature that will never fail in this way!", I say, "Look, there is a reason that every intelligent creature, INCLUDING ME, sometimes fails in this way, yet succeeds in its projects most of the time--what is the strategy that usually succeeds but sometimes leads to this failure, and how can I distill the strategy down to its essentials and reproduce it, accepting that it will sometimes fail?" (I anticipate objections: No, I'm not deliberately trying to build something THAT FAILS. I am NOT following a research strategy based on the principle "to err is human". After all, there are an infinite number of ways to get things wrong--how do I know that any random way of getting things wrong is just the way that intelligent creatures tend to get it wrong? Instead, I'm asking why something that usually succeeds sometimes fails in very specific ways. If I can build something that usually succeeds by a given strategy but sometimes fails in just these sorts of ways, then I'm doing it *right*. I suspect that a consequence of proceeding in this way is that I need not worry ahead of time about what I will do when I get into an infinite behavior loop. I've found that what usually happens is that, when I do it right, these sorts of terrible problems you are trying to hedge against either don't come up at all or are trivially handled by the agent itself.)
|