THE COMPUTER'S SUBCONSCIOUS

Abstract

Human beings are not rational planners. Decision-making is unconscious and rationalized consciously after a decision has been is made. Cases, known consciously and unconsciously, drive the human decision-making process. Any intelligent computer system that ignores how people make decisions will behave in a fashion that, while logical, may miss the forest for the trees. Real decision making relies on the power of a complex, well-indexed case base.

INTRODUCTION

Planning in artificial intelligence (AI) has often made the assumption that the modeling of human planning would be a good place to start. Unfortunately, it is not clear that AI folks have paid all that much attention to how humans actually plan. The common assumption in AI planning work which models humans is that people make conscious decisions, thinking out what they intend to accomplish and how they will achieve their goal. Certainly, Newell and Simon's (1972) work on protocol analysis made this assumption and this assumption stayed the same throughout all the work on expert systems. But this notion of conscious planning does not really serve as an accurate picture of human planning. We like to believe that people think things out, but do they?

If people created new plans for every situation that they encounter, working from first principles as myriad books on decision-making suggest they do, planning would look different than it does. We would just know how to plan and according to the issues at hand, create plans that work for a given situation. But it seems more likely that people do not actually create plans—they adapt plans. Without realizing it, people have access to a plan library and they choose ones they think will work. The plan library we use is usually not known to us (with the obvious exception being the military and sports, where plans are written about and chosen from).

Suppose, for example, that you want to visit your mother. You know where she lives and you have been there many times before. You know various methods of getting to her house. Which one you choose depends upon many factors, money perhaps, time perhaps, traffic perhaps, how long you plan to stay perhaps, what else you plan to do enroute perhaps, who else will be coming along perhaps, and so on. Each of these conditions causes you to select one of your known plans, that is, a method you have used before, as the best plan for this particular situation.

You have a playbook for visiting mom but you don't think of it like that. You might not even be able to elucidate all the plans you know about. It doesn't work like that. You simply choose the right one without seeming to think about it much. Planning is like that. We select from a list without actually looking at the list.

This works most of the time and we are usually happy with the results. This ad hoc method of planning, the one we use on a daily basis, I call, “selecting from the non-conscious playbook.” The problem with this planning method is that when it fails it tends to fail big. Now, how much can you fail in the selecting of the right plan for visiting Mom? You can say, after the fact, “I should have taken the turnpike.” But, in the end maybe you cost yourself 15 minutes of extra drive time. Failures in planning don't matter that much a lot of the time.

Sometimes, they matter a great deal. But, since we are in the habit of planning by using the nonconscious playbook, we continue to do what we have always done: not think about it much. The proper plans are obvious, except sometimes they aren't.

Decision-makers fail to make good decisions for a variety of reasons. Some of these are:

1. They are not in possession of all the relevant facts.

2. They have not completely followed the logic of a decision.

3. They have not considered the likely response to a decision (in a competitive environment).

4. They have not done the math correctly.

5. They failed to appreciate how events can interact.

There are many other reasons why decisions turn out badly. But these are ones that depend on getting the facts right and understanding the implications of the facts.

There are many books that claim that they can help with decision-making. A book called Smart Choices (Hammond, Keeney, and Raiffa 1999) suggests that one can use risk profiles to simplify decisions involving uncertainty. They suggest that one should first construct a risk profile that compares the likelihood of possible outcomes, then identify the key uncertainties, then define the outcomes, then assign chances to the different possible outcomes, and lastly, one should clarify the consequences of the different outcomes. Somehow I don't think this would help anyone who actually doesn't have plenty of time allotted to plan construction. But, of course, computers have the time for this. So it seems an easy way out for AI programs. But do people actually plan like this?

What these authors are describing is how to make a conscious decision. I am not sure what they suggest is anything more than the obvious but it makes no difference because very few people ever have the time or the inclination to make a conscious decision. That is why these books exist: to remind you that you are being irrational in your decision-making. The question for people is: can you become rational? For computers the question is: if people are not normally rational is the right course to build computers that are rational and hence improve on the human planning process? Artificial intelligence people love to improve on humans so this seems to be an easy question to answer. But is it?

People see themselves as making rational decisions of course. The more information they have the better. They want help in making better decisions and they want to be able to follow logical rules for deciding. That's why all these decision-making books exist. The fact that this doesn't work attests to why there are so many of these volumes.

AI research has long held the view that people plan and that a smart computer program would be able to plan as well. Much work has been done in getting computers to plan logically. The problem is that the premise is wrong. There really is little evidence that people are capable of actual planning (as opposed to the ability to select from a list of plans).

Selecting from a set of plans and actually planning are two different things. When a sports coach or a military official invents a new play or a new strategy, you might look at it as if it were a new plan. But no new play or strategy is really original after all. They are all variants of existing plans. In fact, that is why they work when there is an extant counter plan. If a play or strategy looks like another play or strategy, it might be misconstrued by defenders and they might be lured into a trap. Thus, no new plan in a counterplanning situation is, or ought to be, wholly new. A good way to create a new plan is by tinkering with an old one.

Planning is never an original act. My assertion is that planning, or more accurately the ability to plan, is an illusion. Similarly, the idea that we can make rational decisions may also be an illusion.

But, of course, people believe that they plan and they believe that they have thought carefully about something and have made a rational decision. In fact, they may spend a fair amount of time planning and deciding.

Recently, a man that I know who is pretty intelligent but rather uneducated, asked me for advice. His situation was difficult. He lives with his wife, who used to be a stripper, her daughter, who is 16, and a 6-year-old son that he had with his wife. The daughter had run away from home (although he knew where she was). The son had been sexually molested by the daughter's boyfriend. He had come to ask me whether he should kill the boyfriend.

Not an easy decision of course. A horrific question in fact. But it was, I am sure we can agree, a planning problem. A decision was called for, and in asking my help, he was hoping that the decision might be rationally made.

This, to my mind anyway, is what planning looks like for the average person. I am not saying killing is a typical plan. Rather, I am saying that dealing with this level of complexity is common, and attempting to plan a solution is something that happens fairly frequently in people's lives. This is the stuff of what we call thinking. He was trying to think about what to do.

The story was simply that his daughter had been seeing this guy (who is 15) for some time. He hated him from the first and thought he was weird and effeminate. He had left the daughter minding her brother and the boyfriend visited. There had been violence before between boyfriend and girlfriend and he had forbidden these visits. After the 6-year-old reported the molestation to him, he confronted his stepdaughter but she denied it ever took place. The 6-year-old said she was sleeping at the time. His son told him things that he believed he could not possibly have fabricated. His son had trouble sleeping for a week after the incident and that is when he told his father about it. By this time his stepdaughter had run away.

His response was to go find the 15 year old and beat him up. On his way to his house, he ran into his stepdaughter. He said that he took this as a sign of some sort—that fate had intervened somehow—and he began to deal with his daughter instead of beating up the 15 year old. He was now coming to me for advice on what to do about the situation. Apparently he had already contacted people who could do away with the kid or at least force him and his single parent (and apparently drug-addicted) mother to move out of the state.

My friend had a set of plans available to him and he was selecting among them. What does one do when one's stepdaughter's boyfriend molests one's 6-year-old son? While “killing” may not be a plan of yours or mine, it was a plan in my friend's world and he had it among others to select from.

My problem was to get him to choose from something other than the “revenge” plan library. The “getting out of harm's way” plan library that I would have used had little appeal to him however. Either way, we are talking plan selection here, rather than plan creation. The end result of any conversation we had might have had created an entirely new plan in the sense that it had never been enacted before in the same way that the sentence I am writing has never been written before. But it wouldn't have been all that novel.

And of course, no decision is all that novel either. Most decisions are about things that have been decided hundreds of times before, like where to eat or what to do. Any decision is really the result of a selection process.

But what happens when the decision is entirely novel and important? Presumably my friend had never killed anyone before, so this was a novel idea to him in some way.

It is novel problem-solving which is, or ought to be, the province of AI of course. The “Tower of Hanoi problems thinking” which has dominated AI in the last 50 years has tended to misfocus the community on non-novel (and inherently logical) plans, when novel planning is the real issue.

Let's consider marriage. People usually spend some time on this once-(or twice) in-a-lifetime decision.

My daughter recently got married. She married a man who, while they were dating was undecided as to what his professional life should be. While they were dating he decided, with some encouragement from her, to become a professor. He is now in graduate school as I write this. My daughter was not entirely thrilled with this choice. She realizes that as a humanities professor he is unlikely to earn a great deal of money and this means she will always have to work. She is okay with this some of the time, but some of the time she is not.

The man my daughter married is of more or less the same socio-economic ethnic background as she is. He is also 3 years younger than she is.

The other day we were discussing her choice of husbands now that the dust has settled a year later. I asked if she thought it interesting that she had selected to be the wife of a professor (as her mother was) and to marry someone 3 years younger than her (as her mother did). She said she thought it was sort of random. I responded that marrying a 65-year-old truck driver from Kentucky would have been random and that this was anything but. She asked if I thought that there was something inside her saying that marriage meant marrying someone 3 years younger and that fathers of children should be professors and I said “of course.” She said that she always thought maybe it was important that her brother (her only sibling and frequent playmate in childhood) was also 3 years younger than her. “I wasn't looking for that,” she replied. “That's what you think,” I responded.

So, how did she make this decision? Well, I know one thing. My daughter has only the vaguest idea about how she did it. She had no available plan library to select a plan from. When that happens, all hell breaks loose. The subconscious takes over. Computers do not have a subconscious of course. Hmm. Is this a good thing or a bad thing?

The creation of an available subconscious in AI has never been seriously considered, or even mentioned, as far as I am aware. Yet, people cannot plan without relying upon subconscious memories, desires, needs, etc. Should we just say that the modeling of humans is irrelevant in AI or should we think for a while about how people really plan?

What we do know, is that when people select from their plan library, they are often not at all sure what the optimal plan is in that library. When there is no library at all, people get confused and ask for help (like my friend did). When subconscious drives are strong people do not ask for help. They “know” what they need to do. Unfortunately, they don't usually know why they need to do it.

DREAMING TO PLAN

I was intending to call my mother's doctor. My mother is 88 years old and losing it mentally. One cause of her confusion I thought, might be her medication. She takes clonapin for anxiety. I really did not see why she took this drug but she insisted that the doctor told her to take it and she was going to take it. She said that I could call her doctor and discuss it and she would abide by the decision. I kept putting off the call. I hate talking on the phone and I dislike doctors, particularly ones who prescribe drugs to old people without thinking very hard about what they are doing. I simply did not make the call. I found myself speaking to my mother less because the conversations were becoming incoherent.

One night just before I was to visit her and have yet another odd meeting with this increasingly confused woman, I had the following dream. I dream that my daughter was having difficulty with her medications. I was very upset about this and I called her doctor. I said the doctor needed to take her off clonapin and that it was absurd that she was taking this drug. He pointed out that she was not taking clonapin but said that I had made a very good argument for why she should not be.

I thought in the dream that this conversation with the doctor had gone very well. When I awoke I realized that I had been practicing the conversation with my mother's doctor in easier circumstances. It is easier for me to inquire about my child's health than my parent's health. Talking to your child's doctor seems much more natural and I am used to it. Further it was easier talking to a doctor who had not made a serious error in prescribing this drug. I awoke thinking that I needed to call my mother's doctor today and that the conversation would not be so difficult for me.

This kind of “dreaming to plan” is very common, common enough that one can find it in literature and in the movies quite frequently. In one movie, “Willie and Phil” (1980) a woman with whom they are both involved, sharing a house in California, says:

I had a dream. Remember that first day we met? I dreamt that I was in Washington Square Park—and I was lonely, I wanted to go to the movies—only nobody showed up—no Willie, no Phil, but Daddy was there. And I was 10 and I was sitting on my pony and all of a sudden I was crying because everyone was laughing at me, so I just started to ride away. I rode faster and faster until I was flying. I was flying over Manhattan and Daddy was screaming at me to come down—that it was real dangerous and I could get killed, but I just kept right on flying. I mean there was no way I was going to come down. I wasn't afraid of anything.

I called Helena a few minutes ago and I told her I would go to work in New York. I am taking Zelda. I have no idea what's going to happen next but I know I am doing what I have to do.

Although this is fiction it has a sense of reality to it with respect to how people create plans. They dream and then they see what to do next. This usually works much faster in the movies than it does in real life, but the process is the same. What we have is a case of the conscious hearing what the subconscious is thinking. The conscious defers to the subconscious as being the better planner.

Should computers dream? I wonder how we could even think about the creation of intelligent entities that do not dream. Of course, it helps to know why they should dream before we build in this kind of ability. One reason is to create and test out novel plans in fictional circumstances prior to having to execute a real plan. This kind of fictional ideation is critical to thinking, but yet again, not normally part of AI research.

RATIONALIZATION OF SUBCONSCIOUS DECISIONS

Why do people choose the mates they do? Countless books have been written on choosing the right life partner. But all the advice in the world will not help you avoid certain ways of choosing what you have little or no control over. Here is something from Harville Hendrix (2001), the author of Getting the Love You Want:

People try to exorcise their denied negative traits by projecting them onto their mates. Or, to put this another way, they look at their partners and criticize all the things they dislike and deny in themselves. Taking a negative trait and attributing it to their partners is a remarkably effective way to obscure a not-so-desirable part of the self.

Of course, no one thinks, “I am a really angry person so I try to repress my anger but if I marry someone who is really angry just like dear old dad then maybe I can work on my anger issues.”

Nevertheless, this is exactly what Hendrix says he sees in patients all the time. People manage to make one of the most important decisions of their lives—who to marry—by:

1. not understanding why they have made the choice they have made;

2. choosing the exact opposite of what they really need;

3. attempting to satisfy deep-seated psychological issues they have repressed by way of the decision;

4. not really comprehending who they are trying to satisfy with the decision they have made;

5. choosing to live with an unfortunate decision despite the fact that it is very painful to do so.

If people do all this for big decisions, is it simply an anomaly, or is it possible that they behave in this way for nearly all decisions? If they do behave this way for less important decisions, or if they have real power (like being President of the U.S., for example) for potentially even more important decisions, then their decision process would seem to be really broken.

To put this simply, if the subconscious is making the decisions and the conscious is rationalizing them, what kind of process is this? Is it possible that the subconscious is really bad at decision-making and cannot be reasoned with? Really important decisions ought not be left to a decision-making process that is mostly concerned with healing old childhood injuries. Yet, this is pretty much the way it is. Books detailing how to make rational decisions using utility theory will not help. The subconscious can't read.

Nevertheless, business people who apparently read a great deal about decision-making (given the number of books out there on the subject) have been talking about both rational decision-making (using a kind of calculus for making choices, e.g., Smart Choices by Hammond, Keeney and Raiffa) and emotional decision-making. This latter idea stresses using “seat of the pants” gut reasoning. “Many executives have learned to tap into their right-brained thinking by jogging, daydreaming, listening to music, or using other meditative techniques” (Hayashi 2001).

So the good news is that people are starting to realize that their subconscious is the real decision-maker and that they are allowing the subconscious to make the decisions. The bad news is that the subconscious is none too good at this.

Hendrix, in discussing how the nonconscious controls perception, says; “The reason that we are such instant judges of character is that we rely on what Freud called “unconscious perception.” We intuitively pick up much more about people than we are aware of. When we meet strangers, we instantly register the way they move, the way they seek or avoid eye contact, the clothes they wear,…in a matter of minutes. Just by looking at people we can record vast amounts of information.”

But we also record whether they seem to remind us of the parent we had trouble with, according to Hendrix, and therein lies the rub. How can we make sensible decisions if we are guided by a subconscious that is busy worrying about why it wasn't loved enough as a child? Hendrix asks how we can make better mate choices given this issue. I am asking how we can decide anything at all sensibly, given this issue. To put this another way—perhaps the reason people have such difficulty thinking clearly about most anything is that they have trouble separating the details of an issue from the deep psychological issues that are lurking in the background. The subconscious does the thinking, and it is not particularly concerned with the issues at hand.

A trivial example of this happened to me recently. I was deciding on dessert in a fancy restaurant in England. Instead of dessert they offered a “savourie” called Welsh Rarebit. I immediately wanted Welsh Rarebit. It sounded great. I hardly ever saw it on a menu and I was delighted to have something I knew I liked. My wife asked me why I wanted what amounted to the “cheese on toast” she remembered from the Stouffer's frozen variety. I said I didn't know, but that I wanted it, and this being a fancy restaurant, they were probably not serving the Stouffer's offering.

Between this conversation and the arrival of the food, I had some time to think. Suddenly I was thinking about train rides and eating on a train. I realized that I had first eaten Welsh Rarebit on a train to Florida when I was 9. The food was an odd one, I liked the name, but I also had never seen it before, so it stuck in my memory. But what stuck in my memory more was eating on a train. It was a very exciting experience for a 9 year old. Eating soup while the train rattled along was great fun.

I realized while I awaited my food, that I was ordering Welsh Rarebit because of a long ago train ride. This didn't seem a good reason for making this ordering decision, but neither was it a bad one. The Welsh Rarebit was extraordinarily good by the way—nothing like either the train or the Stouffer's variety.

If unconscious psychological influences determine small decisions like what to have for dessert, how can they fail to be part of more significant decisions? And if we fail to comprehend why we decide what we decide, are we likely to be making the best decisions? The subconscious child in me was demanding Welsh Rarebit and the adult gourmet in me was acceding to his wishes. So what? It was just dessert. But sometimes it isn't just dessert and the same interactions apply. In that case very unfortunate decisions can be made.

It looks therefore, that the lesson for AI is: stay as far away from modeling the human subconscious as possible. Nothing good comes out of it for humans and surely no computer need imitate this nonsense.

But, is this right? Is there no value in satisfying subconscious desires? Is this simply a bug in human reasoning and decision-making ability?

DREAMING TO REASON

People reason using cases. One situation reminds you of another. While thinking about what to do in the new situation, the old situation that comes to mind serves as a guide, sometimes about what worked in the past and sometimes about what did not.

I have written about case-based reasoning (CBR) many times in the last 20 years (starting with Schank (1981)). There are many people within AI who think about CBR. Case-based reasoning is a good way to make computers reason better. But what about people?

People use cases in two important ways. They use generalized cases to help them deal with common situations and they use specific cases to help them deal with new or problematic situations. I always find it convenient to use restaurants to explain both types of CBR.

Let's imagine that I decide to take you out to a restaurant and that you have never been to McDonald's so I decide to take you there. You arrive expecting a restaurant of the standard sort. You have a generalized restaurant case available (usually I call these scripts). In this case, compiled from all your restaurant experiences combined, there are tables and hostesses and menus and waiters and checks and a set of events that take place in a certain order. You have compiled this set of expectations about what takes place in a restaurant from years of experience. These expectations are not necessarily thought about as you enter the restaurant, but they are conscious nevertheless. If I asked you what you expected was going to happen after we were seated by the hostess you would be able to tell me that you expected someone to come over and take your order. This knowledge lives within your restaurant script which is, as I said, a generalized case.

But I have taken you to McDonald's so none of this is going to happen. In fact, every expectation that you have about what will happen next at any given time will be violated. There is no hostess, there are no menus. You stand in line to order. No waitress brings you your food. You have to pay before you get your food. You feel confused. You say that this doesn't seem like a restaurant at all.

The next day, I say that if you liked McDonald's you will surely like Burger King. As we enter you say that you are reminded of McDonald's and you bring up your expectations, newly formed from yesterday's experience, about how to order and when to pay. This is called learning. You have corrected some failed expectations, derived from an inapplicable generalized case, and you have substituted new expectations derived from a new specific case.

Case-based reasoning is neither good nor bad; it is simply how we reason from prior experience. Each new situation causes us to look for guidance from prior experience. Sometimes we have exactly the right prior experience to guide us. Something just like this has happened before and the conclusion was successful so we do the same thing. Sometimes what happened before had a bad outcome so we use the case as guidance on what not to do. And quite often, we are in a situation we have encountered so many times before that we do not think about previous cases at all but our unconscious does and it causes us to expect certain things. We only notice these expectations when they fail.

Case-based reasoning works well when there are many different cases to reason from. When someone has had only one prior case, on a second airplane trip, for example, one expects the new experience to copy the old one exactly. When it does not, we adapt and decide (unconsciously) whether we now have two cases or one case with a minor variation. Every new airplane trip causes us to build a stronger generalized case and to store and be reminded of, odd experiences within that case.

Of course, reasoning from limited cases can be a disaster because they are often not really relevant, but simply all we have. Is Iraq like Vietnam? Probably not. Nevertheless it is difficult not to think about previous-like situations when reasoning about what to do in a new one. So sometimes, CBR can mislead you, but cases are all we have to reason from, so that is what we use. The more cases one has in one's memory, the better one is at reasoning about a given situation, assuming, and this is a big assumption, that one has categorized these cases in a useful way. The categorization process, labeling what you have just experienced so that you can find it again when you need it, is a large part of what it means to be intelligent.

The accumulation of cases in our memories is the basis of all decision-making. Whenever we decide to do anything, we are making a prediction about how what we actually do will turn out. We are, in essence, predicting the results of our actions. Unless we are making these predictions on some scientific basis, scanning probability tables using statistics, for example, we are simply predicting by finding what we have deemed to be the most relevant case we know about. Prediction of future outcomes depends upon past outcomes, no surprise here.

But where do cases come from? Certainly they come from experience. But, of course, not all our experiences are accessible. We don't have perfect access to our experience base. The subconscious may have better access.

Cases can come from the experiences of others as well. We hear stories. We watch movies. Other people tell us what happened to them. These too become cases we can reason from if these cases were very well told, or had some emotional impact on us.

But there is a third source of cases that is very important. These cases exist in memory but the experience that they are based upon may never have actually taken place. Here is something curious that Freud has to say (from The Interpretation of Dreams, pp. 44–45):

It may happen that a piece of material occurs in the content of a dream which in the waking state we do not recognize, as forming a part of our knowledge or experience. We remember, of course, having dreamt the thing in question, but we cannot remember whether, or when, we experienced it in real life.

If we cannot remember whether we have actually experienced something in real life, how can we determine if what we recall as a case to reason from that has actually happened? Perhaps we are reasoning from a total fantasy. Or, perhaps we are reasoning from something that did happen but that conscious minds are unaware of having happened. Freud (1900) comments on quite a few dreams that were reported where the dreamer dreamt about some fantastic things that only years later they realized were actually real events in their memories:

The fact that dreams have at their command memories which are inaccessible in waking life is so remarkable and of such theoretical importance that I should like to draw still more attention to it by relating some further “hypermnesiac” dreams.

Freud relates a number of dreams to illustrate what he means. One of these, as a typical example, is a dream that Freud read about where the dreamer dreams of a beautiful woman in some detail but upon waking, although he could still recollect her face, he had no idea who it might be. He went back to sleep, and asked her who she was, and she replied that they had met at the beach in Pornic. Upon waking from this dream he was able to recall the experience.

Cases can come from dreams and dreams can come from cases. While Freud is discussing here inaccessible events only accessible through dreams, the opposite is obviously also the case. Dreams can create cases that never happened. These are also potentially part of our case library. So, while we reason from cases, the cases have multiple origins, not all of them steeped in reality.

One issue is how nonconscious cases, like dreams that reside in our memories as cases, and like cases that reside in our memories only accessible in dreams, can drive our decision-making process. Another issue is what drives the dreams we create. Freud again (p. 593):

Unsolved problems, tormenting worries, overwhelming impressions—all of these carry thought-activity over into sleep and sustain mental processes in the system that we have named the preconscious. If we wish to classify the thought impulses which persist in sleep, we may divide them into the following groups: (1) what has not been carried to a conclusion during the day owing to some chance hindrance; (2) what has not been dealt with owing to the insufficiency of our intellectual power—what is unsolved (Freud lists three more of these).

So, we are left with an image here of problematic and unfinished cases, ones therefore of some real significance to a future decision-maker, being resolved by fantasy of what might have happened or what the dreamer might have done. To put this another way, we may well be relying upon, when reasoning from cases, on cases that were very difficult and important for us to resolve, that we resolved by making the answer up!

My point, of course, is not whether reasoning of this kind is good or bad, as I have said, but merely that this is how reasoning takes place. Taking this further, we can well imagine that when we seek a mate we are looking for someone with certain qualities that we have found to be very pleasant and satisfying in our dreams! What a way to make a marriage decision. Seems to be a bad way to decide things. We even have an expression for this—“looking for the girl of my dreams.”

On the other hand, we have all these utility means end analysis folks who would have you lay out the positive qualities that you seek and the negative ones you would like to avoid and compare each new candidate to see how well they score on your checklist. I once had an ex-cousin who did exactly this. He was deciding on whether to divorce my cousin and wrote down reasons for divorce versus reasons against divorce on a piece of paper (which my cousin later discovered and which precipitated the divorce). Curiously, he is a clinical psychologist.

So, the opposite kind of decision-making process, the supposedly analytical one, really doesn't seem like a vast improvement over letting your dreams guide your life.

Is it wrong to choose a career because you know it would make your parents so proud? Should you vote for the candidate who seems the most like your dad? Should you decide to go to school because you keep dreaming about being in a classroom? Should you choose a house because it looks like the one in “Leave it to Beaver”? Is this really the way people make decisions?

Well, yes, it is. This is exactly how people make decisions. The question is what one can do about it, if anything. And, for AI, the question is whether cases like those found in dreaming and common culture, when absent from an intelligent system, wind up giving that system a very nonhuman intelligence.

Worse than that. Perhaps following an inherently logically based system, full of rational thinking and actual events that have taken place, would simply give us a model that is not intelligent at all. I am not being a romantic here. I am simply trying to understand what human intelligence is about. And it seems to me, that it is a jumble of needs and experiences, which we draw upon when we need to figure out what to do next. People are not even close to logical. I am simply proposing that an intelligent computer ought not be logical either. It needs to have a subconscious and it needs to dream and it needs to rewrite its actual experiences that were distasteful into ones that were happier. In short, it needs to do what people do. Why? Because this blooming buzzing confusion is what intelligence is all about.

CBR FOR DECISION-MAKING

To think about this better, we need to ask what exactly is wrong with using CBR as a decision-making technique? Using prior cases is a perfectly reasonable way to make a decision if the cases being relied upon in the process are:

• reliable

• relevant

• real

• properly labelled

• recognizable

Cases need not be conscious to meet any of the above criteria.

In scientific, medical, legal, and other arenas where there is an attempt to be systematic in reasoning, the question that is taken quite seriously is which cases should be admitted to the case library. There is, after all, in each of these fields, an accepted body of cases that are learned and used when reasoning about a given situation. The difference between this kind of attempt to be precise and systematic in reasoning and everyday reasoning is that there is no governing body attempting to determine what is admissible in your personal case library. There is only you, and as we have seen, you are not very good at it. In fact you do not even know what is in there.

So if decision-making is not really a conscious activity, and if decision-making ability depends upon a case base that is not consciously known, then it follows that to improve one's decision-making abilities, one must:

1. be able to control what gets admitted to one's case base;

2. be able to properly label what is admitted to one's case base;

3. upon retrieval of a plan from one's case base, be able to reject that plan on some sound basis;

4. be able to access one's dreams well enough so that your conscious knows what your unconscious is thinking.

At first glance, this seems like an impossible task. How do we control what gets in or analyze what comes out if we really are unaware of both of these things?

We do not of course. The other day I was on a boat that belongs to a rather eccentric friend of mine who comes from a long line of people in the shipping business. His personal boat is very strange and rather uncomfortable. I like it nevertheless and have made many trips with him on that boat. He always talks about how well engineered it is, how it can withstand any storm, and other things he says that make it clear that people who are just into frivolous things don't appreciate his boat. The oddness of his boat serves as conversation for anyone who knows him, especially in the community of shipping people in which his social life takes place.

Nevertheless, the other day, he mentioned to me how much he likes the wood on his boat and how the lines of his boat especially appeal to him. I asked him what the boat he spent time on when he was little was like. He said, somewhat surprised by the question, that “yes, his image of what a boat should be must have come from his childhood boat because it had similar wood and similar lines to this one.”

This stuff is not that hard really—is it? We make decisions based on our subconscious case base and then rationalize those decisions consciously with all kinds of reasons that had nothing to do with why we actually made the choices we made.

The answer to this involves considering what I call Freud's Fatal Flaw. What was this flaw? Thinking that subconsciously directed thinking was a form of pathology.

THE BATTLE BETWEEN THE NONCONSCIOUSAND THE CONSCIOUS

The subconscious relies upon cases the conscious does not know it has to help it make decisions the conscious does not know it has made. The conscious finds its subconscious partner helping it make wrong decisions. So, it reads books like Get Out of Your Own Way: Overcoming Self-Defeating Behavior, 10 Bad Choices that Ruin Women's Lives, and Learning to Love Yourself.

There are literally thousands of books trying to help people stop making decisions that are bad for them. There really are only two reasons why anyone would make a decision that was bad for them. The first and most obvious is that they didn't know that the result of what they chose would be bad. But, of course, there is a second reason—that they didn't make the choice, their subconscious made that choice.

To see what I mean here, I want to talk about a friend of mine's love life. My friend has been married three times and is now single again. He has had many relationships while single, but none of them seem to work out. The only women who last in his life are those who realize that they can take advantage of him in some way. So, he always has some much younger woman who he travels with, eats dinner with, and helps out in various ways. These women are not girlfriends, but he always wishes they were. They always reject his romantic overtures but keep seeking him out for other things.

Not so long ago, we were talking about his parents and it became obvious to me that he had a deep-seated problem with his (recently deceased) mother. He never said this, of course, but he had been abandoned by her for some time when he was little because of World War II and he complained about how in her old age she mistreated his sister. I wondered aloud if she had mistreated him when he was young and after a while got a statement about how she had never shown him much affection. Could it be possible, I asked, that he sought out women who would also deny him affection? He was not amused. Why would he do that he said. I was crazy. Later, he said maybe I was not crazy and that he would think about it.

Certainly, he never decided to find women who would mistreat him. At least his conscious did not decide that. His conscious could easily express the anger he felt towards what I had said because I had implied that he (that is his conscious) was crazy. It made terrible choices. But I meant no such thing of course. I think his subconscious is making decisions along the lines that Hendrix suggests, trying to find a woman who will deny him affection so he can work on that issue with her. His conscious is directed by his subconscious as it is in most choices involving the opposite sex.

In some sense we accept this crazy behavior in our friends. We watch them make bad choices in a mate, ones that every one but them seems to realize immediately are doomed to failure, and we sometimes decide to intervene. But by and large, it is hopeless. We are reasoning with the conscious and the subconscious is driving the car.

In some sense there is a war between the conscious and the subconscious and the subconscious seems to win most of the time. It makes the important decisions and the conscious rationalizes them. The conscious is left to explain why an obviously bad choice is really good. It often convinces no one but the speaker. One's friends after all, cannot hear your nonconscious. They only hear your conscious which, they feel, is making no sense.

So it would seem that the trick to being a good thinker is to free oneself from the nonconscious decision-maker and let the conscious do the reasoning. There is only one problem with this plan: the conscious is actually rather bad at reasoning. The conscious really has trouble making any decision at all, much less making a really complex or important decision.

Wittgenstein once said that all the important thought he ever had was conceived of in one of the three B's—bed, bath, or bus. To put this another way, whenever he let his mind wander, when he was not trying to think of anything much, good thoughts came to him. In other words, when his conscious was relaxed he could hear what his nonconscious had to say. And his nonconscious was way better at thinking than his conscious was. Great writers and great scientists report this phenomenon all the time. They dream of ideas or they wake up with a good idea. Or, while they are writing, they don't even know what is going to appear on the page—they just write and are amazed to read what they wrote. This is the nonconscious at its best—hardly a decision-making apparatus to be taken lightly.

Listening to the nonconscious causes you to make terrible love choices but be really creative. Is this the choice we have to make?

Freud wrote about this issue exactly, but he was thinking primarily about people whose subconscious was driving them to behave strangely and were therefore dysfunctional in some way. But this dysfunctionality is a normal part of what it means to be human. The conscious and the subconscious are in serious conflict in even the healthiest of people.

So, should AI people study Freud in order to build smart machines? I suppose it wouldn't hurt but that is really not my point. Artificial intelligence people need to study people—really study people. They need to figure out how real decisions get made—not decisions about chess moves or plan selection from a library—but decisions about what to have for dessert or who to marry.

Ah. There is the rub. Computers do not eat dessert nor do they get married. Yes, precisely. All our conceptions of intelligent machines come from models in which machines live rather dull lives. They do not have real drives and they do not have real conflicts. In short, they do not have the kinds of experiences that make people intelligent, which cause people to have to try and figure stuff out for themselves. There is no war between the subconscious and the conscious in our conception of intelligent machines because those computers are not thought of as having been raised by their creator for many years, eventually running into the kinds of conflicts that every child runs into with his parents and the society in which he is about to enter.

Any first-rate computer intelligence would have to have those sorts of experiences however. It would have to dream, resolve complex personal issues, make decisions that make its friends and caretakers unhappy, and then worry about the consequences. That is the stuff that makes us intelligent, at least the kind of intelligent we turn out to be.

We can, of course, choose to create computer intelligences that are simply logical and not representative of human intelligence at all. We could, but it would be boring.