Responsible Thinking: Appendix


Back to home page

Printer friendly

Human Thought

Note: I apologize if the material here is not too intelligible. It represents an attempt to briefly summarize the way I look at intelligence, based mostly on ideas from when I was a graduate student working on Artificial Intelligence. It is far too brief to do a good job of explaining things, but hopefully some people will find some useful ideas here.

Innate Mechanisms vs. Learned Mechanisms

In this essay I want to consider the basic mechanisms of thought which enable learning to take place. These are elements babies would start life with. Adult behavior involves complex thinking strategies, most of which, I would assume, were learned ultimately by making use of the inborn learning mechanisms. In many cases it is difficult to know whether a thinking process that almost everybody uses is something that was inborn or whether it was learned. In this essay I will try to suggest mechanisms that are so basic that it seems unlikely that they could have been learned from even simpler principles. If it turns out that the mechanisms suggested here are inadequate to explain some things that people can actually do, then we will have to try to determine what additional basic functions are necessary.

Pattern Recognition

The fundamental mechanism of thinking is pattern recognition. Patterns of information are received by the senses and compared with previously stored knowledge in the brain, which retrieves related patterns and sends appropriate motor signals to the muscles. The patterns our senses observe are far too complicated to ever match previously observed patterns exactly, so recognition is always a matter of finding what matches best, not what matches perfectly.

Sensory specific processing

There are parts of the brain that process sensory data in ways specific to the nature of each of the senses, for example the visual cortex. Since we are interested in the most general purpose learning mechanisms, the patterns we are concerned with will be assumed to be the results of specialized sensory processing. The recognition we will mainly be concerned with is what takes place in the conscious mind, presumably in the cerebrum.

Active Concepts

In addition to current sensory information, the pattern the conscious mind attempts to recognize at any moment must include the results of other recent recognitions. We'll call these recent recognitions "active concepts" . These provide a context for better recognition of the new patterns. If I feel something on the back of my neck, my interpretation of what that means is different if I just saw a bee fly by than if I just saw a fly go by.

Very recent active concepts have more effect on the recognition than ones from further in the past. This would make sense from a practical viewpoint, since recent observations are more relevant to the present situation than observations that we made further in the past. The mosquito we saw ten seconds ago is more relevant than the bee we saw an hour ago, and the car we saw our friend in yesterday is more likely to be the one she is driving now than the one we saw her in five years ago. To some extent all concepts stored in our brains are active, but will vary greatly in degree from important ones we observed seconds ago to minor things we observed a long time in the past. The degree of activity would seem to vary with the importance we attach to any concept as well as how recently it was recognized.

New Patterns vs. Old Patterns

Any observed situation is potentially a new pattern or concept to be stored in our brains depending on the extent to which it seems important. It consists of a combination of sensory inputs and existing active concepts, and becomes a concept itself which could be a part of future patterns.


The results of a pattern recognition must include not only active concepts, but outputs which affect our behavior. What we do is very much determined by what we recognize from our senses plus the active concepts which help provide a context for that information.

Sensory buffers

There seems to be another source of input to the mental lookup. If I say "visualize a horse", most people will be able to do this. If I then ask, "which way is the horse facing", the person will usually be able to give me a specific direction. There seems to be an input to the thought process from a "visualization" area of the brain which acts somewhat like the input we get from our eyes, except we ourselves determine what goes into this area. Similarly, there is an ability to "visualize" sounds. We can think of the sound of glass breaking, for example. This affects our thinking a little bit like the actual sound of glass breaking. We seem to have the ability to construct sensory impressions which we can then observe. The areas for handling these impressions I'll refer to as sensory "buffers".

This is particularly vivid for the construction of words. Often people feel they think in words. This would seem to be because we create strings of words for our hearing buffer which we can then "listen" to, perhaps to see what implications they have. We might then try out an argument on ourselves before speaking it out loud to someone else, or we might just repeat a telephone number to ourselves between the time we see it written down and when we dial it on the telephone.

So the three types of input which are used for the mental pattern recognition would be actual sensory information, sensory buffer information, and concepts activated by recent recognitions. Similarly there are three "outputs" from the pattern recognition - active concepts, motor actions, and the outputs that enter new visualizations into the sensory buffers.

Stream of consciousness

In the absence of especially interesting sensory stimuli or the need for physical action, our mind uses the current set of active concepts and inputs from the sensory buffers to do recognitions and generate new active concepts and outputs to the sensory buffers. This produces a sort of stream of consciousness, in which thoughts progress from one thing to another related thing while we sit "thinking".

Freudian slips

Suppose a man at a cocktail party is distracted by the low neckline of the woman he is talking to. He intends to say "I know you'll do your best" but mistakenly says "I know you'll do your breast". This would be an example of a "Freudian slip". Rather than being the result of some mysterious unconscious activity, it can be seen as a simple consequence of the way patterns are recognized in the brain. The active concepts which would normally lead to the word "best" are combined with the man's thoughts about "breasts" and it turns out that "breast" is a better match for the next word than "best" is, so that's what he says.

Learning behavior based on expected reward

It would be possible for an animal to have mechanisms such as those described above and have all the patterns built in and unchanging from the time they are born until the time they die. Obviously this is not true for people, who learn by observing and experiencing the world around them. This makes people (and to lesser degrees, many other animals) much more flexible in dealing with unfamiliar situations. They do not have to be born knowing how to respond to every situation they might encounter, but can adapt their behavior to unexpected circumstances.

Let's imagine a situation where a barefoot, hungry person has several choices. She can stay in a comfortable patch of dirt, walk through some brambles to some food, walk over sharp pebbles to food, walk over hot rocks to food, or walk over cool grass to food. If she has never experienced any of these choices before, she might make any one of them, but once she has experienced them all, she would be likely to walk on the cool grass. The reason is that this choice would give her the most pleasure with the least amount of pain.

With a very simple animal like an insect, behavior is usually a particular response to particular situations (as detected by the senses), but with a complex organism like a human, nature seems to have taken a more subtle strategy. Instead of being wired to do certain things in certain situations, we are wired to identify a certain degree of pleasure or pain with every situation. Then our learning ability is organized to produce behavior that leads to the greatest possible pleasure and the least possible pain. Clearly the things that produce pleasure are the things that are beneficial to survival and continuation of the species, and the things that lead to pain are things that are likely to hinder survival.

It seems to me that virtually all human behavior is oriented around increasing pleasure and reducing pain. Behavior change based on reward and punishment is known as operant conditioning by behavioral psychologists.

Time duration of reward

At any particular moment in time a person may have several conditions existing which affect his level of reward. His toe may hurt and the sun may be in his eyes, both mild sources of pain, while at the same time he is getting a compliment from someone he wants to impress, which is a source of reward. It seems logical to add these together, with the rewards being positive and the pains being negative, to arrive at a single "reward level" that exists at any moment in time.

However, since a person lives over a period of time, it makes sense that the length of time a reward or pain is in effect should be proportional to its importance or its influence on behavior. So an enjoyable back massage that lasts for ten minutes should be ten times a strong a reward as the same massage that lasts for only one minute, and similarly a toothache that lasts for one hour should only be one tenth as bad as one that lasts for ten hours.

Learning with rewards

I can think of two ways that reward level could affect learning. First, whenever we must make a decision we could visualize the likely outcome of each alternative and select the one that is best. This is fairly complicated since we would not normally be totally sure of the outcomes and so must consider the various possibilities and their likelihoods and how good or bad each would be for any one given action. If I need to decide whether to go to a party tonight or stay home, I must make a judgement about how much fun the party will be and how much I will enjoy staying home. While I may have a pretty good idea of how it will be to stay home, the party might be a lot of fun or very dull depending on things I don't know about, like who else goes.

The second possibility is that whenever we have experienced a certain outcome the level or reward associated with it either supresses or encourages associations between the decision situation and the action, so that actions that had been successful in the past would be more likely to become active concepts than actions that had unpleasant results in the past. If there is a plate of rice cakes and a plate of chocolate chip cookies in front of me, I might simply look at the situation and immediately "reach for the cookies" becomes an active concept rather than "reach for the rice cakes" because I have tried both approaches in the past and "reach for the cookies" has been rewarded more than "reach for the rice cakes."

Clearly we do sometimes reason about decisions as in the first mechanism, but it's possible that we have learned to do such reasoning through small steps which were themselves learned by the second method.

Secondary reinforcement

When we are making decisions about actions that might lead to different alternative situations, we need to evaluate how desirable each situation is in order to do the comparison. It makes sense that together with our memory of the pattern that defines a situation, we also remember the reward or pain that was associated with it. However it may be that some situations don't involve any direct reward or pain, but make it more likely that we will get reward or pain in the future. Breaking a window with a rock does not cause pain at the time the window is broken, but it greatly increases the likeliness of unhappiness at a later time. If some circumstance has typically led to pleasure in the past, we would then associate pleasure with that circumstance. We might, for example, normally enjoy going to a show, since when we go, we typically find it pleasurable. If we are given tickets to a show, we evaluate that situation as desirable since it is likely to bring about the rewarding situation of going to the show. Having reward associated with circumstances that precede reward allows actions to be rewarded even though the actual reward might not be experienced until a later time. If we did someone a favor that resulted in her giving us tickets to a show, we could learn that doing such a favor is beneficial, even before we got the actual reward of seeing the show. In effect our behavior is reinforced by the expectation of reward rather than by the reward itself. The same applies, of course for punishment. Potentially this sort of learning can involve many steps - we might recognize a situation is good because it leads to a situation that leads to a situation that leads to a situation that is rewarding.

The property described is effectively what is called secondary reinforcement by behavioral psychologists.

It would seem that our memory maintains a catalog of experiences (patterns that we recognize) and how rewarding each is. Notice that for this to work properly, getting an already expected reward is not very reinforcing. In fact even if a reward is positive, but it's less that what we would expect from previous situations, we ought to suspect that our action was in error and the action is, in effect, punished. We measure the benefit of our actions by whether our expectation of reward increases or decreases.

Generalization and differentiation

Our behavior is based on patterns that we recognize. This is much more difficult than it may sound. One problem is that there are many things sensed at any point in time, and identifying what parts of our sensory experience are the ones relevant to our behavior can be difficult. A second problem is that the object or situation that we want to identify may not always appear the same at different times. The same word may be pronounced differently by different people, and people may even use different words for the same thing, and we need to recognize when these different sensory patterns represent essentially the same situation.

The first problem is solved to some degree by the learning principle known as habituation, in which sensory stimuli which are always present eventually are ignored. Our situation is a little more complex, in that we want to ignore stimuli when they have no effect on recognizing one pattern, but we may want to take them into account when recognizing something else.

The second problem is related to the learning principle known as classical conditioning. If two sensory patterns usually occur together, such as the bark of a dog and the appearance of a dog (or in Pavlov's famous experiment, the sound of a bell and the appearance of food), we are justified in assuming they are part of the same phenomenon and we ought to react the same way to both.

Superstition and probability

We are constantly searching for patterns that we can use. If we eat a certain kind of nut and later feel ill, we may hypothesize that the nut was bad for us. If we push a button and a light goes on, we might assume that the button controls the light. If we touch an oddly shaped stone and shortly thereafter step in a hole and twist an ankle, we might hypothesize the stone caused the bad luck. Educated people in a modern society might reason that there is no mechanism that would relate touching a stone to bad luck. However, in order to give us the greatest chance to survive in unanticipated situations, our brains do not rule out such possibilities. If they did, they might rule out the idea that pushing a button could turn on a light, since until very recently in human history, that would have been impossible. Even if the next time we find such a rock (or a black cat crosses our path), we don't suffer the same misfortune, it still is a reasonable mental strategy to remember it in case some other factor is involved in the rock's unlucky effect. Or perhaps the rock will cause some other misfortune at a later time.

In any case, we can see how superstition can arise as a result of a normal psychological learning mechanism. We are constantly looking for cause-effect relationships and frequently chance events cause us to assume relationships that aren't valid. Often we might forget these over time, but if the original event was important in our lives, or the event reoccurs, or other people report a similar belief, we may become fairly convinced of the truth of the superstition.

As people with an advanced culture, we recognize certain events as having a specific probability, such as the chance of throwing a "6" on a die as one in six. It seems unlikely that we have the estimation of probabilities as a basic mental function, however, since it would seem to be a better strategy to try to identify other causal factors for an outcome of interest. If we go to the store and find that it is closed approximately one day in seven, we should not just chalk it up to a 6/7 chance of being open. We might find that whether it is open is related to the day of the week.


Emotional states seem to occur when things of particular importance occur in our lives. Fear occurs when our happiness or survival is threatened. Anger occurs when it might be important to take an action against someone else. Love and lust are related to the need to reproduce. It seems likely that an emotional situation triggers the release of chemicals in the bloodstream that alter the functioning of the brains neurons and their connections, and perhaps other bodily functions as well. In effect, the situation requires some special behavior different from what we would do in the normal course of events. I would guess that one of the biggest effects of emotion would be to change our inherent determiners of pleasure and pain. When we are angry, for example we would find pleasure in hurting the person we are angry at, and might be less concerned about our own pain if we were to get into a fight, whereas normally we would not want to hurt the person and would avoid any fights.