Summary of posts 1 to 8: A theory of Artificial General Intelligence (AGI) must begin by focusing on invisible, automatic mental behaviours that drive higher-level, complex ones. One such behaviour is “becoming aware of something”, either something external or within one’s own mind. This results in the formation of concrete memories/thoughts (sights and sounds). Re-experiencing such stimuli as thoughts can help you solve problems. Thus awareness can be interpreted as a type of problem-solving.
Learning, in order to be useful, must serve a purpose. So what purpose does learning serve in the human mind? Many possible answers have been suggested; morality, hedonism, truth, survival, power, and others have all at some time been proposed as the purpose of life. Ultimately, each of these affects your motivations by influencing what your mind treats as valuable¹. They can all be viewed as overarching goals that are realized as positive and negative “values”. This post will begin to dig into the concrete details of how this influence works.
Our goal (yet again) is to find automatic, unconscious mental behaviours that drive conscious machinations like goal-setting and personal ambition. Higher-level mental behaviours, like long-term planning, or breaking problems down into sub-problems, are complex and consciously learned or developed. The mechanism we’re looking for, on the other hand, must be absolutely universal; it must act directly in every instance, without the possibility of conscious interference. This disqualifies any motives, like survival, or hedonism that can be overcome, negated, or ignored based on other conscious intentions.
It’s true that survival, hedonism, etc. could be reinterpreted in a way to make them seem universal. For example, you could say that ascetic monks who reject hedonism are in some way still serving it, or that self-sacrificing martyrs are still acting in aid of survival in a broad sense. However, using the words “hedonism” or “survival” are inadequate since, on the surface at least, they can be used to justify any type of behaviour, even mutually contradictory ones. We should look for a more accurate, and more granular way to describe what is happening.
As discussed in an earlier post, motivations can only really be known by the way they influence actions or thoughts. The best way to begin, then, is to figure out what actions or thoughts they influence, and find a pattern. Here, things get difficult. Although you may be able to spot a pattern in how your thoughts and actions change over a long time, such as when you learn how to speak a language, each individual instance of learning something — e.g. learning to say the word “paper” — usually leaves no trace of how or why it was learned. Think of the thousands of words you know. For how many of them can you remember when you first learned it? It takes conscious effort to notice you have learned something new; and that meta-information itself would be a new instance of learning, thus demanding its own explanation.
Thus far in the series we have focused on one type of learning, that of “becoming aware of something”. This type of learning includes memories, and even the act of combining thoughts together. Although we have not (yet) asserted that this is the only type of learning, it is a useful one with which to start studying motivations, since it takes place in a single moment, and has specific content. Such learning is also motivated — though that fact can often be obscured since, as mentioned, we usually don’t remember why we learned something, and often need to invent explanations after the fact.
In the previous post we showed how becoming aware can be seen as a type of problem-solving: a problem is set up, it looks for a solution that fits its criteria, and finally the mind “becomes aware” of the content of the solution; for example, when you are asked to look for a red-coloured item that is in front of you. However, the words “problem” and “solution” imply a lot of complexity and ambiguity, so we should reduce them only to the set of specific changes they cause. In the simplest case, this pattern begins with a trigger of some kind; perhaps you perceive that a question is being asked, or sense an uncomfortable noise or sight, or something needs explaining, etc. You wait for a solution to present itself, in either thought or external experience which, once perceived, is remembered. Implicit in this is some way of knowing that the problem was solved — how did you know that what you saw qualified as a “red object”?
There is one other missing piece which is not as obvious. When you learn something new, whether it is an action or a thought, you’d want what you learned to help you next time, without having to solve the same problem again. If you’re looking for your keys, and find them under a cushion, the next time you want to know where your keys are, you’d want the image of them under the cushion to pop into your mind. You may then use that memory to reach under the cushion. Whether the keys are actually there or not would be irrelevant at the time when you first learned it, since you can’t know beforehand if this is a constant pattern (you always leave your keys there) or if it is a one-off which can soon be forgotten. You may notice from this example that there is a grey area between memory and knowledge — though it was a single instance (a memory), it could implicitly serve as the first example of more general knowledge. With enough repetition it would become knowledge.
This leads us to a curious property of imagined solutions: whether or not they are generalizable, the fact is that you wish they were. The next time you couldn’t find your keys, it would very much help you if you could look under your cushion and find your keys there, since you could easily solve the problem at hand. If it weren’t there, you’d have to search extensively all over again. This means that, since it solved your problem last time, this pattern of experiences is something you’d hope will be true again and again. A solution pattern, once found, ideally would repeat since you’d want to escape the stress of not being able to find your keys. It is more than just an expectation or prediction. It is a positive, i.e. desired expectation. The sights and sounds that have solved your problems before are things that you would want to see again; they are “solutions” in the broad sense of the term.
There is no existing English word for this type of mental event. What name would you give to an imagined sound or image, that reflects something you experienced earlier, that solved a problem, that you predict or expect will happen again, and at the same time want to happen again? “Solution” seems close, but it implies that it objectively solves the general problem, when all it does is recall a set of sights and sounds you’d like to re-experience. The closest word I can think of is “intention”, or perhaps “ideal”.
To summarize the discussion so far, there are certain concrete thoughts which are created in a moment of awareness, during which they solved a problem for you. These thoughts reappear in your head later, in similar contexts. If you’re having difficulty wrapping your mind around this mechanism, consider the exact same pattern in terms of physical actions. Think of learning to move your arm away from a hot stove. You see a stove near your arm (circumstances), you accidentally touch the stove and experience pain (problem), your built-in reflexes helpfully retract your arm (action) and the immediate pain subsides (solution). The next time you see a stove near your arm (circumstances), you immediately carry out the solving action and retract your arm (action); thus skipping over the pain (problem). Thoughts, as described above, follow an identical pattern— the only difference is instead of an action, it is a sight or sound that is recreated in your mind; aka a thought.² Any experience that allows you to skip over the problem the next time is remembered as a thought.
Using the word “problem” for the initial driving force may be confusing, since that word can be used to indicate any kind of complex perceived issue, such as a political disagreement. What we are looking for is a simple, basic unit of motivation. The words “pain”, or “stress” are also inaccurate, since some problems are psychological, and you don’t necessary experience a stressful feeling or somatic response alongside a problem. Even the words “negative” or “bad” are somewhat misleading since they imply a conscious value judgment, as though the mind knows that these are evils. Finally, the words “desire” or “need” include both the problem and solution sides of this process, so they also don’t fit. In the interest of clarity, the best word I have found for the above force is “tension”. A tension is an impulse or imbalance that automatically tries to remove itself through some learned change. This happens when the mind learns an action or thought that overcomes it, and uses that response next time as a way to prevent it occurring again. A tension is a signal for something you’d do well to avoid. It is self-effacing.
Indeed, there need not be any negative conscious experience involved in a tension at all, nor any knowledge that the event is “bad”. The only thing that defines a tension at all, and therefore designates its content as “bad”, is that the mind automatically tries to overwrite it. This model of motivation reverses customary notions of how problems work on the mind. It is commonly believed that we avoid bad things because they “feel” unpleasant. However, such a theory opens a can of worms when you ask what it means for something to “feel” unpleasant to your consciousness, and how this causes avoidance behaviours. It requires the mind to have created concepts of “bad” or “desire”, and then to work with those, before you can experience any sort of motivation. It makes the whole process dependant on conscious awareness and puts it under an individual’s control to respond, thus raising questions of choice and free will. It no longer becomes an automatic, built-in process, but a complex, high-level, and conscious one.
The model outlined in this post addresses this issue by reversing the assumed cause and effect. In short, we don’t avoid things because they consciously feel bad, it is the fact that we automatically avoid them that makes something consciously considered bad. The conscious recognition and apprehension of something as “unpleasant” is a type of awareness — you become aware that something is unpleasant before you react. And awareness is a type of learning which, as shown above, is motivated. Therefore you learn to think of something as unpleasant because such thoughts are effective for avoiding that situation again. This is the only way to coherently address the riddle of why things “feel” bad, without also creating a circular argument.
There is no other way to resolve the paradox of why things “feel bad”; or why they “feel good” for that matter. This approach also opens a door to explaining conscious experiences in general, which was shut to us before. In fact, this door was closed because our own foot was up against it. By treating conscious experience (e.g. of good or bad feelings) as a cause of actions or of learning, we set up a paradox that could not be resolved, since we had to explain mental behaviour in terms of consciousness, which itself is a kind of mental behaviour.
This observation has consequences for theories of awareness and therefore of consciousness. It implies that moments of conscious awareness don’t drive motivations; conscious awareness is itself created by underlying drives and motivations, as an explicit manifestation of their influence on thinking. Your tensions force you to be aware. Where tensions cease to exist, conscious awareness disappears as well.
The next post will continue this investigation and dig into the details of how tensions work on the mind. As mentioned before, motivations are nuanced and idiosyncratic; they help define a person’s character. This raises the question of how and why they are formed and shaped in an individual in the first place. We’ll look into how, through experience, certain situations themselves come to trigger tensions.
Next post: Creating motivations
¹ This includes “truth”. Objective truth is never the self-sustaining driver of learning or belief; if it were, no discussion would ever get “heated”. “Heated” implies conflicting motivations. On a personal level too, accurately understanding the world is not the default or norm; it requires effort, and skills such as critical thinking. Effort and skill imply a motivating impulse. If you don’t care about being accurate or truthful, your beliefs will skew towards wishful thinking.
² In the case of learning thoughts, you may not have had to do any physical action for the problem to be solved. This is an early clue as to what differentiates thought solutions from action solutions (see upcoming posts).