On Philosophy

October 20, 2007

A Theory About Thinking

Filed under: Mind — Peter @ 12:00 am

Today I am going to present an initial hypothesis about how thinking works, motivated partly by my claim yesterday that certain features of language exist because they influence they way we think, and not because they add to the descriptive content of the sentence. Obviously if that is the case then a complete account of language will require a complementary account of how we think in order to connect linguistic features with the thought processes that they influence. Of course to present a hypothesis about thinking I must first indicate in some way what thinking is, and that is not exactly an easy task. In a rough way we can understand a thought as simply the part of consciousness that is not a product of perception but rather of the inner workings of the mind (produced by the minds operation on its own state). Some thoughts are very precise and may be accompanied by inner speaking of some kind, but this is not the only kind of thought we can have. There are also “vaguer” thoughts, unaccompanied by any kind of inner speaking, as well as mental images, and both can be considered thoughts. But pain, for example, is not a thought because it is a form of perception. This distinction may seem arbitrary in some ways (why not simply add perceptions in as well?) but it exists in the service of this theory, which is about thinking, how one thought is connected to another. Perceptions are, from such a point of view, a special case since they are not derived from other thoughts. And, similarly, how thoughts produce actions is also outside the scope of this theory.

The fundamental connection between one thought and another is association. All thoughts have content of some kind. For example, the thought that it is raining outside contains content about the world, specifically about rain and the fact that it is raining. Now by content I do not mean to indicate what the thought refers to, that is a different matter (and something that does not depend on the mind alone). It is hard to say exactly what content is, but I can shed some more light on the matter by pointing out that a mental image of raining and the thought that “it is raining” accompanied by some kind of inner speech have substantial overlaps in content. Thus two thoughts may be alike in sharing some of the same content, or their content may be similar (similarity between content may be a primitive relationship, or it may be that content has inner complexities and that similar content have some identical parts). Association then is a process by which thoughts give rise to further thoughts with content that overlaps or is similar in some way. I call this the fundamental process by which thinking works because it is how we think when we think without self-reflection, one thought follows another via the process of association.

The mechanism by which one thought comes to follow another is necessarily unconscious at some level, and thus I am unable to say exactly how association works. In fact the claim that association proceeds by a similarity in content is simply a generalization from my own thinking. I will make a few more observations about how association works, also such generalizations, but I will leave them for later. For now I would simply like to observe that the content which association proceeds upon can range from a small part of the content to all of the content. When association is proceeding on the basis of a small part of the content of the thought the subsequent thoughts appear very loosely connected; when association proceeds on the basis of most or all of the content it appears that we are turning over the same thought in our heads and looking at it from different “angles” (different ways of structuring that content so that different aspects are brought to attention), such reflections are also how vague unstructured thoughts can be turned into structured ones accompanied by some kind of inner speech. Finally, I would like to observe that as we can have multiple thoughts at the same time association may also proceed on the basis of several of these thoughts. For example, if we had two thoughts in mind association may produce our next thought by yielding a new thought with some new content but which has some content similar to one of those previous thoughts and some content similar to the other.

Thinking by association is a very “primitive” way of thinking. As I have mentioned previously it usually occurs when we aren’t consciously reflecting on our thought process itself. And I suspect that associative thinking, or something very much like it, is present in many animals (although the content of their thoughts is much more restricted than ours is). In addition to associative thinking there is also what I call formalistic thinking, which can be described as thinking according to some plan (the plan itself being contained in some other thought or thoughts). Any time when we are producing one thought from another on the basis of a conscious process we are probably engaged in such thinking. For example, logical deduction is an example of formalistic thinking. Causal or predictive thinking as we engage in it is also an example of formalistic thinking, we consider the situation at hand, the laws governing it, and then apply those laws to produce the situation we think will result from it. (Of course predictive thinking can also be done via association, by association with similar situation and from there the association with the situations that followed that one, but such thinking is much less accurate.)

Finally there is unconscious thinking, which, together with associative and formalistic thinking, I believe exhausts the ways in which thinking proceeds. Unconscious thinking may be a bit of a misnomer, because obvious the thoughts that result from “unconscious” thinking are themselves conscious. Unconscious thinking is really a hypothesis to explain the rare thoughts that seem to pop into mind without any connection to those that have immediately gone before them. Specifically I suppose that there exist things that might be called unconscious thoughts, similar to thoughts in their structure and they way they are processed by the mind, but unlike them in that they are not available to conscious reflection. And I further suppose that when some thoughts move out of conscious they become such unconscious thoughts rather than disappearing from the mind completely, and that associative thinking applies to these unconscious thoughts as well, producing further unconscious thoughts. And finally, for some unknown reason, sometimes these unconscious thoughts return to consciousness. Thus the apparently unconnected thoughts that pop into mind are not really unconnected at all, by my hypothesis, rather they are connected by a chain of associative thinking that is simply hidden from us.

This concludes my survey of how thinking occurs. Which brings me to the topic that is in interest in connection with language: what factors influence how thinking proceeds. For formalistic thinking the answer is obvious: the plans for thinking that we have available to us. And obviously language can influence such thinking, in a roundabout way, by describing plans for thinking or by suggesting that a certain plan be employed. More interesting is associative thinking, about which I am able to say much less. Obviously there must be something we might call “focus” that determines which content is the basis of further associations, and where the focus is seems to depend on what we are interested in, but beyond that it is hard to say much more. It may also be possible to say something about where new thoughts spring from. For example, we may observe that new content in thoughts produced by associative thinking often comes from memories, but again it is hard to pin down all the sources of this content. And finally we may observe that our emotional state leads associative thinking to produce new thoughts that are “compatible” with that state (for example, being angry will result in associative thinking tending to produce “negative” thoughts). Here I will put a close to this investigation for now, partly because I have not reflected on the possibilities sufficiently, and partly because here we are beginning to get into the domain of psychology, where our generalizations are specific enough that they must be tested experimentally and not by simple reflection.

October 13, 2007

Investigating Consciousness

Filed under: Mind — Peter @ 12:00 am

When investigating anything it is extremely important to tie both the development of the theory itself and the content of the theory to experience. If we are investigating something then there is a distinction between our theories and the subject matter itself and so we need a way to evaluate our theories to see how well they conform to the subject matter. And the only way to do that is through experience of some kind, because as far as I can tell we lack any other way to get at things, except through our experiences. (This can be contrasted to a completely constructive enterprise, such as mathematics, where the standards are completely internal, meaning that the nature of the theory alone determines if it is good or bad.) Before turning to consciousness let us first look a little more closely at how we need to tie both the evaluative and constructive aspects of our theorizing to experience.

First let’s examine why we need to connect our theories in some way to experience in the context of evaluation, because it is easiest to see the need for such a connection in this context. In order to evaluate a theory I claim that each theoretical entity needs to be able to be experienced in some way. Now obviously many of the things we theorize about, such as atoms, are things we can’t directly look at. But that is not a problem, all our theory needs to do is to provide some indirect way to experience atoms (through their effects), which various instruments and experimental set-ups do in fact allow. With this connection we are then able to verify, at least in some cases, whether atoms are behaving as the theory claims they do, whether they are present or absent as claimed, whether they are moving quickly or slowly as claimed, and so on. To see why this connection is necessary simply consider a theory that lacks one, such as the hypothesis that all atoms are made of a mixture of substance X and substance Y in various proportions, and that the amount of X and Y present give each atom its properties. Unless X and Y are independently made to be specifically responsible for various properties we have no way to get at X and Y in any way, and thus the theory is essentially vacuous (it only seems to make assertions while it really asserts nothing).

There is also a need to associate our constructive theoretical activities with experience in some way (although a less rigorous one). Obviously when we begin an investigation we do so with some “theoretical” expectations. For example, we may be looking for a specific kind of thing, which means that we are looking for a class of things united by sharing certain features. But our starting point can’t be completely theoretical, we need something material to latch onto and to serve as our connection with experience for our initial hypothesizing. For example, we might be looking for a theory about the kind of things that appear a certain way or the kind of things that cause a certain event (causation being an additional “theoretical” expectation). Without any such connection the investigation can’t get started. Consider, for example, investigating some X, defined as the cause of Y and Z, Y and Z being the mirror images of each other. Obviously X, Y, and Z are all defined completely theoretically (without any reference to experience), which is why I am not naming them, to prevent accidentally associating them with experience in some way. There is simply no way to investigate this X, we can only elaborate on the theoretical commitments we began with, which obviously doesn’t add anything new to our understanding of X, nor does it bring us any closer to a theory that can be evaluated (as mentioned above).

With that in mind we are now ready to consider how we should investigate consciousness. I would argue that we should divide consciousness into two concepts, because the word is used in two distinct ways, with the hope of reuniting them later. Specifically I would separate “conscious” used as a description of other people from “our consciousness” used in an introspective sense to describe our own experience. The first sense of “conscious” is relatively easy to investigate, so long as we don’t trip ourselves up and cloud the issue with ideas stemming from “our consciousness”. “Conscious” is something that we ascribe to people primarily on the basis of their behavior, especially verbal behavior. Even more specifically we seem to call people conscious because they display introspective behavior, behavior that reflects on their own inner states, feelings, attitudes, goals, and so on. Obviously such behavior can be experienced, and so it can serve as a starting point for an investigation. We thus begin our investigation of consciousness, in the first sense, as that which is the cause of this behavior and which produces it as an accurate reflection of its operation (a theoretical requirement added because our expectation is that this behavior reflects what is really going on inside the person, and not an illusion, such as might be produced by a recording of such behavior). Such an investigation probably concludes with some kind of functional description of the kinds of systems that count as conscious.

Obviously investigating “our consciousness” starts in a quite different way. There are no specific experiences that are associated with “our consciousness”, rather every experience we have is part of it. Indeed even distinguishing our consciousness from the rest of the world requires a bit of theoretical sophistication, specifically accepting the hypothesis that there is an objective world out there and that we only experience some aspects of some parts of it (if someone was a solipsist and didn’t buy into this then “the world” and “consciousness” would designate the same thing, which would end the investigation rather early). What we are looking for then is something that unites or causes all of our experiences but is absent from everything that isn’t experienced. Obviously there are many directions to go at this point (because what this common feature is isn’t at all obvious), some worth considering, some absurd (absurd because they leave behind any hope of confirming the theory developed through experience). We might hypothesize at this point that we are one of the systems that is called conscious in the sense investigated above. If this is indeed that case then what unites all of our experiences would be the fact that they are all available to a certain kind of self-reflective activity (whatever exactly the theory developed by the previous investigation concludes). And that possibility can be confirmed (or disconfirmed) through experience by observing ourselves using our objective tools of inquiry and seeing whether we are such a system and whether our experiences coincide exactly with those that the theory predicts will be.

October 11, 2007

Moments Of Consciousness

Filed under: Mind — Peter @ 12:00 am

On previous occasions I have argued that consciousness is a kind of process, a time extended phenomenon. By time extended phenomenon I mean something that is defined in part by the changes it undergoes. For example, the property of “getting larger” is a time extended one, we can only say that something is getting larger if at previous moments it was smaller and at future moments it will be larger. And since “getting larger” is a time extended phenomenon a single moment is not enough to determine if something has this property, because in a single moment it is not changing in size. Of course it remains to show that consciousness is such a property, but I don’t think that is too hard of a task. Consider, for example, a person who is frozen in time for some duration, who doesn’t change physically in the least. Such a person, when unfrozen, would obviously consider themselves to be unconscious during that period, since from their perspective it would appear as if time had suddenly jumped forwards. Clearly then consciousness doesn’t depend on just the physical properties at a moment of time, otherwise they would have been conscious for that entire period, but on how those properties are changing. But of course that is simply an appeal to a kind intuition, there is also a theoretical reason to think that consciousness is a time extended phenomena. The theoretical reason comes from the definition of what constitutes a conscious experience. As far as we can determine an experience is conscious if and only if it is in some way able to be reflected upon or built upon by future mental states (future experiences). An experience that fails to fulfill this condition will not be considered conscious by the person who is supposedly experiencing it, as they will be unaware of it. Since experience is a necessary part of consciousness, and experiences are a time extended phenomenon, so consciousness must be as well.

But, despite these arguments, consciousness is not experienced as a time extended phenomenon. From the inside, so to speak, it appears that we have consciousness that exists from moment to moment, and which in no way depends on previous or past moments to exist. It appears to us that we are conscious now, and, since we can make that judgment based only on a single moment’s worth of experience, consciousness must not be a time extended phenomena, otherwise we wouldn’t be able to know if we were conscious by considering just that moment. Of course none of this is really an airtight argument for the claim that consciousness is not a time extended phenomenon, but it certainly seems to us that it isn’t. And certainly we should be granted some authority with respect to our own consciousness, at the very least it would seem that we are owed an explanation as to how we can be so drastically in error about it.

As I see it the solution is not to give in to our intuitions about consciousness, but rather to explain them and explain why they can be in error. To do that we must first reflect about where we can legitimately invoke our first person authority with respect to consciousness. I would agree that we are the final judges when it comes to the content of consciousness, what we are conscious of. If we think that we are conscious of something then indeed it is part of our consciousness, and if we would deny that we are aware of it then it is unconscious. But there is more to consciousness than its content. Obviously there must be unconscious forces at play that determine what we are and aren’t conscious of. And when we attempt to capture consciousness in our theories we try to describe more than what we are conscious of, but these unconscious forces that structure and give rise to consciousness itself as well. Because simply describing the phenomenon as experienced is a poor explanation, we already know what we experience, what we want to know is why we experience what we do, as we do, and why we experience anything at all instead of nothing. Obviously answering these questions requires more than simply a description of experience.

Thus my explanation of the fact that consciousness seems like a phenomenon that exists in singular moments is to claim that it simply seems that way to us because each experience is centered on a specific moment. Obviously experiences contain some remnants of previous experiences, because they are connected with each other and not totally separate events. But most of the content of any particular experience is focused on the new thoughts and new sensations that occurred at that moment. Thus the content of consciousness, what we are conscious of, is primarily particular moments. That means that in our experience of consciousness what we encounter is moments, and so, in experiencing the moment, we conclude that the moment is conscious, and thus that we are conscious in that moment, regardless of the future or past. But that is simply an artifact of the way we experience things, not a reflection of the way that consciousness works. We can imagine beings which do not have experiences centered around particular moments, instead their experiences might contain all the information that was experienced during that particular period of consciousness, and that this experience simply grows as time progresses, without losing past moments. To such beings consciousness would seem stretched over an ever-growing length of time. But this is no more of a reflection on how consciousness really works than how we experience things is. And so I conclude that, despite the fact that consciousness may appear to us as if it exists in single moments independently of other moments, there is no problem with holding that consciousness is really a time extended phenomenon.

September 20, 2007

Intelligence = Learning

Filed under: Mind — Peter @ 12:00 am

We can divide the capabilities of an agent into two categories: its skills and its ability to learn. In terms of the agent’s short-term performance it is skills that are the most important, and thus which make the greatest impact on us. However, I would maintain that intelligence is a product of the agent’s capacity to learn, although not every agent that displays some ability to learn would be classified as intelligent. If this is correct then a good deal of work on artificial intelligence may be going down the wrong path, creating agents that possess skills but lack intelligence.

Of course I can’t appeal to any special direct access to the essence of intelligence to say what it really is, but I can point out that the ability to learn will always, in the long run, be superior to skills. No matter what skill an agent possesses an agent that can learn can always acquire that skill, perhaps not directly, but learning has no limits, and so the agent that can learn will, eventually, be able to produce machines to do whatever the agent with skills can. That is the power of the ability to learn, not necessarily to be able to internalize everything, but to abstract and to solve any problem piece by piece, part of which may involve developing tools to solve the problem. Indeed this is how human intelligence works in many ways, we develop intellectual tools, such as math and logic, which allow us to tackle problems that we could never deal with head on. Math and logic are mindless abstractions; in many ways they are like machines we have constructed in our heads to allow us to do with formal rules what we cannot with our native abilities. The fact that we use math to solve certain hard problems is not cheating, and thus doing the same thing in an external fashion, by constructing a machine isn’t either.

But I digress. I consider it sufficient to point out that while skills can solve a fixed number of problems the ability to learn allows any problem to be solved. And thus learning is more appropriately associated with intelligence than skills that can be mastered by “dumb” machines. Of course there are different kinds of learning too, and many of the simplest hardly seem like any kind of intelligence at all. Simple learning we might call single skill learning. Single skill learning is rare in nature, but it is almost the only kind produced by artificial intelligence researchers. Single skill learning is when an agent is able to improve at a specific skill, but does not have the ability to transfer over any of that learning to other areas. For example, a computer that “learns” to play a game by eliminating every losing move it comes across, and never playing that move again, is improving its skill at playing that game, but is getting no better at playing any other game, or learning hot to play games in general. Another example of such learning is developing the ability to sort items into categories by comparing them to examples which experience has shown to be or not to be members. Obviously such a capability results in improved categorization over time, but again this improvement doesn’t extend to other skills. Such a limited ability to learn is not necessarily better than a set of skills, granted it is likely that such single skill learning will, in the long run, result in being the best possible at that skill, but such a narrow ability is not necessarily going to put the agent in a better position to deal with the world than a wide range of skills.

With single skill learning set to one side we can turn our attention to “real” learning. Unfortunately I can’t say with complete precision what “real” learning entails. If I could it would be a short step from there to building human-like artificial intelligence and, as a direct result, being showered with money. I can, however, make some broad observations about some of the capabilities it must include, three of which are: flexible abstractions, meta-learning, and strategies. If an agent has flexible abstractions this means two things. First it is able to develop new mental abstractions, meaning that a category, such as “large red triangles” which first arises as the result of deliberation gradually becomes more and more automatic, until it is eventually a judgment that is produced automatically. Being able to generate such abstractions means that the agent is, over time, able to apply their intelligence at a higher level, at a distance from all the details (even though they maintain the ability to look at the details if they wish). The second thing flexible abstractions entails is the ability to transfer reasoning about one domain over to another. What exactly the claims are about is flexible, allowing reasoning by analogy, the simplest kind of reasoning. If an agent has the capability for meta-learning it means that it is able to observe not only the external world, but its own thinking processes as well, and is thus able to revise those processes on the basis of their success or failure (or to alter them as the situation demands). Meta-learning then entails a capacity to learn to learn, another way in which improvements in one skill can translate to improvements in completely unrelated skills. Finally, we come to the ability to form strategies. Again, this implies two things. First is the ability to mentally evaluate a course of action (which could be a line of reasoning), estimating the likelihood of its success. The second is the ability to follow a strategy (either in action or in reasoning) but, at the same time, have the ability to deviate from it as well if circumstances demand. In other terminology, such an agent might be said to have the ability to create and revise its own abstract goals.

I doubt that list is comprehensive, but it is a start in the right direction for understanding “real” learning, meaning the completely general kind. Of course not even humans work exclusively on the basis of learning. It appears that we come with a number of built in skills, such as 3D mapping and language use, or at least a predisposition for them, making us a combination of skills and the capacity for “real” learning. In purely practical terms it makes a great deal of sense, because learning takes time, and people have only limited life spans. And so evolution will generally lead to the essential skills becoming built-in. But when it comes to computers there is no such restriction, once we have successfully taught a single machine we can just make copies of it. So in terms of reward for our efforts developing machines with the capacity for “real” learning seems infinitely more worthwhile then working on giving them specific skills, or even single skill learning.

September 16, 2007

Consciousness Is Relatively Unimportant

Filed under: Mind — Peter @ 12:00 am

Obviously we study consciousness, like we study everything, because we are curious about it. Consciousness, however, seems to draw a lot of attention; certainly more than is given to many subjects. I think it does so because we naively divide the world into inner and outer parts, the inner world being us and our experiences, and the outer world being everything else. Given that we have a fairly good grip on the outer world that just leaves the inner world to contend with. And I think that people assume that by understanding consciousness we can understand that inner world.

That is a bad assumption. Everyone admits that there is more to the inner world then just consciousness. Which leads to a division of the inner world into conscious and unconscious parts. But there is a tendency to push the unconscious to one side, to treat it as a footnote to consciousness. Because we are not constantly aware of its existence there is the temptation to assume that it doesn’t really matter, and that the mysteries of the inner world are to be unraveled by a consciousness centric approach. And that is a mistake, both in the clean division between conscious and unconscious mind and the assumption that consciousness is the more important of the two.

To make an analogy I would claim that consciousness is like the everyday world of light, sound, and texture. That world is simply an “illusion”, not a false picture, but an incomplete picture. Really understanding the external world requires understand what is going on behind the scenes, that which gives rise to the familiar phenomena. Even though the familiar phenomena reflect what is really going on we can’t simply read off that information from them; how the world really works has to be discovered through laborious experiment, it is not obvious from the use of our senses. I would argue that we are in a similar situation with respect to consciousness. Obviously consciousness is what we immediately confront when we deal with the inner world, just as we immediately confront light and sound when dealing with the outer world. And I would contend that we are unlikely to understand consciousness, or the inner world at a fundamental level by studying it directly, just as we would never have discovered anything interesting about the external world by reflecting on our experience of it. In both cases, it seems to me, what is required is a study of the things that aren’t directly available, and from them an understanding of the immediately accessible phenomena will emerge.

But of course the fact that I can make this analogy doesn’t make it a good analogy. Thus let me provide some of my reasons for thinking that consciousness is a small and relatively unimportant part of the inner world. Consider then the fact that we do not define ourselves as a particular consciousness but as a particular person (agent), who is defined by their memories, desires, dispositions, and so on. Consciousness is a fleeting phenomena, it disappears in sleep, when zoning out, or from a moderate blow to the head. But such disruptions in consciousness, by themselves, don’t bother us; we don’t care that one consciousness exists now and that another will exist some time later, what matters is that they are both expressions of the same person. We only worry about consciousness because it is the vehicle through which who we are is expressed. And the person is a mental construct that exists primarily outside of consciousness. Although who we are shapes our consciousness, the direction of our thoughts, only a small fraction of it is ever part of our consciousness at a single time (due to its complexity, it is too “big” for us to be aware of all the details at once). So, from a stance of what is important to us, consciousness itself is relatively unimportant, at least compared to the rest of the mind.

Another reason to believe that consciousness is less important then we might think is the observation that consciousness is not totally in control. Intuitively we think of consciousness as the driver of the mind, a natural way to look at things since we consider ourselves in charge and consciousness constitutes our perspective on the world. But consciousness is not in charge, although it is not powerless neither is it like a driver who makes all the decisions. Consciousness is directed, in many ways, by other parts of the mind. What information makes its way into consciousness, which ideas hold our attention, which new thoughts spring unbidden into mind, all these things are ways in which other parts of the mind direct consciousness, and thus ways in which consciousness is not in control. Of course we are in control, since, as established earlier, we do not identify ourselves with an episode of consciousness, but as a person, and it is the mostly unconscious structures that define this person that are directing consciousness from behind the scenes (at least in many cases). This makes sense if we understand consciousness as playing a particular function within the mind, as a process that brings intelligence to bear on the problems at hand. Certainly this seems to agree with our experience of consciousness. When intelligence is not needed (such as when we completely zone out) there is no consciousness. And what we are conscious of is usually only things that need thinking about. For example, I am conscious of the content of what I intend to write, but the exact words are supplied by what seems to be an unconscious process, with consciousness then getting a chance to review them in order to improve tricky sentences, and so on. Of course this isn’t a scientifically verified conclusion, just an informed guess. But if it is somewhere close to the truth then it explains a good deal, and points to the fact that consciousness is simply one part of the inner world among many.

So to understand the inner world what we want is a better understanding of the person, what I would call in general “agents”. To understand an agent is not to understand the physics behind how the agent works, we already have a pretty good grip on that in the human case. What we want is an understanding of the agent’s internal structure. How it models the world, how it learns, how it processes information, how its desires influence its behavior and are in turn influenced by experiences. When we understand these things then we will understand ourselves and our inner world. Obviously consciousness is one part of the puzzle, but it doesn’t stand by itself; and even if it could be understood by itself it still wouldn’t provide the understanding of ourselves or our inner world that presumably motivates its study in the first place.

« Previous PageNext Page »

Blog at WordPress.com.