Self-Reference

How to Build an Artificial Human

entirelyuseless Science , , , , , , , , , , , , , , , , , , , , , , , , , , ,

I was going to use “Artificial Intelligence” in the title here but realized after thinking about it that the idea is really more specific than that.

I came up with the idea here while thinking more about the problem I raised in an earlier post about a serious obstacle to creating an AI. As I said there:

Current AI systems are not universal, and clearly have no ability whatsoever to become universal, without first undergoing deep changes in those systems, changes that would have to be initiated by human beings. What is missing?

The problem is the training data. The process of evolution produced the general ability to learn by using the world itself as the training data. In contrast, our AI systems take a very small subset of the world (like a large set of Go games or a large set of internet text), and train a learning system on that subset. Why take a subset? Because the world is too large to fit into a computer, especially if that computer is a small part of the world.

This suggests that going from the current situation to “artificial but real” intelligence is not merely a question of making things better and better little by little. There is a more fundamental problem that would have to be overcome, and it won’t be overcome simply by larger training sets, by faster computing, and things of this kind. This does not mean that the problem is impossible, but it may turn out to be much more difficult than people expected. For example, if there is no direct solution, people might try to create Robin Hanson’s “ems”, where one would more or less copy the learning achieved by natural selection. Or even if that is not done directly, a better understanding of what it means to “know how to learn,” might lead to a solution, although probably one that would not depend on training a model on massive amounts of data.

Proposed Predictive Model

Perhaps I was mistaken in saying that “larger training sets” would not be enough, at any rate enough to get past this basic obstacle. Perhaps it is enough to choose the subset correctly… namely by choosing the subset of the world that we know to contain general intelligence. Instead of training our predictive model on millions of Go games or millions of words, we will train it on millions of human lives.

This project will be extremely expensive. We might need to hire 10 million people to rigorously lifelog for the next 10 years. This has to be done with as much detail as possible; in particular we would want them recording constant audio and visual streams, as well as much else as possible. If we pay our crew an annual salary of $75,000 for this, this will come to $7.5 trillion; there will be some small additions for equipment and maintenance, but all of this will be very small compared to the salary costs.

Presumably in order to actually build such a large model, various scaling issues would come up and need to be solved. And in principle nothing prevents these from being very hard to solve, or even impossible in practice. But since we do not know that this would happen, let us skip over this and pretend that we have succeeded in building the model. Once this is done, our model should be able to fairly easily take a point in a person’s life and give a fairly sensible continuation over at least a short period of time, just as GPT-3 can give fairly sensible continuations to portions of text.

It may be that this is enough to get past the obstacle described above, and once this is done, it might be enough to build a general intelligence using other known principles, perhaps with some research and refinement that could be done during the years in which our crew would be building their records.

Required Elements

Live learning. In the post discussing the obstacle, I noted that there are two kinds of learning, the type that comes from evolution, and the type that happens during life. Our model represents the type that comes from evolution; unlike GPT-3, which cannot learn anything new, we need our AI to remember what has actually happened during its life and to be able to use this to acquire knowledge about its particular situation. This is not difficult in theory but you would need to think carefully about how this should interact with the general model; you do not want to simply add its particular experiences as another individual example (not that such an addition to an already trained model is simple anyway.)

Causal model. Our AI needs not just a general predictive model of the world, but specifically a causal one; not just the general idea that “when you see A, you will soon see B,” but the idea that “when there is an A — which may or may not be seen — it will make a B, which you may or may not see.” This is needed for many reasons, but in particular, without such a causal model, long term predictions or planning will be impossible. If you take a model like GPT-3 and force it to continue producing text indefinitely, it will either repeat itself or eventually go completely off topic. The same thing would happen to our human life model — if we simply used the model without any causal structure, and forced it to guess what would happen indefinitely far into the future, it would eventually produce senseless predictions.

In the paper Making Sense of Raw Input, published by Google Deepmind, there is a discussion of an implementation of this sort of model, although trained on an extremely easy environment (compared to our task, which would be train it on human lives).

The Apperception Engine attempts to discern the nomological structure that underlies the raw sensory input. In our experiments, we found the induced theory to be very accurate as a predictive model, no matter how many time steps into the future we predict. For example, in Seek Whence (Section 5.1), the theory induced in Fig. 5a allows us to predict all future time steps of the series, and the accuracy of the predictions does not decay with time.

In Sokoban (Section 5.2), the learned dynamics are not just 100% correct on all test trajectories, but they are provably 100% correct. These laws apply to all Sokoban worlds, no matter how large, and no matter how many objects. Our system is, to the best of our knowledge, the first that is able to go from raw video of non-trivial games to an explicit first-order nomological model that is provably correct.

In the noisy sequences experiments (Section 5.3), the induced theory is an accurate predictive model. In Fig. 19, for example, the induced theory allows us to predict all future time steps of the series, and does not degenerate as we go further into the future.

(6.1.2 Accuracy)

Note that this does not have the problem of quick divergence from reality as you predict into the distant future. It will also improve our AI’s live learning:

A system that can learn an accurate dynamics model from a handful of examples is extremely useful for model-based reinforcement learning. Standard model-free algorithms require millions of episodes before they can reach human performance on a range of tasks [31]. Algorithms that learn an implicit model are able to solve the same tasks in thousands of episodes [82]. But a system that learns an accurate dynamics model from a handful of examples should be able to apply that model to plan, anticipating problems in imagination rather than experiencing them in reality [83], thus opening the door to extremely sample efficient model-based reinforcement learning. We anticipate a system that can learn the dynamics of an ATARI game from a handful of trajectories,19 and then apply that model to plan, thus playing at reasonable human level on its very first attempt.

(6.1.3. Data efficiency)

“We anticipate”, as in Google has not yet built such a thing, but that they expect to be able to build it.

Scaling a causal model to work on our human life dataset will probably require some of the most difficult new research of this entire proposal.

Body. In order to engage in live learning, our AI needs to exist in the world in some way. And for the predictive model to do it any good, the world that it exists in needs to be a roughly human world. So there are two possibilities: either we simulate a human world in which it will possess a simulated human body, or we give it a robotic human-like body that will exist physically in the human world.

In relationship to our proposal, these are not very different, but the former is probably more difficult, since we would have to simulate pretty much the entire world, and the more distant our simulation is from the actual world, the less helpful its predictive model would turn out to be.

Sensation. Our AI will need to receive input from the world through something like “senses.” These will need to correspond reasonably well with the data as provided in the model; e.g. since we expect to have audio and visual recording, our AI will need sight and hearing.

Predictive Processing. Our AI will need to function this way in order to acquire self-knowledge and free will, without which we would not consider it to possess general intelligence, however good it might be at particular tasks. In particular, at every point in time it will have predictions, based on the general human-life predictive model and on its causal model of the world, about what will happen in the near future. These predictions need to function in such a way that when it makes a relevant prediction, e.g. when it predicts that it will raise its arm, it will actually raise its arm.

(We might not want this to happen 100% of the time — if such a prediction is very far from the predictive model, we might want the predictive model to take precedence over this power over itself, much as happens with human beings.)

Thought and Internal Sensation. Our AI needs to be able to notice that when it predicts it will raise its arm, it succeeds, and it needs to learn that in these cases its prediction is the cause of raising the arm. Only in this way will its live learning produce a causal model of the world which actually has self knowledge: “When I decide to raise my arm, it happens.” This will also tell it the distinction between itself and the rest of the world; if it predicts the sun will change direction, this does not happen. In order for all this to happen, the AI needs to be able to see its own predictions, not just what happens; the predictions themselves have to become a kind of input, similar to sight and hearing.

What was this again?

If we don’t run into any new fundamental obstacle along the way (I mentioned a few points where this might happen), the above procedure might be able to actually build an artificial general intelligence at a rough cost of $10 trillion (rounded up to account for hardware, research, and so on) and a time period of 10-20 years. But I would call your attention to a couple of things:

First, this is basically an artificial human, even to the extent that the easiest implementation likely requires giving it a robotic human body. It is not more general than that, and there is little reason to believe that our AI would be much more intelligent than a normal human, or that we could easily make it more intelligent. It would be fairly easy to give it quick mental access to other things, like mathematical calculation or internet searches, but this would not be much faster than a human being with a calculator or internet access. Like with GPT-N, one factor that would tend to limit its intelligence is that its predictive model is based on the level of intelligence found in human beings; there is no reason it would predict it would behave more intelligently, and so no reason why it would.

Second, it is extremely unlikely than anyone will implement this research program anytime soon. Why? Because you don’t get anything out of it except an artificial human. We have easier and less expensive ways to make humans, and $10 trillion is around the most any country has ever spent on anything, and never deliberately on one single project. Nonetheless, if no better way to make an AI is found, one can expect that eventually something like this will be implemented; perhaps by China in the 22nd century.

Third, note that “values” did not come up in this discussion. I mentioned this in one of the earlier posts on predictive processing:

The idea of the “desert landscape” seems to be that this account appears to do away with the idea of the good, and the idea of desire. The brain predicts what it is going to do, and those predictions cause it to do those things. This all seems purely intellectual: it seems that there is no purpose or goal or good involved.

The correct response to this, I think, is connected to what I have said elsewhere about desire and good. I noted there that we recognize our desires as desires for particular things by noticing that when we have certain feelings, we tend to do certain things. If we did not do those things, we would never conclude that those feelings are desires for doing those things. Note that someone could raise a similar objection here: if this is true, then are not desire and good mere words? We feel certain feelings, and do certain things, and that is all there is to be said. Where is good or purpose here?

The truth here is that good and being are convertible. The objection (to my definition and to Clark’s account) is not a reasonable objection at all: it would be a reasonable objection only if we expected good to be something different from being, in which case it would of course be nothing at all.

There was no need for an explicit discussion of values because they are an indirect consequence. What would our AI care about? It would care roughly speaking about the same things we care about, because it would predict (and act on the prediction) that it would live a life similar to a human life. There is definitely no specific reason to think it would be interested in taking over the world, although this cannot be excluded absolutely, since this is an interest that humans sometimes have. Note also that Nick Bostrom was wrong: I have just made a proposal that might actually succeed in making a human-like AI, but there is no similar proposal that would make an intelligent paperclip maximizer.

This is not to say that we should not expect any bad behavior at all from such a being; the behavior of the AI in the film Ex Machina is a plausible fictional representation of what could go wrong. Since what it is “trying” to do is to get predictive accuracy, and its predictions are based on actual human lives, it will “feel bad” about the lack of accuracy that results from the fact that it is not actually human, and it may act on those feelings.

Some Remarks on GPT-N

entirelyuseless Metaphysics, Probability, Science , , , , , , , , , , , , , , , ,

At the end of May, OpenAI published a paper on GPT-3, a language model which is a successor to their previous version, GPT-2. While quite impressive, the reaction from many people interested in artificial intelligence has been seriously exaggerated. Sam Altman, OpenAI’s CEO, has said as much himself:

The GPT-3 hype is way too much. It’s impressive (thanks for the nice compliments!) but it still has serious weaknesses and sometimes makes very silly mistakes. AI is going to change the world, but GPT-3 is just a very early glimpse. We have a lot still to figure out.

I used “GPT-N” in the title here because most of the comments I intend to make are almost completely general, and will apply to any future version that uses sufficiently similar methods.

What it does

GPT-3 is a predictive language model, that is, given an input text it tries to predict what would come next, much in the way that if you read the first few words of this sentence with the rest covered up, you might try to guess what would be likely to come next. To the degree that it does this well, it can be used to generate text from a “prompt,” that is, we give it something like a few words or a few sentences, and then add whatever it predicts should come next. For example, let’s take this very blog post and see what GPT-3 would like to say:

What it doesn’t do

While GPT-3 does seem to be able to generate some pretty interesting results, there are several limitations that need to be taken into account when using it.

First and foremost, and most importantly, it can’t do anything without a large amount of input data. If you want it to write like “a real human,” you need to give it a lot of real human writing. For most people, this means copying and pasting a lot. And while the program is able to read through that and get a feel for the way humans communicate, you can’t exactly use it to write essays or research papers. The best you could do is use it as a “fill in the blank” tool to write stories, and that’s not even very impressive.

While the program does learn from what it reads and is quite good at predicting words and phrases based on what has already been written, this method isn’t very effective at producing realistic prose. The best you could hope for is something like the “Deep Writing Machine” Twitter account, which spits out disconnected phrases in an ominous, but very bland voice.

In addition, the model is limited only to language. It does not understand context or human thought at all, so it has no way of tying anything together. You could use it to generate a massive amount of backstory and other material for a game, but that’s about it.

Finally, the limitations in writing are only reinforced by the limitations in reading. Even with a large library to draw on, the program is only as good as the parameters set for it. Even if you set it to the greatest writers mankind has ever known, without any special parameters, its writing would be just like anyone else’s.

The Model

GPT-3 consists of several layers. The first layer is a “memory network” that involves the program remembering previously entered data and using it when appropriate (i.e. it remembers commonly misspelled words and frequently used words). The next layer is the reasoning network, which involves common sense logic (i.e. if A, then B). The third is the repetition network, which involves pulling previously used material from memory and using it to create new combinations (i.e. using previously used words in new orders).

I added the bold formatting, the rest is as produced by the model. This was also done in one run, without repetitions. This is an important qualification, since many examples on the internet have been produced by deleting something produced by the model and forcing it to generate something new until something sensible resulted. Note that the model does not seem to have understood my line, “let’s take this very blog post and see what GPT-3 would like to say.” That is, rather than trying to “say” anything, it attempted to continue the blog post in the way I might have continued it without the block quote.

Truth vs Probability of Text

If we interpret the above text from GPT-3 “charitably”, much of it is true or close to true. But I use scare quotes here because when we speak of interpreting human speech charitably, we are assuming that someone was trying to speak the truth, and so we think, “What would they have meant if they were trying to say something true?” The situation is different here, because GPT-3 has no intention of producing truth, nor of avoiding it. Insofar as there is any intention, the intention is to produce the text which would be likely to come after the input text; in this case, as the input text was the beginning of this blog post, the intention was to produce the text that would likely follow in such a post. Note that there is an indirect relationship with truth, which explains why there is any truth at all in GPT-3’s remarks. If the input text is true, it is at least somewhat likely that what would follow would also be true, so if the model is good at guessing what would be likely to follow, it will be likely to produce something true in such cases. But it is just as easy to convince it to produce something false, simply by providing an input text that would be likely to be followed by something false.

This results in an absolute upper limit on the quality of the output of a model of this kind, including any successor version, as long as the model works by predicting the probability of the following text. Namely, its best output cannot be substantially better than the best content in its training data, which is in this version is a large quantity of texts from the internet. The reason for this limitation is clear; to the degree that the model has any intention at all, the intention is to reflect the training data, not to surpass it. As an example, consider the difference between Deep Mind’s AlphaGo and AlphaGo Zero. AlphaGo Zero is a better Go player than the original AlphaGo, and this is largely because the original is trained on human play, while AlphaGo Zero is trained from scratch on self play. In other words, the original version is to some extent predicting “what would a Go player play in this situation,” which is not the same as predicting “what move would win in this situation.”

Now I will predict (and perhaps even GPT-3 could predict) that many people will want to jump in and say, “Great. That shows you are wrong. Even the original AlphaGo plays Go much better than a human. So there is no reason that an advanced version of GPT-3 could not be better than humans at saying things that are true.”

The difference, of course, is that AlphaGo was trained in two ways, first on predicting what move would be likely in a human game, and second on what would be likely to win, based on its experience during self play. If you had trained the model only on predicting what would follow in human games, without the second aspect, the model would not have resulted in play that substantially improved upon human performance. But in the case of GPT-3 or any model trained in the same way, there is no selection whatsoever for truth as such; it is trained only to predict what would follow in a human text. So no successor to GPT-3, in the sense of a model of this particular kind, however large, will ever be able to produce output better than human, or in its own words, “its writing would be just like anyone else’s.”

Self Knowledge and Goals

OpenAI originally claimed that GPT-2 was too dangerous to release; ironically, they now intend to sell access to GPT-3. Nonetheless, many people, in large part those influenced by the opinions of Nick Bostrom and Eliezer Yudkowsky, continue to worry that an advanced version might turn out to be a personal agent with nefarious goals, or at least goals that would conflict with the human good. Thus Alexander Kruel:

GPT-2: *writes poems*
Skeptics: Meh
GPT-3: *writes code for a simple but functioning app*
Skeptics: Gimmick.
GPT-4: *proves simple but novel math theorems*
Skeptics: Interesting but not useful.
GPT-5: *creates GPT-6*
Skeptics: Wait! What?
GPT-6: *FOOM*
Skeptics: *dead*

In a sense the argument is moot, since I have explained above why no future version of GPT will ever be able to produce anything better than people can produce themselves. But even if we ignore that fact, GPT-3 is not a personal agent of any kind, and seeks goals in no meaningful sense, and the same will apply to any future version that works in substantially the same way.

The basic reason for this is that GPT-3 is disembodied, in the sense of this earlier post on Nick Bostrom’s orthogonality thesis. The only thing it “knows” is texts, and the only “experience” it can have is receiving an input text. So it does not know that it exists, it cannot learn that it can affect the world, and consequently it cannot engage in goal seeking behavior.

You might object that it can in fact affect the world, since it is in fact in the world. Its predictions cause an output, and that output is in the world. And that output and be reintroduced as input (which is how “conversations” with GPT-3 are produced). Thus it seems it can experience the results of its own activities, and thus should be able to acquire self knowledge and goals. This objection is not ultimately correct, but it is not so far from the truth. You would not need extremely large modifications in order to make something that in principle could acquire self knowledge and seek goals. The main reason that this cannot happen is the “P in “GPT,” that is, the fact that the model is “pre-trained.” The only learning that can happen is the learning that happens while it is reading an input text, and the purpose of that learning is to guess what is happening in the one specific text, for the purpose of guessing what is coming next in this text. All of this learning vanishes upon finishing the prediction task and receiving another input. A secondary reason is that since the only experience it can have is receiving an input text, even if it were given a longer memory, it would probably not be possible for it to notice that its outputs were caused by its predictions, because it likely has no internal mechanism to reflect on the predictions themselves.

Nonetheless, if you “fixed” these two problems, by allowing it to continue to learn, and by allowing its internal representations to be part of its own input, there is nothing in principle that would prevent it from achieving self knowledge, and from seeking goals. Would this be dangerous? Not very likely. As indicated elsewhere, motivation produced in this way and without the biological history that produced human motivation is not likely to be very intense. In this context, if we are speaking of taking a text-predicting model and adding on an ability to learn and reflect on its predictions, it is likely to enjoy doing those things and not much else. For many this argument will seem “hand-wavy,” and very weak. I could go into this at more depth, but I will not do so at this time, and will simply invite the reader to spend more time thinking about it. Dangerous or not, would it be easy to make these modifications? Nothing in this description sounds difficult, but no, it would not be easy. Actually making an artificial intelligence is hard. But this is a story for another time.

And Fire by Fire

entirelyuseless Metaphysics, Uncategorized , , , , , , , , , , , , ,

Superstitious Nonsense asks about the last post:

So the answer here is that -some- of the form is present in the mind, but always an insufficient amount or accuracy that the knowledge will not be “physical”? You seem to be implying the part of the form that involves us in the self-reference paradox is precisely the part of the form that gives objects their separate, “physical” character. Is this fair? Certainly, knowing progressively more about an object does not imply the mental copy is becoming closer and closer to having a discrete physicality.

I’m not sure this is the best way to think about it. The self-reference paradox arises because we are trying to copy ourselves into ourselves, and thus we are adding something into ourselves, making the copy incomplete. The problem is not that there is some particular “part of the form” that we cannot copy, but that it is in principle impossible to copy it perfectly. This is different from saying that there is some specific “part” that cannot be copied.

Consider what happens when we make “non-physical” copies of something without involving a mind. Consider the image of a gold coin. There are certain relationships common to the image and to a gold coin in the physical world. So you could say we have a physical gold coin, and a non-physical one.

But wait. If the image of the coin is on paper, isn’t that a physical object? Or if the image is on your computer screen, isn’t your screen a physical object? And the image is just the colors on the screen, which are apparently just as “physical” (or non-physical) as the color of the actual coin. So why we would say that “this is not a physical coin?”

Again, as in the last post, the obvious answer is that the image is not made out of gold, while the physical coin is. But why not? Is it that the image is not accurate enough? If we made it more accurate, would it be made out of gold, or become closer to being made out of gold? Obviously not. This is like noting that a mental copy does not become closer and closer to being a physical one.

In a sense it is true that the reason the image of the coin is not physical is that it is not accurate enough. But that is because it cannot be accurate enough: the fact that it is an image positively excludes the copying of certain relationships. Some aspects can be copied, but others cannot be copied at all, as long as it is an image. On the other hand, you can look at this from the opposite direction: if you did copy those aspects, the image would no longer be an image, but a physical coin.

As a similar example, consider the copying of a colored scene into black and white. We can copy some aspects of the scene by using various shades of gray, but we cannot copy every aspect of the scene. There are simply not enough differences in a black and white image to reflect every aspect of a colored scene. The black and white image, as you make it more accurate, does not become closer to being colored, but this is simply because there are aspects of the colored scene that you never copy. If you do insist on copying those aspects, you will indeed make the black and white image into a colored image, and thus it will no longer be black and white.

The situation becomes significantly more complicated when we talk about a mind. In one way, there is an important similarity. When we say that the copy in the mind is “not physical,” that simply means that it is a copy in the mind, just as when we say that the image of the coin is not physical, it means that it is an image, made out of the stuff that images are made of. But just as the image is physical anyway, in another sense, so it is perfectly possible that the mind is physical in a similar sense. However, this is where things begin to become confusing.

Elsewhere, I discussed Aristotle’s argument that the mind is immaterial. Considering the cases above, we could put his argument in this way: the human brain is a limited physical object. So as long as the brain remains a brain, there are simply not enough potential differences in it to model all possible differences in the world, just as you cannot completely model a colored scene using black and white. But anything at all can be understood. Therefore we cannot be understanding by using the brain.

I have claimed myself that anything that can be, can be understood. But this needs to be understood generically, rather than as claiming that it is possible to understand reality in every detail simultaneously. The self-reference paradox shows that it is impossible in principle for a knower that copies forms into itself to understand itself in every aspect at once. But even apart from this, it is very obvious that we as human beings cannot understand every aspect of reality at once. This does not even need to be argued: you cannot even keep everything in mind at once, let alone understand every detail of everything. This directly suggests a problem with Aristotle’s argument: if being able to know all things suggests that the mind is immaterial, the obvious fact that we cannot know all things suggests that it is not.

Nonetheless, let us see what happens if we advance the argument on Aristotle’s behalf. Admittedly, we cannot understand everything at once. But in the case of the colored scene, there are aspects that cannot be copied at all into the black and white copy. And in the case of the physical coin, there are aspects that cannot be copied at all into the image. So if we are copying things into the brain, doesn’t that mean that there should be aspects of reality that cannot be copied at all into the mind? But this is false, since it would not only mean that we can’t understand everything, but it would also mean that there would be things that we cannot think about at all, and if it is so, then it is not so, because in that case we are right now talking about things that we supposedly cannot talk about.

Copying into the mind is certainly different from copying into a black and white scene or copying into a picture, and this does get at one of the differences. But the difference here is that the method of copying in the case of the mind is flexible, while the method of copying in the case of the pictures is rigid. In other words, we have a pre-defined method of copying in the case of the pictures that, from the beginning, only allows certain aspects to be copied. In the case of the mind, we determine the method differently from case to case, depending on our particular situation and the thing being copied. The result is that there is no particular aspect of things that cannot be copied, but you cannot copy every aspect at once.

In answer to the original question, then, the reason that the “mental copy” always remains mental is that you never violate the constraints of the mind, just as a black and white copy never violates the constraints of being black and white. But if you did violate the constraints of the black and white copy by copying every aspect of the scene, the image would become colored. And similarly, if you did violate the constraints of the mind in order to copy every aspect of reality, your mind would cease to be, and it would instead become the thing itself. But there is no particular aspect of “physicality” that you fail to copy: rather, you just ensure that one way or another you do not violate the constraints of the mind that you have.

Unfortunately, the explanation here for why the mind can copy any particular aspect of reality, although not every aspect at once, is rather vague. Perhaps a clearer explanation is possible? In fact, someone could use the vagueness to argue for Aristotle’s position and against mine. Perhaps my account is vague because it is wrong, and there is actually no way for a physical object to receive copied forms in this way.

Tautologies Not Trivial

entirelyuseless Logic, Mathematics, Metaphysics, Science, Uncategorized , , , , , , , , , , , , , , , , ,

In mathematics and logic, one sometimes speaks of a “trivial truth” or “trivial theorem”, referring to a tautology. Thus for example in this Quora question, Daniil Kozhemiachenko gives this example:

The fact that all groups of order 2 are isomorphic to one another and commutative entails that there are no non-Abelian groups of order 2.

This statement is a tautology because “Abelian group” here just means one that is commutative: the statement is like the customary example of asserting that “all bachelors are unmarried.”

Some extend this usage of “trivial” to refer to all statements that are true in virtue of the meaning of the terms, sometimes called “analytic.” The effect of this is to say that all statements that are logically necessary are trivial truths. An example of this usage can be seen in this paper by Carin Robinson. Robinson says at the end of the summary:

Firstly, I do not ask us to abandon any of the linguistic practises discussed; merely to adopt the correct attitude towards them. For instance, where we use the laws of logic, let us remember that there are no known/knowable facts about logic. These laws are therefore, to the best of our knowledge, conventions not dissimilar to the rules of a game. And, secondly, once we pass sentence on knowing, a priori, anything but trivial truths we shall have at our disposal the sharpest of philosophical tools. A tool which can only proffer a better brand of empiricism.

While the word “trivial” does have a corresponding Latin form that means ordinary or commonplace, the English word seems to be taken mainly from the “trivium” of grammar, rhetoric, and logic. This would seem to make some sense of calling logical necessities “trivial,” in the sense that they pertain to logic. Still, even here something is missing, since Robinson wants to include the truths of mathematics as trivial, and classically these did not pertain to the aforesaid trivium.

Nonetheless, overall Robinson’s intention, and presumably that of others who speak this way, is to suggest that such things are trivial in the English sense of “unimportant.” That is, they may be important tools, but they are not important for understanding. This is clear at least in our example: Robinson calls them trivial because “there are no known/knowable facts about logic.” Logical necessities tell us nothing about reality, and therefore they provide us with no knowledge. They are true by the meaning of the words, and therefore they cannot be true by reason of facts about reality.

Things that are logically necessary are not trivial in this sense. They are important, both in a practical way and directly for understanding the world.

Consider the failure of the Mars Climate Orbiter:

On November 10, 1999, the Mars Climate Orbiter Mishap Investigation Board released a Phase I report, detailing the suspected issues encountered with the loss of the spacecraft. Previously, on September 8, 1999, Trajectory Correction Maneuver-4 was computed and then executed on September 15, 1999. It was intended to place the spacecraft at an optimal position for an orbital insertion maneuver that would bring the spacecraft around Mars at an altitude of 226 km (140 mi) on September 23, 1999. However, during the week between TCM-4 and the orbital insertion maneuver, the navigation team indicated the altitude may be much lower than intended at 150 to 170 km (93 to 106 mi). Twenty-four hours prior to orbital insertion, calculations placed the orbiter at an altitude of 110 kilometers; 80 kilometers is the minimum altitude that Mars Climate Orbiter was thought to be capable of surviving during this maneuver. Post-failure calculations showed that the spacecraft was on a trajectory that would have taken the orbiter within 57 kilometers of the surface, where the spacecraft likely skipped violently on the uppermost atmosphere and was either destroyed in the atmosphere or re-entered heliocentric space.[1]

The primary cause of this discrepancy was that one piece of ground software supplied by Lockheed Martin produced results in a United States customary unit, contrary to its Software Interface Specification (SIS), while a second system, supplied by NASA, expected those results to be in SI units, in accordance with the SIS. Specifically, software that calculated the total impulse produced by thruster firings produced results in pound-force seconds. The trajectory calculation software then used these results – expected to be in newton seconds – to update the predicted position of the spacecraft.

It is presumably an analytic truth that the units defined in one way are unequal to the units defined in the other. But it was ignoring this analytic truth that was the primary cause of the space probe’s failure. So it is evident that analytic truths can be extremely important for practical purposes.

Such truths can also be important for understanding reality. In fact, they are typically more important for understanding than other truths. The argument against this is that if something is necessary in virtue of the meaning of the words, it cannot be telling us something about reality. But this argument is wrong for one simple reason: words and meaning themselves are both elements of reality, and so they do tell us something about reality, even when the truth is fully determinate given the meaning.

If one accepts the mistaken argument, in fact, sometimes one is led even further. Logically necessary truths cannot tell us anything important for understanding reality, since they are simply facts about the meaning of words. On the other hand, anything which is not logically necessary is in some sense accidental: it might have been otherwise. But accidental things that might have been otherwise cannot help us to understand reality in any deep way: it tells us nothing deep about reality to note that there is a tree outside my window at this moment, when this merely happens to be the case, and could easily have been otherwise. Therefore, since neither logically necessary things, nor logically contingent things, can help us to understand reality in any deep or important way, such understanding must be impossible.

It is fairly rare to make such an argument explicitly, but it is a common implication of many arguments that are actually made or suggested, or it at least influences the way people feel about arguments and understanding.  For example, consider this comment on an earlier post. Timocrates suggests that (1) if you have a first cause, it would have to be a brute fact, since it doesn’t have any other cause, and (2) describing reality can’t tell us any reasons but is “simply another description of how things are.” The suggestion behind these objections is that the very idea of understanding is incoherent. As I said there in response, it is true that every true statement is in some sense “just a description of how things are,” but that was what a true statement was meant to be in any case. It surely was not meant to be a description of how things are not.

That “analytic” or “tautologous” statements can indeed provide a non-trivial understanding of reality can also easily be seen by example. Some examples from this blog:

Good and being. The convertibility of being and goodness is “analytic,” in the sense that carefully thinking about the meaning of desire and the good reveals that a universe where existence as such was bad, or even failed to be good, is logically impossible. In particular, it would require a universe where there is no tendency to exist, and this is impossible given that it is posited that something exists.

Natural selection. One of the most important elements of Darwin’s theory of evolution is the following logically necessary statement: the things that have survived are more likely to be the things that were more likely to survive, and less likely to be the things that were less likely to survive.

Limits of discursive knowledge. Knowledge that uses distinct thoughts and concepts is necessarily limited by issues relating to self-reference. It is clear that this is both logically necessary, and tells us important things about our understanding and its limits.

Knowledge and being. Kant rightly recognized a sense in which it is logically impossible to “know things as they are in themselves,” as explained in this post. But as I said elsewhere, the logically impossible assertion that knowledge demands an identity between the mode of knowing and the mode of being is the basis for virtually every sort of philosophical error. So a grasp on the opposite “tautology” is extremely useful for understanding.

 

Form and Reality II

entirelyuseless Metaphysics, Uncategorized , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

This is a followup to this earlier post, but will use a number of other threads to get a fuller understanding of the matter. Rather than presenting this in the form of a single essay, I will present it as a number of distinct theses, many of which have already been argued or suggested in various forms elsewhere on the blog.

(1) Everything that exists or can exist has or could have some relationship with the mind: relationship is in fact intrinsic to the nature of existence.

This was argued here, with related remarks in several recent posts. In a sense the claim is not only true but obviously so. You are the one who says or can say “this exists,” and you could not say or understand it unless the thing had or could have some relationship with your mind.

Perhaps this seems a bit unfair to reality, as though the limits of reality were being set by the limits of the thinker. What if there were a limited being that could only think of some things, but other things could exist that it could not think about? It is easy to see that in this situation the limited being does not have the concept of “everything,” and so can neither affirm nor deny (1). It is not that it would affirm it but be mistaken. It would simply never think of it.

Someone could insist: I myself am limited. It might be that there are better thinkers in the world that can think about things I could never conceive of. But again, if you have concept of “everything,” then you just thought of those things: they are the things that those thinkers would think about. So you just thought about them too, and brought them into relationship with yourself.

Thus, anyone who actually has the idea of “everything,” and thinks about the matter clearly, will agree with (1).

(2) Nothing can be true which could not in principle (in some sense of “in principle”) in some way be said to be true.

Thesis (1) can be taken as saying that anything that can be, can also be understood, at least in some way; and thesis (2) can be taken as saying that anything that can be understood, can also be said, at least in some way.

Since language is conventional, this does not need much of an argument. If I think that something exists, and I don’t have a name for it, I can make up a name. If I think that one thing is another thing, but don’t have words for these things, I can make up words for them. Even if I am not quite sure what I am thinking, I can say, “I have a thought in my mind but don’t quite have the words for it,” and in some way I have already put it into words.

One particular objection to the thesis might be made from self-reference paradoxes. The player in the Liar Game cannot correctly say whether the third statement is true or false, even though it is in fact true or false. But note two things: first, he cannot do this while he is playing, but once the game is over, he can explicitly and correctly say whether it was true or false. Second, even while playing, he can say, “the third statement has a truth value,” and in this way he speaks of its truth in a generic way. This is in part why I added the hedges to (2), “at least in some way”, and “in principle.”

(3) Things do not have hidden essences. That is, they may have essences, but those essences can be explained in words.

This follows in a straightforward way from (1) and (2). The essence of a thing is just “what it is,” or perhaps, “what it most truly is.” The question “what is this thing?” is formed with words, and it is evident that anyone who answers the question, will answer the question by using words.

Now someone might object that the essence of a thing might be hidden because perhaps in some cases the question does not have an answer. But then it would not be true that it has an essence but is hidden: rather, it would be false that it has an essence. Similarly, if the question “where is this thing,” does not have any answer, it does not mean the thing is in a hidden place, but that the thing is not in a place at all.

Another objection might be that an essence might be hidden because the answer to the question exists, but cannot be known. A discussion of this would depend on what is meant by “can be known” and “cannot be known” in this context. That is, if the objector is merely saying that we do not know such things infallibly, including the answer to the question, “what is this?”, then I agree, but would add that (3) does not speak to this point one way or another. But if it is meant that “cannot be known” means that there is something there, the “thing in itself,” which in no way can be known or expressed in words, this would be the Kantian error. This is indeed contrary to (3), and implicitly to (1) or (2) or both, but it is also false.

People might also think that the essence cannot be known because they notice that the question “what is this?” can have many legitimate answers, and suppose that one of these, and only one, must be really and truly true, but think that we have no way to find out which one it is. While there are certainly cases where an apparent answer to the question is not a true answer, the main response here is that if both answers are true, both answers are true: there does not need to be a deeper but hidden level where one is true and the other false. There may however be a deeper level which speaks to other matters and possibly explains both answers. Thus I said in the post linked above that the discussion was not limited to “how many,” but would apply in some way to every question about the being of things.

(4) Reductionism, as it is commonly understood, is false.

I have argued this in various places, but more recently and in particular here and here. It is not just one-sided to say for example that the universe and everything in it is just a multitude of particles. It is false, because it takes one of several truths, and says that one is “really” true and that the other is “really” false.

(5) Anti-reductionism, as it is commonly understood, is false.

This follows from the same arguments. Anti-reductionism, as for example the sort advocated by Alexander Pruss, takes the opposite side of the above argument, saying that certain things are “really” one and in no way many. And this is also false.

(6) Form makes a thing to be what it is, and makes it to be one thing.

This is largely a question of definition. It is what is meant by form in this context.

Someone might object that perhaps there is nothing that makes a thing what it is, or there is nothing that makes it one thing. But if it is what it is of itself, or if it is one of itself, then by this definition it is its own form, and we do not necessarily have an issue with that.

Again, someone might say that the definition conflates two potentially distinct things. Perhaps one thing makes a thing what it is, and another thing makes it one thing. But this is not possible because of the convertibility of being and unity: to be a thing at all, is to be one thing.

(7) Form is what is in common between the mind and the thing it understands, and is the reason the mind understands at all.

This is very distinctly not a question of definition. This needs to be proved from (6), along with what we know about understanding.

It is not so strange to think that you would need to have something in common with a thing in order to understand it. Thus Aristotle presents the words of Empedocles:

For ’tis by Earth we see Earth, by Water Water,

By Ether Ether divine, by Fire destructive Fire,

By Love Love, and Hate by cruel Hate.

On the other hand, there is also obviously something wrong with this. I don’t need to be a tree in order to see or think about a tree, and it is not terribly obvious that there is even anything in common between us. In fact, one of Hilary Lawson’s arguments for his anti-realist position is that there frequently seems to be nothing in common between causes and effects, and that therefore there may be (or certainly will be) nothing in common between our minds and reality, and thus we cannot ultimately know anything. Thus he says in Chapter 2 of his book on closure:

For a system of closure to provide a means of intervention in openness and thus to function as a closure machine, it requires a means of converting the flux of openness into an array of particularities. This initial layer of closure will be identified as ‘preliminary closure’. As with closure generally, preliminary closure consists in the realisation of particularity as a consequence of holding that which is different as the same. This is achieved through the realisation of material in response to openness. The most minimal example of a system of closure consists of a single preliminary closure. Such a system requires two discrete states, or at least states that can be held as if they were discrete. It is not difficult to provide mechanical examples of such systems which allow for a single preliminary closure. A mousetrap for example, can be regarded as having two discrete states: it is either set, it is ready, or it has sprung, it has gone off. Many different causes may have led to it being in one state or another: it may have been sprung by a mouse, but it could also have been knocked by someone or something, or someone could have deliberately set it off. In the context of the mechanism all of these variations are of no consequence, it is either set or it has sprung. The diversity of the immediate environment is thereby reduced to single state and its absence: it is either set or it is not set. Any mechanical arrangement that enables a system to alternate between two or more discrete states is thereby capable of providing the basis for preliminary closure. For example, a bell or a gate could function as the basis for preliminary closure. The bell can either ring or not ring, the gate can be closed or not closed. The bell may ring as the result of the wind, or a person or animal shaking it, but the cause of the response is in the context of system of no consequence. The bell either rings or it doesn’t. Similarly, the gate may be in one state or another because it has been deliberately moved, or because something or someone has dislodged it accidentally, but these variations are not relevant in the context of the state of system, which in this case is the position of the gate. In either case the cause of the bell ringing or the gate closing is infinitely varied, but in the context of the system the variety of inputs is not accessible to the system and thus of no consequence.

A useful way to think about Lawson is that he is in some way a disciple of Heraclitus. Thus closure is “holding that which is different as the same,” but in reality nothing is ever the same because everything is in flux. In the context of this passage, the mousetrap is either set or sprung, and so it divides the world into two states, the “set” state and the “sprung” state. But the universes with the set mousetrap have nothing in common with one another besides the set mousetrap, and the universes with the sprung mousetrap have nothing in common with one another besides the sprung mousetrap.

We can see how this could lead to the conclusion that knowledge is impossible. Sight divides parts of the world up with various colors. Leaves are green, the sky is blue, the keyboard I am using is black. But if I look at two different green things, or two different blue things, they may have nothing in common besides the fact that they affected my sight in a similar way. The sky and a blue couch are blue for very different reasons. We discussed this particular point elsewhere, but the general concern would be that we have no reason to think there is anything in common between our mind and the world, and some reason to think there must be something in common in order for us to understand anything.

Fortunately, the solution can be found right in the examples which supposedly suggest that there is nothing in common between the mind and the world. Consider the mousetrap. Do the universes with the set mousetrap have something in common? Yes, they have the set mousetrap in common. But Lawson does not deny this. His concern is that they have nothing else in common. But they do have something else in common: they have the same relationship to the mousetrap, different from the relationship that the universes with the sprung mousetrap have to their mousetrap. What about the mousetrap itself? Do those universes have something in common with the mousetrap? If we consider the relationship between the mousetrap and the universe as a kind of single thing with two ends, then they do, although they share in it from different ends, just as a father and son have a relationship in common (in this particular sense.) The same things will be true in the case of sensible qualities. “Blue” may divide up surface reflectance properties in a somewhat arbitrary way, but it does divide them into things that have something in common, namely their relationship with the sense of sight.

Or consider the same thing with a picture. Does the picture have anything in common with the thing it represents? Since a picture is meant to actually look similar to the eye to the object pictured, it may have certain shapes in common, the straightness of certain lines, and so on. It may have some colors in common. This kind of literal commonness might have suggested to Empedocles that we should know “earth by earth,” but one difference is that a picture and the object look alike to the eye, but an idea is not something that the mind looks at, and which happens to look like a thing: rather the idea is what the mind uses in order to look at a thing at all.

Thus a better comparison would be between the the thing seen and the image in the eye or the activity of the visual cortex. It is easy enough to see by looking that the image in a person’s eye bears some resemblance to the thing seen, even the sort of resemblance that a picture has. In a vaguer way, something similar turns out to be true even in the visual cortex:

V1 has a very well-defined map of the spatial information in vision. For example, in humans, the upper bank of the calcarine sulcus responds strongly to the lower half of visual field (below the center), and the lower bank of the calcarine to the upper half of visual field. In concept, this retinotopic mapping is a transformation of the visual image from retina to V1. The correspondence between a given location in V1 and in the subjective visual field is very precise: even the blind spots are mapped into V1. In terms of evolution, this correspondence is very basic and found in most animals that possess a V1. In humans and animals with a fovea in the retina, a large portion of V1 is mapped to the small, central portion of visual field, a phenomenon known as cortical magnification. Perhaps for the purpose of accurate spatial encoding, neurons in V1 have the smallest receptive field size of any visual cortex microscopic regions.

However, as I said, this is in a much vaguer way. In particular, it is not so much an image which is in common, but certain spatial relationships. If we go back to the idea of the mousetrap, this is entirely unsurprising. Causes and effects will always have something in common, and always in this particular way, namely with a commonality of relationship, because causes and effects, as such, are defined by their relationship to each other.

How does all this bear on our thesis (7)? Consider the color blue, and the question, “what is it to be blue?” What is the essence of blue? We could answer this in at least two different ways:

  1. To be blue is to have certain reflectance properties.
  2. To be blue is to be the sort of thing that looks blue.

But in the way intended, these are one and the same thing. A thing looks blue if it has those properties, and it has those properties if it looks blue. Now someone might say that this is a direct refutation of our thesis, since the visual cortex presumably does not look blue or have those properties when you look at something blue. But this is like Lawson’s claim that the universe has nothing in common with the sprung mousetrap. It does have something in common, if you look at the relationship from the other end. The same thing happens when we consider the meaning of “certain reflectance properties,” and “the sort of thing that looks blue.” We are actually talking about the properties that make a thing look blue, so both definitions are relative to the sense of sight. And this means that sight has something relative in common with them, and the relation it has in common is the very one that defines the nature of blue. As this is what we mean by form (thesis 6), the form of blue must be present in the sense of sight in order to see something blue.

In fact, it followed directly from thesis (1) that the nature of blue would need to include something relative. And it followed from (2) and (3) that the very same nature would turn out to be present in our senses, thoughts, and words.

The same argument applies to the mind as to the senses. I will draw additional conclusions in a later post, and in particular, show the relevance of theses (4) and (5) to the rest.

Self Reference Paradox Summarized

entirelyuseless Logic, Mathematics, Metaphysics, Uncategorized , , , , , , , , , , , , , , ,

Hilary Lawson is right to connect the issue of the completeness and consistency of truth with paradoxes of self-reference.

As a kind of summary, consider this story:

It was a dark and stormy night,
and all the Cub Scouts where huddled around their campfire.
One scout looked up to the Scout Master and said:
“Tell us a story.”
And the story went like this:

It was a dark and stormy night,
and all the Cub Scouts where huddled around their campfire.
One scout looked up to the Scout Master and said:
“Tell us a story.”
And the story went like this:

It was a dark and stormy night,
and all the Cub Scouts where huddled around their campfire.
One scout looked up to the Scout Master and said:
“Tell us a story.”
And the story went like this:

It was a dark and stormy night,
and all the Cub Scouts where huddled around their campfire.
One scout looked up to the Scout Master and said:
“Tell us a story.”
And the story went like this:
etc.

In this form, the story obviously exists, but in its implied form, the story cannot be told, because for the story to be “told” is for it to be completed, and it is impossible for it be completed, since it will not be complete until it contains itself, and this cannot happen.

Consider a similar example. You sit in a room at a desk, and decide to draw a picture of the room. You draw the walls. Then you draw yourself and your desk. But then you realize, “there is also a picture in the room. I need to draw the picture.” You draw the picture itself as a tiny image within the image of your desktop, and add tiny details: the walls of the room, your desk and yourself.

Of course, you then realize that your artwork can never be complete, in exactly the same way that the story above cannot be complete.

There is essentially the same problem in these situations as in all the situations we have described which involve self-reference: the paradox of the liar, the liar game, the impossibility of detailed future prediction, the list of all true statements, Gödel’s theorem, and so on.

In two of the above posts, namely on future prediction and Gödel’s theorem, there are discussions of James Chastek’s attempts to use the issue of self-reference to prove that the human mind is not a “mechanism.” I noted in those places that such supposed proofs fail, and at this point it is easy to see that they will fail in general, if they depend on such reasoning. What is possible or impossible here has nothing to do with such things, and everything to do with self-reference. You cannot have a mirror and a camera so perfect that you can get an actually infinite series of images by taking a picture of the mirror with the camera, but there is nothing about such a situation that could not be captured by an image outside the situation, just as a man outside the room could draw everything in the room, including the picture and its details. This does not show that a man outside the room has a superior drawing ability compared with the man in the room. The ability of someone else to say whether the third statement in the liar game is true or false does not prove that the other person does not have a “merely human” mind (analogous to a mere mechanism), despite the fact that you yourself cannot say whether it is true or false.

There is a grain of truth in Chastek’s argument, however. It does follow that if someone says that reality as a whole is a formal system, and adds that we can know what that system is, their position would be absurd, since if we knew such a system we could indeed derive a specific arithmetical truth, namely one that we could state in detail, which would be unprovable from the system, namely from reality, but nonetheless proved to be true by us. And this is logically impossible, since we are a part of reality.

At this point one might be tempted to say, “At this point we have fully understood the situation. So all of these paradoxes and so on don’t prevent us from understanding reality perfectly, even if that was the original appearance.”

But this is similar to one of two things.

First, a man can stand outside the room and draw a picture of everything in it, including the picture, and say, “Behold. A picture of the room and everything in it.” Yes, as long as you are not in the room. But if the room is all of reality, you cannot get outside it, and so you cannot draw such a picture.

Second, the man in the room can draw the room, the desk and himself, and draw a smudge on the center of the picture of the desk, and say, “Behold. A smudged drawing of the room and everything in it, including the drawing.” But one only imagines a picture of the drawing underneath the smudge: there is actually no such drawing in the picture of the room, nor can there be.

In the same way, we can fully understand some local situation, from outside that situation, or we can have a smudged understanding of the whole situation, but there cannot be any detailed understanding of the whole situation underneath the smudge.

I noted that I disagreed with Lawson’s attempt to resolve the question of truth. I did not go into detail, and I will not, as the book is very long and an adequate discussion would be much longer than I am willing to attempt, at least at this time, but I will give some general remarks. He sees, correctly, that there are problems both with saying that “truth exists” and that “truth does not exist,” taken according to the usual concept of truth, but in the end his position amounts to saying that the denial of truth is truer than the affirmation of truth. This seems absurd, and it is, but not quite so much as appears, because he does recognize the incoherence and makes an attempt to get around it. The way of thinking is something like this: we need to avoid the concept of truth. But this means we also need to avoid the concept of asserting something, because if you assert something, you are saying that it is true. So he needs to say, “assertion does not exist,” but without asserting it. Consequently he comes up with the concept of “closure,” which is meant to replace the concept of asserting, and “asserts” things in the new sense. This sense is not intended to assert anything at all in the usual sense. In fact, he concludes that language does not refer to the world at all.

Apart from the evident absurdity, exacerbated by my own realist description of his position, we can see from the general account of self-reference why this is the wrong answer. The man in the room might start out wanting to draw a picture of the room and everything in it, and then come to realize that this project is impossible, at least for someone in his situation. But suppose he concludes: “After all, there is no such thing as a picture. I thought pictures were possible, but they are not. There are just marks on paper.” The conclusion is obviously wrong. The fact that pictures are things themselves does prevent pictures from being exhaustive pictures of themselves, but it does not prevent them from being pictures in general. And in the same way, the fact that we are part of reality prevents us from having an exhaustive understanding of reality, but it does not prevent us from understanding in general.

There is one last temptation in addition to the two ways discussed above of saying that there can be an exhaustive drawing of the room and the picture. The room itself and everything in it, is itself an exhaustive representation of itself and everything in it, someone might say. Apart from being an abuse of the word “representation,” I think this is delusional, but this a story for another time.