Lying? Stupid? just untrustworthy

We still find ourselves debating whether an obviously false statement is due to lying or stupidity. I hoped this question would become less relevant with the end of the Bush administration but I was over-optimistic, as the recent health care “debate” has shown.

But trying to make this distinction only helps those uttering the obvious falsehoods. They don’t care about informing us, have no real interest in what’s true or false. Their statements fit Harry Frankfurt’s definition of bullshit.

So let’s just call these people “untrustworthy”. It doesn’t matter if they are lying or stupid. It doesn’t matter why they say these things. We can’t trust them to guide us or inform us. We should pay them as little attention as possible.

Untrustworthy — and unworthy of our regard.

Social fixed points

Austin Henderson in his comment on Dancing toward the singularity starts by remarking on an issue that often troubles people when dealing with reflexive (or reflective) systems:

On UI, when the machine starts modeling us then we have to incorporate that modeling into our usage of it. Which leads to “I think that you think that ….”. Which is broken by popping reflective and talking about the talk. Maybe concurrently with continuing to work. In fact that may the usual case: reflecting *while* you are working.

We need the UIs to support his complexity. You talk about the ability to “support rapid evolution of conventions between the machine and the human,”. …. As for the “largely without conscious human choice” caveat, I think that addresses the other way out of the thinking about thinking infinite regress: practice, practice, practice.

I think our systems need to be reflexive. Certainly our social systems need to be reflective. But then what about the infinite regress that concerns Austin?

There are many specific tricks, but really they all boil down to the same trick: take the fixed point. Fixed points make recursive formal systems, such as lambda calculus, work. They let us find stable structure in dynamic systems. They are great.

Fixed points are easy to describe, but sometimes hard to understand. The basic idea is that you know a system is at a fixed point when you apply a transformation f to the system, and nothing happens. If the state of the system is x, then at the fixed point, f(x) = x — nothing changes. If the system isn’t at a fixed point, then f(x) = x’ — when you apply f to x, you “move” the system to x’.

A given system may have a unique fixed point — for example, well behaved expressions in the lambda calculus have a unique least fixed point. Or a system may have many fixed points, in which case it will get stuck at the one it gets to first. Or it may have no fixed points, in which case it just keeps changing each time you apply f.

Now suppose we have a reflective system. Let’s say we’re modeling a computer system we’re using (as we must to understand it). Let’s also say that at the same time, the system is modeling us, with the goal of (for example) showing us what we want to see at each point. We’d like our behavior and the system’s behavior to converge to a fixed point, where our models don’t change any more — which is to say, we understand each other. If we never reached a fixed point, we’d find it very inconvenient — the system’s behavior would keep changing, and we’d have to keep “chasing” it. This sort of inconvenience does arise, for example, in lists that try to keep your recent choices near the top.

Actually, of course, we probably won’t reach a truly fixed point, just a “quiescent” point that changes much more slowly than it did in the initial learning phase. As we learn new aspects of the system, as our needs change, and perhaps even as the system accumulates a lot more information about us, our respective models will adjust relatively slowly. I don’t know if there is a correct formal name for this sort of slowly changing point.

People model each other in interactions, and we can see people finding fixed points of comfortable interaction, that drift and occasionally change suddenly when they discover some commonality or difference. People can also get locked into very unpleasant fixed points with each other. This might be a good way to think about the sort of pathologies that Ronald Laing called “knots”.

Fixed points are needed within modeling systems, as well as between them. The statistical modeling folks have recently (say the last ten years) found that many models containing loops, which they previous thought were intractable, are perfectly well behaved with the right analysis — they provably converge to the (right) fixed points. This sort of reliably convergent feedback is essential in lots of reasoning paradigms, including the set of compression / decompression algorithms that come closest to the the Shannon bounds on channel capacity.

Unfortunately we typically aren’t taught to analyze systems in terms of this sort of dynamics, and we don’t have good techniques for designing reflexive systems — for example, UIs that model the user and converge on stable, but not excessively stable fixed points. If I’m right that we’re entering an era where our systems will model everything they interact with, including us, we’d better get used to reflexive systems and start working on those design ideas.

Language and populations

After posting my ignorant non-argument about philosophical theories of reference, I was happy to see a post on The Use Theory of Meaning at The Leiter Reports that seemed highly relevant, with abundant comments. Alas, on reading it I found that it was unhelpful in two ways. First, generally the arguments seemed to assume as background that the meaning of linguistic expressions was universal (much more on this below). Second, the discussion was obviously confused – by which I mean that participants disagreed about the meaning of the terms they used, how others paraphrased their positions, etc.

If the first problem is correct, then the whole discussion is fairly pointless. Furthermore I think the first problem creates the conditions for the second, because an assumption of universal meanings for expressions is so far from any actual situation in language that an attempt to base theories on it is likely to lead to chaos.

Here is a clear example of this assumption of “universal meaning”: William Lycan in Some objections to a simple ‘Use’ theory of meaning says “[A rule for the meaning of a name must be] a rule that every competent speaker of your local dialect actually obeys without exception, because it is supposed to constitute the public linguistic meaning of the name.” “A rule that every competent speaker… obeys” is universal in just the sense I mean.

Now, this simply isn’t an accurate way to look at how people actually use language. I hope any readers can see this if they think about some examples of creating and understanding expressions, but I’m not going to argue for it now – maybe in another post. I can imagine all sorts of responses: Chomsky’s competence stance, claims that we have to talk that way to have a meaningful (!) or useful (!) theory, statements that it is some sort of harmless idealization (of a different sort from competence), etc. However given the messes in the philosophy of language now which are (in my opinion) largely due to this background assumption, and the concrete results in linguistics and machine learning that show we can get along just fine without it, I reject any such claim. Again, I’m not going to try to substantiate these bald claims right now – but I’m confident I can, and the Steels paper in the earlier post is a good example.

As my earlier post says, what we actually have is a population. To take the story further, each member has dispositions (rules if you will) about how to use a term, how to compose terms to create more complex meanings, or decompose expressions to recover their meanings, etc. But the dispositions of each member of the population will in general be different in all sorts of ways from those of other members. There is no requirement that these dispositions be completely describable, any more than your disposition to shape your hand as you reach for a cup is completely describable – though they might be remarkably consistent in some ways. As a result, no matter how narrowly we define the circumstances, two members of the population will quite likely differ in some details of their use of expressions in those circumstances.

Even with no total agreement in any particular, language works because (again as mentioned in the earlier post) people can resort to context and can create more context through interaction while trying to understand or make themselves understood. This resort prompts us to adjust our usage dispositions over time to bring them closer together, when we find such adjustment helpful and not too difficult. However it also implies the meaning of any given expression may depend in an unbounded way on its context.

I’ll end this with comments on two related issues. First, even apparently consonant ideas, such as Wittgenstein’s “family resemblances”, typically embed the background “universal meaning” assumption. In Wittgenstein’s metaphor the word “game” refers to a particular family, held together only by those resemblances – but the family is treated as a universally accepted meaning for the term, albeit not conveniently delimited by necessary and sufficient conditions. My use of overlapping (and largely consonant) dispositions is not equivalent to this, as I hope is obvious, perhaps with a little thought. However of course overlapping dispositions can easily give rise to meanings that fit Wittgenstein’s “family resemblances”, and the relationship between two different speakers usage dispositions for a given term should perhaps be seen as a family resemblance.

Second, such things as Gettier problems and difficulties with vagueness seem to me to arise quite directly from this assumption of universal meaning. Given the context dependence of meaning in my proposed (very fundamental) sense, it is not surprising that unusual contexts induce incoherence in our intuitions about meaning. The interpretation of our claims that we’ve seen a barn will depend on whether the listener knows there are lots of fake barns about (and knows that we know or don’t know). A population with varying dispositions about the boundaries of Everest will produce something very like supervaluation, and our actual use of language will take that into account. And so forth.

Solving Kuhn’s problems with reference

Thomas Kuhn expressed grave doubts about whether the protagonists on opposite sides of a given scientific revolution even “live in the same world”. His doubts were based on many historical examples where the opposite sides disagreed deeply about the reference of important terms that they shared. In later essays he emphasized that the participants could understand each other, and rationally make choices about what theory to adopt based on reasonable criteria, but he never gave up this fundamental concern.

Kuhn’s examples from scientific revolutions are especially well documented cases of the sort of shifts of reference that occur all the time in our language. Kuhn’s analyses make clear the stakes: if we want to understand how beliefs can and should change, we need a concept of reference that adequately supports these kinds of changes, and that gives us some basis for judging whether the changes are appropriate.

We can approach the needed concept by observing that a reference can’t be understood in isolation. A speaker’s reference can’t succeed unless it picks out a referent for some listener. Success depends on a shared context; typically others understanding references or making them, but in some cases a context of the same person’s earlier or later referencing. (As a software developer, I am very familiar with the need to understand my earlier self’s choice of references – “When I named this variable, what was I thinking?”)

The typical philosophical definition of reference elides this need for context. For example, the Stanford Encyclopedia of Philosophy begins its article on reference by saying “Reference is a relation that obtains between expressions and what speakers use expressions to talk about.” The article continues with an exploration of the relationship between expressions (typically words) and things; speakers drop out of the picture. And listeners were never really in the picture.

The causal theory of reference in a way admits this need for context, but quickly jumps to defining rules for using very stereotyped historical context to pick out the correct target of a reference. In effect, the elision performed rhetorically in the article is performed formally in these theories of reference – speakers and listeners are simply carrying out their formal roles (more or less perfectly). Reference depends only on the very narrow aspects of the situation specified in the formal account.

This may work as an approximation to the use of some kinds of references in stable, unproblematic situations, but it fails badly in situations where the references of terms are unstable, murky, and being negotiated – that is, in precisely the sort of situations Kuhn documents, and more broadly, the sort of situations where we need to understand how reference actually works! This seems to me to be a philosophical case of looking under the lamp-post when we know the car keys were dropped in the alley.

Other philosophers have taken this goal of a formal account of reference as valid, but then either deemed it “inscrutable” (Quine) or vacuous (deflationists). In either case, we are left with no useful tools to interpret or judge situations involving reference conflict or change. This would be a case of denying that there ever were such things as car keys, or if there were, that they did anything useful.

I think Kuhn suffers from a related problem. He seems to have felt that formal accounts fail in scientific revolutions (as they clearly do), and that only a formal account could explain reference. So during revolutions something very mysterious happens, and people come to live in “different worlds”. Within each world reference could perhaps be formally described; across worlds formal descriptions break down.

Let us use our original intuition to take the analysis of reference in a different direction. “Reference is a relation that obtains between expressions and what speakers use expressions to talk about.” If a speaker is using an expression to talk about something, then there is a listener (at least hypothetically) who will have to understand it (correctly) for the speaker to be successful. Correct understanding, in this sense, means that it picks out an aspect of the world that satisfies the purposes of the speaker. This is the minimal context of reference.

Considering this context leads to a view of reference which is neither formal nor useless. Expressions in context do refer. Given enough context, people are very good at figuring out what a speaker refers to. But the aspects of the situation that could be relevant are unbounded and the process of figuring out can’t be specified by a fixed set of rules. In particular, if participants find that their ability to refer successfully is breaking down for some reason, they resort to searching for potentially relevant new aspects of the situation, and arguing about what sort of rules should be followed – behaviors characteristic of scientific revolutions, as well as other episodes of belief and meaning change.

In this account, even though no complete formal account of reference is possible, reference is not mysterious, opaque or incomprehensible. In fact, the process of achieving consensus on the reference of terms is simple and robust enough that it has been implemented in populations of robots. In these experiments, each robot did live in a “different world” – each robot sensed and classified the world in a unique way. Furthermore, there was never a fixed procedure for them to negotiate agreement on which terms to use. But very quickly, through rough pointing and a desire to coordinate their references, the whole population converged on common terms, often after some disagreement (i.e. different subgroups temporarily agreed on different terms). Under some circumstances (such as a large influx of new robots who didn’t know the consensus terms) references again came “up for grabs” and sometimes got reassigned. None the less, the terms used during any given period of consensus did refer; a speaker could use an expression to pick out a piece of the world, and a listener would reliably guess the correct referent.

In some sense, I am pointing at (but by no means adequately setting forth) a causal theory of reference. However the causal process is one of approximate (and shifting) consensus in a population of speaker/listeners, not one that can be stereotyped as “dubbing” or reduced to any other formal account based on specified factors. I hope my description gives some sense of why I think this addresses Kuhn’s concerns. In addition it provides a natural way to address issues of vague reference, reference to fictional entities, etc. — but such bald assertion needs some followup which won’t be forthcoming right now.

I have a sense, however, that this account would be profoundly unsatisfactory to most philosophers who are concerned with reference. If such a philosopher finds it unsatisfactory (in its goals and means, not its current state of extreme sketchiness), my question is Why? Why have philosophers spent so much effort on such problematic formal approaches, when a relatively simple account based on actual (and simulated) use of reference will do the job?

Learning syntax

This paper by Elman does a good job of showing two things highly relevant to the philosophy of mind (as currently pursued):

  • How statistical learning can acquire compositional structure, and
  • How structural properties of language can be learned without innate syntax.

I see that Gary Marcus has criticized Elman from a (more or less) Fodorian perspective, but Elman has been able to generate exactly the results that were supposed to refute him. The pattern seems to be that critics assume connectionist models that are much weaker than we can actually build today, and much weaker than the facts of human biology and learning would suggest.

Can we declare the poverty of stimulus argument dead now?