Pur Autre Vie

I'm not wrong, I'm just an asshole

Tuesday, November 13, 2012

You Want the Truth?

I am not trained in philosophy, and I am not well-read in it either.  Moreover, I'm going to make no real effort to represent ideas accurately or fairly, fit my ideas into philosophical traditions, or defend them against what may very well be devastating critiques.  So take this post with a huge grain of salt, and apologies for mangling various ideas.  That said, this is basically what I believe about the world.

1.  Start with a "correspondence theory of truth."  That is, truth is a correspondence between statements/ideas and a mind-independent "real world."  So if there is an apple on a table in the "real world," then the statement "There is an apple on a table" is true, regardless of whether anyone has seen the apple on the table.  On the other hand, if there is no apple on a table, then "There is an apple on a table" is false, regardless of whether people believe it to be true.

I won't worry about the precise nature of the "correspondence" required by this theory.  However, let's focus for a moment on "mind-independent."  The idea is that the real world is "out there," and while you can change the world (for instance by picking up the apple), you cannot change it internally (merely by observing it or thinking about it differently).  My very crude understanding is that quantum physics puts some pressure on this concept, but in any case we will attack it from a different angle.

Note that "reference" becomes very important in this framework, because you can make true statements only if you can successfully refer to things "out there" in the real world.  This raises interesting issues, but I will ignore them for now.

Note also that if you use the correspondence theory to argue about what is true and what is not, you are probably arguing about what is in the "real world" and what isn't.  For instance, does the "real world" include "moral facts," such that one can make "true" statements along the lines of, "It is wrong to make fun of James's physiology"?  And in fact, it turns out that these unanswerable questions ultimately undermine the whole theory, as I will argue in the next step.

But before moving on, I hope that the appeal of the traditional "correspondence" approach is obvious.  And in fact, I think that when all is said and done, we operate on the basis of something very like the traditional approach.  It's true that I will argue for discarding it, but having done that, I think we have to adopt something very similar.  I think this is why the "naive" view of the world is sufficient for all day-to-day purposes, and people don't generally get into trouble by taking it literally.

2.  The problem with the correspondence theory of truth is that it makes truth inaccessible to human minds, because the "real world" may be remote from our observations.  An easy way to see this is to use the traditional "brains in a vat" thought experiment.  Imagine that we are all just brains in a vat, being fed electrical impulses that give us sensations of tables, chairs, apples, etc.  In reality, nothing exists but the vat, the brains, and wires feeding electrical signals into the brains.  In this world, one could not make true statements about apples on tables, because in the "real world" there would be no apples or tables.

Moreover, there is no way to distinguish our world from a brain-in-vat world, so we don't really know whether you can make truthful statements about apples on tables.  It is hard to see how you can convict anyone of perjury in a world that adheres to a correspondence theory of truth.  "Truth" is a concept that cannot be grasped by human minds.

3.  So the next step is to abandon the "mind-independent" aspect of the "real world."  Instead, we turn to an "empirical world" tied to observations that humans make.  This means that even brains in vats can live in a rich "empirical world" populated by tables, apples, etc., and they can make true and false statements about those objects.  Truth is something that humans can understand and evaluate based on empirical evidence.  We can resume putting people on trial for perjury.  (Note that even if you reject the kind of exotic thought experiments on which I based point 2, you still have the problem that understanding the "real world" requires access to that world, and so at best we can only discuss a model of the "real world" derived from our observations, at which point the jump to the "empirical world" is basically accomplished.)

4.  But while the empirical world is based on observational data, a person's empirical world is not merely the sum of his observations.  This is because the observed world is "thin," containing little of the information that we need to evaluate statements and beliefs.  For instance, I have no direct observational basis to believe that bones contain calcium, or that Romania is north of Bulgaria.  On a deeper level, any kind of functioning empirical world must assume that the future will resemble the past, for the reasons advanced by Hume.  And yet we cannot observe the future to confirm that this is so.

Instead of being an aggregation of data, the empirical world is a model extrapolated from observational data (some of which, by the way, is discarded because it doesn't fit the existing model—for instance, dreams and hallucinations).  And now we come to a key point:  modeling is not an exercise in reproducing the world with exactitude.  First of all, as discussed, there is no meaningful sense in which there is a world to reproduce.  But more importantly, modeling is more art than science, and modelers make all kinds of trade-offs for which there is no "right" choice.  Modelers must try to achieve some balance of predictive power, tractability, "fit" with other models, etc.  These are value judgments, and different models will be appropriate for different circumstances and different people (just as one person might want a road map and another might want a topographical map).

5.  So the empirical world is "value-laden" in that it reflects the modeler's subjective choices.  The fact/value distinction must either be abandoned or must be stated with considerably more nuance than it traditionally has been.  (Alan, take note.)  Moreover, there is no longer any separation between statements/beliefs and the world they describe.  You can think of the empirical world as a filter that is applied to beliefs/statements, and if they pass the bar, they become part of the empirical world (and are then used to evaluate other beliefs/statements).  It is like filling in a crossword puzzle:  the solution to one puzzle becomes a clue in another.

Now the empirical world comprises a bunch of statements/beliefs, and so "truth" doesn't inhere in the relationship between the two.  Rather, "true" is a squishy word used to describe an empirical world that is "healthy" or "useful" or simply "good."  (Pragmatism about truth.)  But what makes one empirical world better than another?  Well, an empirical world is probably not very good if it directly contradicts a lot of observational data.  But once you zoom in and try to make fine-grained distinctions, you rapidly lose traction.  This means that there is, in the abstract, potential for a kind of everything-goes subjectivism that repels most people.  In practice, we deal with this in a variety of ways (more below).

6.  It is useful to bifurcate the discussion into two branches:  the individual pursuit of "truth" and truth as a social convention.  For individual purposes, information that comes from other people is merely additional data, and the individual has to assess it according to his model.  You probably want to build a model in which other people exist, and in which observations that they report can sometimes clear the hurdle and become part of your empirical world.  But basically each person carries around his own empirical world with him.

Communities care a lot about truth, though, so they have developed their own empirical worlds, which include a "persuasive" or "rhetorical" element that is not as important for individuals.  That is, provability becomes important in a social context, because we don't want to take action until something has been demonstrated to be true (to our satisfaction).  Thus reasonable doubt and peer review and replication studies.

However, different communities can build different empirical worlds, and those empirical worlds can compete.  In fact, in a sense a lot of ideological fights come down to struggles over the definition of a society's empirical world.  Moreover, it is quite common for people to use rhetoric from the correspondence theory of truth when fighting over society's empirical world.  For instance, the "reality-based community" (the term raises some interesting questions about how the public views "the truth").  In general, this fight over our shared empirical world is fought on many fronts, with practically every weapon imaginable.  A lot of people use deplorable tactics, but there really isn't a way to keep the fight "pure."  A lot is at stake, and it is almost impossible to carry the burden of proof for very long.  So a tremendous amount of effort goes into burden-shifting and re-framing.

7.  In practice, it is hard to see any problem with using the correspondence theory of truth as shorthand for a more nuanced view.  This is basically the model that most of us adopt as our "empirical world" in any case—a "real world" that we observe and try to predict, as if it were the mind-independent world we started with.  Dropping mind-independence is conceptually interesting but doesn't change very much in daily life.  I am quite ready to call people "wrong" and to base this claim on the fact that their statements don't align with reality.  So what is the practical upshot of all of the preceding discussion?

I can only speak for myself, but I have found this way of looking at things to be liberating.  I am less judgmental about different ways of viewing the world (there is some "signal" to be gained even from viewpoints that are mostly wrongheaded), and I take a broader view about what constitutes knowledge.  There is no "one true rationality," and claims along those lines, which attempt to monopolize the discussion, are immediately suspect.

I am also more willing to be judgmental about "normative" questions, since in my mind there is no longer a sharp distinction between those questions and factual ones.  "Facts" are tools to help you navigate life, and values are roughly the same thing.  Both are to be assessed on their overall "healthiness" and not on some metaphysical property that can't be observed.  By their fruits you shall know them.  But I admit that all of this is more of a re-framing than a revolution.

13 Comments:

Blogger Zed said...

I tend to think the "thickness" of scientific statements is overrated: thick statements like "salt contains sodium" are shorthand for inductive predictions that are empirically "thin" (under a certain well-defined procedure it will blow up). The big organizing narratives of science are just filing systems for such thin inductive predictions. Needless to say there is not much epistemic disagreement about whether one of these predictions has actually been confirmed; i.e., the whole process is obviously _persuasive._ Obviously there are thought experiments of the "grue" type but there do not seem to be any agents out there with grue preferences so (as far as I'm concerned) these don't matter.

With "value" questions I don't really see any _path_ to such a reduction, or any underlying cross-cultural (or maybe even intra-cultural) common ground. Perhaps this is because ethics is a primitive science; perhaps it is because the enterprise is doomed; in either case, what matters is that ethical arguments are essentially never persuasive.

3:38 PM  
Blogger James said...

I may be missing your point about sodium. Are you arguing that it is easy to construct an empirical world out of observational data, with little or no modeling required? That is a claim I find hard to accept. I am ready to acknowledge the less disruptive point that much knowledge is "thin" once you assume a social framework along the lines of scientific procedure. Note, there is still a lot of "thick" knowledge, even if we confine ourselves to the hard sciences (which of course we don't).

On value questions, I am not sure whether you find my point about modeling choices persuasive. What makes a modeling choice good or bad? Presumably it depends on the relative value you attribute to accuracy, tractability, etc.

I do think that in practice moral claims are often just as persuasive as "factual" claims. You will get a lot more consensus that rape is wrong than that we evolved by natural selection. And I think the same pattern will appear even if you exclude poorly-educated people - basic moral propositions can easily be more convincing than scientific "facts."

4:02 PM  
Blogger Zed said...

I just don't think it is necessary to construct an empirical world in order to interpret scientific predictions/empirical statements in general. I.e., science connects thin statements in the past to predictions about thin statements in the future, using thick ideas as an intermediate step which can be interpreted purely as a mnemonic device.

I do think it'd be useful to exclude the social sciences for the point of this discussion because otherwise one confuses internal confusion (_within_ a discipline) with purely philosophical difficulties.

I do not think "rape is wrong" is the sort of statement you would get that much agreement on if you actually started with a precise definition of rape. (Also, I am not sure you would get universal consensus that rape of enemies is bad in times of war. Or that, e.g., rape of convicted sex offenders is bad.) I do think there is a sort of mirror symmetry here between inductive and moral statements: with inductive reasoning, it's the high-level pronouncements that people find difficult to accept; the more specific things are the more agreement you get. With moral issues, it's the other way round: everyone agrees with a bunch of meaningless high-level statements but when you actually come down to specific cases you almost always lose consensus.

I don't really take a view on your view on modeling choices. I do think it is useful to keep models mentally distinct from data. I also think one should generally avoid making ontological claims about the entities in models, as it usually seems to be the case that vastly "different" models give very similar predictions.

4:14 PM  
Blogger Zed said...

Actually perhaps the right thing to say about your view of modeling choices in value judgments is more like this. I am mostly concerned with the Q. of how one persuades other people, who (in general) have different models. With empirical questions this is at least a well-defined point in principle: find something the models disagree about, then test it. (I would claim that at least for some subset of elementary claims the tests are not very theory-laden; everyone agrees about eclipses and falling balls and the like.) For values it seems to me that there is no analogous test; or if there is (something like, this model cannot be made to yield any results that cause a large number of stomachs to turn?), then no existing model comes close to passing it.

4:40 PM  
Blogger James said...

I guess I think there is no real alternative to modeling an empirical world. Take something like "seawater contains sodium." However thin this claim is (and I'm willing to treat it as very thin), it is not a "true" statement in a brain/vat world that doesn't contain any seawater, if we use the correspondence theory of truth.

So then, if we've abandoned the correspondence theory of truth, we have to fall back on something. One answer (which I think is where you are going) is to move to "raw data," but I really think raw data needs substantial extrapolation before it can do much of anything. So just for instance, I think your concept of sodium probably includes a bunch of information that is not evident from the way it shows up in a mass spectrometer or whatever it is you're using to detect it. "Atomic number" is a concept that needs significant modeling before it makes any sense. It basically is a model.

But I don't know if there is anything at stake in this discussion, since I don't think there is any alternative to an "empirical world" approach. Maybe you are driving at some kind of "absolute" or "objective" truth, but I don't think that is a possibility. It may be that any reasonable empirical world would contain facts like "seawater contains sodium," but I don't think we can construct some kind of bare-bones evidence-driven "objectively true" body of knowledge.

I'll defer discussion of values, but I hope to return to it.

5:53 PM  
Blogger James said...

By the way, I believe that some pragmatists define truth in terms of some kind of limit - the result that would be reached in idealized circumstances after diligent investigation or whatever. I don't really like that answer, instead preferring to treat truth as inherently provisional. But anyway that concept is out there. (I emphasize once again my abject ignorance in this area. I like pragmatism in the same way a 14-year-old boy thinks he will like sex.)

5:58 PM  
Blogger James said...

It appears that I have been parroting Putnam more than I intended (and with unfortunate duplication of his terminology, despite the fact that I don't think I am capturing his concepts rightly). Here he is in Reason, Truth, and History:


Of course, if metaphysical realism were right, and one could view the aim of science simply as trying to get our notional world to 'match' the world in itself, then one could contend that we are interested in coherence, comprehensiveness, functional simplicity, and instrumental efficacy only because these are instruments to the end of bringing about this 'match.' But the notion of a transcendental match between our representation and the world in itself is nonsense. To deny that we want this kind of metaphysical match with a noumenal world is not to deny that we want the usual sort of empirical fit (as judged by our criteria of rational acceptability) with an empirical world. But the empirical world, as opposed to the noumenal world, depends upon our criteria of rational acceptability (and, of course, vice versa). We use our criteria of rational acceptability to build up a theoretical picture of the 'empirical world' and then as that picture develops we revise our very criteria of rational acceptability in the light of that picture and so on and so on forever. The dependence of our methods on our picture of the world is something I have stressed in my other books; what I wish to stress here is the other side of the dependence, the dependence of the empirical world on our criteria of rational acceptability. What I am saying is that we must have criteria of rational acceptability to even have an empirical world, that these reveal part of our notion of an optimal speculative intelligence. In short, I am saying that the 'real world' depends upon our values (and, again, vice versa).

6:32 PM  
Blogger Zed said...

No I'm not really interested in absolutes either, I suppose I am just much more a formalist than a pragmatist by inclination. I am fine with meaningless unjustifiable games as long as the rules are clear; it seems to me that this is so with science, and not so with ethics, which is why the distinction between the two seems very sharp.

But I do not think my basic claim -- that you can make extrapolations from past events to future events without invoking "models" to more than the sort of minimal extent needed to justify "time" etc. -- is problematic/unclear. One way of seeing the task of science is that it takes in everything that's happened so far, and using the basic induction idea, spews out everything that will happen in the future. Now in order to persuade yourself that your predictions make sense you might need to refer to a model, but as far as _others_ are concerned you could simply make your prediction and let it do the arguing for you. As previously said, I am not interested in the Q. of how people arrive at answers, merely how they justify them.

7:28 PM  
Blogger Tarun Menon said...

Sarang, I'm not sure if your claim is that justification in science actually works the way you describe or that it should work this way. The former claim is almost certainly false, I think. The rhetoric of scientific justification is complex and involves all sorts of moves that can't be reduced to claims of empirical superiority. For instance, proponents of loop quantum gravity often cite manifest background-independence as a major advantage of their theory, but this has nothing to do with making better predictions. Evolutionary theory makes a lot of good predictions, but it's primarily an explanatory science rather than a predictive one, and its main virtue is that it offers a simple, coherent, entirely naturalistic, and, yes, empirically tested explanation for the development of biological complexity. I've been reading a book on the early history of statistical mechanics, and the sorts of arguments employed by Maxwell, Boltzmann and others (arguments that the Second Law is probabilistic, for instance) make very little reference to empirical predictions.

So I don't think your account of scientific justification works as a descriptive theory. It might be a good prescriptive theory -- maybe scientific justification should work the way you describe and cases where it doesn't are mistakes -- but if that's all it is then I don't think ethics is at a disadvantage. Ethicists have produced a number of accounts of how ethical discourse should work that have reasonably clear rules.

9:39 AM  
Blogger Zed said...

I think what I've outlined is a fair descriptive theory of how physicists and chemists would now justify results that were firmly established by, say, 1930. No one in their right mind would appeal to elegance when justifying Einstein's theory. I agree that other standards of justification are sometimes cited, mostly in speculative areas, but I believe they never are when the ones I've outlined are available.

5:08 PM  
Blogger Grobstein said...

I can't really contribute to this in a sophisticated way, but my impression is that Tarun-like factors are sort of unavoidable in scientific inference, even in settled areas of science.

So for example the Copernican model is preferable to the Ptolemaian model because eventually when you've drawn your nth epicycle, you can't shake the feeling that it's ad hoc bullshit. Right?

I guess the Sarang-like response is that it is the ad hoc-ness of the epicycles that makes them bullshit, not the fact that they are inelegant, etc. If circa 1700 you could use your theory of epicycles to predict the orbits of future undiscovered stellar bodies, the Ptolemaic theory would be golden. You can't do that, though, because you can only draw all the epicycles correctly once you know what the observations will be, and you can't predict the observations because you don't have a working model of kinematics.

On this view, elegance and so on are all close to meaningless. The one thing that really matters is strict separation between training data and testing data. You could use epicycles to predict what you'd already seen, but not what you hadn't seen. In fact, if we observe that ad hoc, "overfitted" theories often have extra parameters, we might simply say that elegance etc. are historically-useful heuristics for predictions that will be good outside of training data.

So to extend S's descriptive theory of scientific inference, we can say that in areas where predictive reliability is not a good discriminator, we use things we hope are leading indicators of predictive reliability. The areas where this comes up will be those where we don't have a good supply of untrained data (i.e., principally things that are currently hard to experimentally test).

4:24 PM  
Blogger Grobstein said...

Why yes I did change my mind partway through my comment.

4:25 PM  
Blogger Unknown said...

michael kors, coach outlet, michael kors, coach outlet, sac guess, nike air max, ralph lauren uk, hermes, burberry, north face, north face, michael kors outlet, hollister pas cher, hollister, replica handbags, vans pas cher, true religion jeans, ray ban uk, michael kors, tn pas cher, lacoste pas cher, hogan, nike free run uk, nike blazer, converse pas cher, oakley pas cher, ugg boots, true religion jeans, burberry outlet online, true religion outlet, nike roshe, michael kors, timberland, air force, vanessa bruno, mulberry, nike air max, new balance pas cher, ray ban pas cher, michael kors outlet, michael kors outlet, true religion jeans, ugg boots, kate spade handbags, coach purses, michael kors outlet, nike air max, michael kors outlet, abercrombie and fitch, lululemon

8:00 PM  

Post a Comment

<< Home