Saturday, January 18, 2014

Against the Ideological Turing Test



In 2011, Bryan Caplan came up with an idea that has gained popularity in certain circles: the Ideological Turing Test. Caplan's original formulation, which was in reference to an argument between himself and Paul Krugman, went like this:
We don't have to idly speculate about how well adherents of various ideologies understand each other. We can measure the performance of anyone inclined to boast about his superior insight.  
How?  Here's just one approach.  Put me and five random liberal social science Ph.D.s in a chat room. Let liberal readers ask questions for an hour, then vote on who isn't really a liberal.  Then put Krugman and five random libertarian social science Ph.D.s in a chat room. Let libertarian readers ask questions for an hour, then vote on who isn't really a libertarian. Simple as that. 
Actual ITTs are, of course, not very feasible. For one thing, they would have to be anonymous. For another thing, they would have to involve the participation of several "authentic" members of an ideology, and the certification of authenticity must be performed by someone other than the tester. For these reasons, actual ITTs are not only occasionally done.

But even if they were done, I don't think ITTs would be a good measure of how much someone understands an ideology. This is because ITTs seem relatively easy to pass using Chinese Room tactics. Unlike intelligence itself (which the original Turing Test tests for), ideology has a finite, circumscribed set of inputs and outputs.

For example, suppose I'm taking an Ideological Turing Test for Austrian economics. The questioner asks: "Why must we accept the axiom that Humans Act?" And I answer: "The action axiom is itself a self-referential proposition; the statement, "Humans act," constitutes an action. The goal of the action is the positive assertion of the action axiom; the means is the statement. The positive assertion of the action axiom can be read thus: "This assertion of the action axiom is itself an action." It is thus a self-referential statement. The attempt to deny the action axiom is also self-referential. It amounts to stating, "Action does not exist; therefore, this statement is not an action." 

Now I have no idea what the heck "Humans act" even means, and that answer, which was copied from the Mises Institute website, makes no sense to me either. I don't understand it. I'm not even sure there is something there to understand. But I could give the answer. I could fiddle with the sentence structure and word choice until it sounded more extemporaneous and less like a chapter-and-verse recitation. And I'm betting that there's a decent chance that an Austrian test administrator would pick me as the real Austrian over someone who really believes in Austrian stuff but hasn't memorized the Mises Institute website quite as carefully. And since ideologies are finite, there are only a finite number of such answers I'd have to memorize.

A good real-life example of someone passing a sort of ITT without understanding the ideology in question is the Sokal Hoax, who published a gibberish paper in a postmodern cultural studies journal without knowing the first thing about cultural studies. The reviewers, who undoubtedly rejected lots of other submissions from authors who did know quite a lot about postmodern cultural studies, were very close to Turing Test administrators, and postmodern studies is not too different from an ideology (*ducks*). In fact, I think it would be a lot easier for some version of the Postmodern Essay Generator to get a paper published than for it to pass a real Turing Test of general intelligence.

If you don't believe me, try it out yourself! Go into some political chat room that subscribes to some ideology you disagree with. Try to convince the room that one of the other chatters is a poser and not a true believer. I bet you can do it. Now ask yourself how well you really understand the ideology you just enforced.

To put it bluntly, ideologies are large parts bullshit to begin with, and so it's possible to bullshit your way through ideological tests.

But this is a bit academic. The fact is, for reasons mentioned above, actual Ideological Turing Tests are impractical. Instead, what you usually hear is people saying to an intellectual opponent: "I bet you couldn't pass an Ideological Turing Test." But since public arguments are very, very far away from an actual ITT, this is just bluster. It's a nerdy-sounding way to say "You're too dumb to understand my point."

In the real world, invoking the name of the ITT is usually just a way to insist that an opponent adopt one's preferred terminology, and/or accept certain of one's own premises, before proceeding with the argument. In other words, it is asking for a handicap in a debate. And if the challenge is accepted - if one party agrees to argue only on the other's terms, just to show how broad-minded they are - it tends to impoverish the debate as a whole. Debates are often more productive, and lead to more actual mutual understanding and learning, when each person argues on their own terms.

So I think the Ideological Turing Test should drop the "Turing" part and just call itself what it really is: an Ideological Test. It was an intriguing idea, but I think it's time to put it to rest.


Update:

There are other problems with ITTs.

For example: Suppose someone passes an ITT and then claims not to understand the ideology on which he was tested (thus claiming to have spoofed the test). Is his claim of non-understanding credible? Remember that real Turing Tests define intelligence as the ability to pass the test! Should we define "understanding" using the ITT itself?

Another problem: What does it mean to "understand" something that is logically incoherent? And aren't some parts of at least some ideologies logically incoherent?

Oh, and in case you're wondering, I DO think it's very good to try to understand the point of view, and the ideas, of one's opponents. If the desire to be able to pass a hypothetical ITT motivates you to try hard to understand your opponents' point of views, well then by all means, do it!

Adam Gurri defends the spirit of the ITT. I agree with all his points. I tend to think literally about this sort of thing, I guess. Also, I recently read Blindsight, by Peter Watts, a science fiction novel that deals with non-self-aware intelligence, and I was excited to apply that idea to a blogosphere debate. :-)

34 comments:

  1. Political chat rooms tend to have PhD's from top institutes? He acknowledges that the idea doesn't work if you try it with ordinary people, so that paragraph is just pushing on an open door.

    ReplyDelete
    Replies
    1. Now ask yourself how many times in your life you have pushed on an open door, and whether it was a bad idea when you did it...

      Delete
  2. Anonymous4:48 PM

    Wow you just failed an ITT and proved his point in this blog post.

    ReplyDelete
    Replies
    1. Implying that the ITT itself is an ideology? Sounds about right.

      Delete
  3. Re: The Sokal hoax, it is worth nothing that the editors of "Social Text" (which was no longer much concerned with postmodernism at the time, FYI) said during the controversy that they would not have published it if it had written by someone with a background in the humanities, because it rehashed a lot of outdated points, but they found it interesting because Sokal was a physicist. In other words, it stuck out as different to them, but the author's "outsider" status led them to their decision. If he was trying to pass an ITT, he failed.

    ReplyDelete
    Replies
    1. Riiiiiiiiiiight. ;-)

      Delete
    2. Yeah, I was going to make the same point. It involves taking the Social Text editor's public statements at face value, but for me they passed the "ring of truth" test (close cousin of the turning) [1], and I don't think their excuses reflect well on them at all: they ran Sokal's paper, despite some evident problems, because of his credentials... they thought they had found a new ally in the science wars. Sokal didn't pass as a double-plus good duckspeaker, he passed as a guy who would be useful to have on your side (look, even a *scientist* agrees with us).

      [1] One of the reasons I believe the "Social Text" explanation is (1) they didn't appear to realize how bad it sounded (2) it's long since disappeared off of the net... I have to quote it from memory.

      Delete
  4. As a firm believer that the Chinese Room would be conscious, I think that in a certain sense, being able to recite the appropriate bits of ideology back sufficiently well does constitute understanding. (Although the understanding isn't something you are totally conscious of, at least for the conventional sense of the word you.) If reciting back text from other websites is all you need to replicate an ideology, then that's all there is. If an ideology is all bullshit, then the bullshitter understands it as well as anyone else.

    But I'd predicate that heavily on being able to be "sufficiently convincing." A very poor AI can trick people into thinking it's a person if people are inclined to think they're talking to an intentionally obtuse and obnoxious person. So similarly, if you can convince people that you're a crank, you can pass the ideological test easily. To understand something properly you want to seem like a relatively intelligent adherent of the ideology. But then, what if an ideology is just inherently stupid?

    ReplyDelete
    Replies
    1. Have you read Blindsight, by Peter Watts? I think you'd like it.

      Delete
    2. FWIW, here is a fascinating interview with Daniel Dennett on why the Chinese Room is not actually a very useful or interesting thought experiment:

      http://philosophybites.com/2013/06/daniel-dennett-on-the-chinese-room.html

      Delete
    3. And of course after I clicked "Publish" on that comment Blogger prompted me to "Please prove you're not a robot!"

      Which is awesome.

      Delete
    4. Hehehe.

      Yep, the Chinese Room doesn't really work for real Turing Tests. Weird fact: I read that Dennett article in Philosophy Camp when I was 13!

      But my point is that ideology is a qualitatively easier thing to spoof than consciousness, so a Chinese Room that passed one would actually be a spoof, not just a relabeling of "understanding".

      Delete
    5. I guess part of the issue is that even if you accept that understanding an idea is nothing more than being able to explain and apply that idea, the way the Intellectual Turing Test is typically described isn't really a very good way to test that. If you work out the specifics in just the right way it might get the job done, but as described it's a bit silly.

      Delete
    6. The problem with the Chinese room argument is that it assumes there exists a set of rules that allow the person in the room to pass the test. The actual issue is whether or not such a set of rules could exist so the argument is ultimately circular.

      Delete
  5. Anonymous5:16 PM

    Well obviously a rule of an ITT would be that you can't steal from someone who could pass the ITT like you do in your Mises example. The point of the ITT is to show you understand the argument. To understand an argument you have to be able to produce it yourself (perhaps not in textbook form).

    I don't really understand why you have always been so hostile to the idea of "being able to pass" an ITT. All that simply means is that you aren't attacking a straw man.

    ReplyDelete
    Replies
    1. Sure you can't plagiarize, but plagiarizing is too easy to spot anyway. Paraphrase. If you outlaw paraphrasing then the test is completely impossible.

      Delete
  6. In general, I find it funny that Bryan Caplan was the one propagating the ITT idea in the first place, because in my experience libertarians are the least likely to actually demonstrate that they understand non-libertarian arguments about anything, ever, and also the most likely to claim that a handful of axiomatic tenets of their ideology invalidate or overrule a wide swatch of complex argumentation.

    ReplyDelete
  7. Your statement that "intellectual Turing Tests are not done" isn't quite right - Leah Libresco has been holding a yearly ITT between Catholics and atheists with interesting results. You can see the records at http://www.patheos.com/blogs/unequallyyoked/ideological-turing-test-contest .

    What I take from them is that below a certain level of intelligence and well-informedness ITTs can distinguish people based on factual knowledge and basic worldview assumptions; after that level everyone knows the facts about their opponents but it is still marginally possible to distinguish people based on emotions, emphasis, and styles of expression.

    ReplyDelete
  8. Libertarians get a lot of counterarguments that they view as simplistic (Libertarianism is childish, who will build the roads?, libertarianism works great in SOMALIA, etc), so they've developed kind of a frustrated "nobody understands our depths" mentality. It's pretty insulting for a person to imply that you've never thought about how roads get built or how to prevent a mad max scenario under your preferred form of societal organization. The ITT complaint is basically a bludgeon against an opponent who isn't giving libertarianism credit as a thought-out thing. Problem is, once you're convinced that nobody understands your depths, then anyone who disagrees with you must just not get it and you wind up sticking to the hive mind. It's best to only do ideologies that don't have steep learning curves, like seahawks fandom and brony.

    ReplyDelete
  9. Regarding the action axiom, I think the idea is that in order to refute the action axiom, you have to act, which would be logically contradictory to the idea that "humans act".

    This is incredibly tortured, absurd logic. And how Mises gets from the action axiom to the other stuff is even more tortured, but that's another story for another day.

    Anyway, I like ideological Turning tests, at least for my own purposes. I use them to try and understand how other people think and the angle they are coming at things from. I couldn't care less about them in any other context, especially pissing contests.

    ReplyDelete
    Replies
    1. It seems to me that denying the action axiom is, at best, only momentarily self-contradictory. An obvious philosophical objection is that you could easily have "not acted" outside of the present Catch-22 moment. An analogy that I have used before:

      MARIO: "Hey, Luigi. Are you awake."
      LUIGI: "No."
      MARIO: "A-ha! But you must be awake, since otherwise you could not have answered me. Proof that you never sleep!"

      When I have pointed this out to Austrian types in the past, their response in part has been to say that Mises was only limiting his analysis to "purposeful" action. Okay, but:

      a) This then just makes the laboured invocation of Kant, synthetic a priori truths and all that completely redundant! You have already specified the terms and limits of your engagement -- i.e. purposeful behaviour -- at the outset and so you don't need to go through the whole rigmarole trying do deny that humans "act". (By doing so you are just repeating the tautology that purposeful action is purposeful action.)

      b) It should certainly invalidate sweeping claims made about praxeology by certain proponents, regarding the extent to which it universally applies in our day-to-day lives... or renders economics an non-empirical science. (And this is completely disregarding the very dubious steps that Mises makes in getting from the action axiom to the rest of his theory, which John alludes to above.)

      I mean, is it really all that different to saying that neoclassical economics is aimed at explaining human behaviour to the extent that agents are rational and narrowly self-interested? I don't see it.

      Delete
  10. "Unlike intelligence itself (which the original Turing Test tests for)"

    No, it tests for the ability to *simulate* intelligence.

    ReplyDelete
    Replies
    1. But the Turing Test is a proposed definition of intelligence; pass it, the theory goes, and you *are* intelligent.

      Delete
  11. Anonymous11:58 PM

    A better ITT: Crack a mean joke at the expense of a certain group -- if the person goes quite or becomes defensive, they are part of said group; if they laugh or reply with a joke of their own, they pass.

    ReplyDelete
  12. A further problem is that it's hard to define what ideology you are testing and therefore to select the chatroom partners. That's a kind of Duhem-Quine-ITT problem. The chatroom partners would have to pass an ITT first, so how do we select them?

    ReplyDelete
  13. Anonymous8:47 AM

    Noah, your argument here is basically that the ITT is dumb because people can just regurgitate answers they've memorized or read without understanding them. This is a bad argument, because this same complaint can be applied to almost all tests or exams.

    Should we eliminate exams in school or university because students can write memorized answers they don't understand?

    ReplyDelete
    Replies
    1. ...now that you mention it...

      Delete
    2. Exams don't test understanding but memory but this is sufficiently correlated with intelligence and knowledge that understanding, or at least understandability, is inferred.

      Delete
    3. My experience, with mathematics exams at least, is that you can only make it so far with memorizing definitions and theorems and practice problems. Professors will usually throw in a sufficiently modified question on an exam which can only be answered if the student has gained some deeper insight into the course material. But I doubt this can be done in all subject areas.

      Delete
  14. Noah,

    Good point.

    I was never convinced that the ideological turing test could be very helpful. If anything, I think it would favor the less logical ideologies. For example, a UFO enthusiast would most probably pass an ideological turing test for UFO skepticistm. "Why don't you believe UFOs are all around?" "There has never been any good evidence of it." But if roles were reversed, a number of arbitrary but consistent (in the sense of being always the same) explanations might be necessary to convince people, and the skeptic would fail.

    ReplyDelete
  15. There is much to be said for understanding alternate points of view however: (1) life is short; (2) there are more points of view than you can understand in a lifetime; and (3) some points of view have no prospect of attracting a wide enough audience to be relevant to the ever changing political compromise at the heart of all western nations. Therefor, when one reads gibberish like the Austrian statement on Action that you quote the rational thing to do is to look no further and dismiss out of hand everything else that the Austrians say.

    ReplyDelete
  16. If you can explain someone else's idea in your own words, such that they agree with your explanation, then you understand that idea well enough to refute it. If not, then generally speaking, you don't. Many, though not all, arguments, get stuck at the stage of not understanding the opponents points. The smarter and more empathetic you are, the more likely it is you can get past this stage at which point the disagreement turns on different sets of facts believed or different fundamental vales. Accusations of not being able to pass such a test are essentially accusations of either low intelligence, lack of empathy, or laziness. Noah, you are making way too much out of this.

    ReplyDelete
    Replies
    1. "Accusations of not being able to pass such a test are essentially accusations of either low intelligence, lack of empathy, or laziness."

      Others above have said the same, but I fail to see why these are the only three options. To me, the most obvious reason a person might not understand a viewpoint, such that they would be accused of not being able to pass an ITT, is that they are simply not sufficiently familiar with it to judge it appropriately.

      Delete