Comments on: Reasonable Disagreement http://www.overcomingbias.com/2006/12/reasonable_disa.html Overcoming Bias is economist Robin Hanson’s blog, on honesty, signaling, disagreement, forecasting, and the far future. Sun, 12 Feb 2012 03:06:37 +0000 hourly 1 http://wordpress.org/?v=3.1.3 By: Robin Hanson http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423998 Robin Hanson Fri, 15 Dec 2006 15:56:20 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423998 Nick, it seems the issue we need to consider more to further explore this topic is the appropriateness of Bayesian-like analysis. But as it is a framework intended to account for a wide range of issues in inference, we should judge it overall in terms of all of the intuitions it may or may not conflict with, relative to other possible frameworks of analysis. Since we expect some of our intuitions to be in error, finding a few conflicts with intuition should not discourage us from embracing Bayesian analysis. I invite you to take the first shot by posting sometime on what you see as the most serious problems with the Bayesian approach. Nick, it seems the issue we need to consider more to further explore this topic is the appropriateness of Bayesian-like analysis. But as it is a framework intended to account for a wide range of issues in inference, we should judge it overall in terms of all of the intuitions it may or may not conflict with, relative to other possible frameworks of analysis. Since we expect some of our intuitions to be in error, finding a few conflicts with intuition should not discourage us from embracing Bayesian analysis. I invite you to take the first shot by posting sometime on what you see as the most serious problems with the Bayesian approach.

]]> By: Nicholas Shackel http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423997 Nicholas Shackel Mon, 11 Dec 2006 20:40:35 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423997 I see this is in effect being further addressed in new posts, so I’ll just answer the last two points and then leave it. Eliezer (apologies for mis-spelling your name in the last post): The answer to your challenge is in the original post, where I offered precisely such an argument. Secondly, I think you are mistaken about where the burden of proof lies. No one is disputing that sometimes there are unreasonable disagreements, which is all that your examples go to show. But you are simply assuming that Bayesianism is true. My point is that it is no less reasonable, and perhaps more reasonable, to start from the premiss that people do reasonably disagree (indeed, some would argue that we are morally required to accept that premiss), and if Bayesianism conflicts with that, so much the worse for Bayesianism. Robin: It is not that I am uninterested or unsympathetic to the formal results, but I am bringing into view ways in which it might be argued that the formal model seems to give the wrong answer. Guy brought out the point about idealisation at the end of enquiry versus our situation. There is a lot to be discussed about whether and when idealisation is a clarification rather than obscuration of philosophical issues. I have been pressing on a different point, namely the the requirement that reasons for a belief should have content that is relevant to the truth of the content of the belief. I have drawn your attention to a specific argument about endurance versus perdurance which turns on the problem of temporary intrinsics. Attending to that argument shows why Lewis thinks temporary intrinsics means perdurance is true whilst Van Inwagen does not, and the reasons that Lewis adduces are other metaphysical doctrines, in particular, doctrines about what it is to be an intrinsic property. The belief in their disagreement has no content that bears. It is quite irrelevant. If you add it as a premiss to either of their arguments it sits as an idle cog. It can neither justify nor defeat any of the reasons they adduce in this dispute. So I have given an example of the way in which disagreement, at least prima facie, has no rational significance, and your reply is to say, well yes, but in my formal model with impossible possible worlds it does. Fine, say I, so much the worse for your model! I see this is in effect being further addressed in new posts, so I’ll just answer the last two points and then leave it.
Eliezer (apologies for mis-spelling your name in the last post): The answer to your challenge is in the original post, where I offered precisely such an argument. Secondly, I think you are mistaken about where the burden of proof lies. No one is disputing that sometimes there are unreasonable disagreements, which is all that your examples go to show. But you are simply assuming that Bayesianism is true. My point is that it is no less reasonable, and perhaps more reasonable, to start from the premiss that people do reasonably disagree (indeed, some would argue that we are morally required to accept that premiss), and if Bayesianism conflicts with that, so much the worse for Bayesianism.
Robin: It is not that I am uninterested or unsympathetic to the formal results, but I am bringing into view ways in which it might be argued that the formal model seems to give the wrong answer. Guy brought out the point about idealisation at the end of enquiry versus our situation. There is a lot to be discussed about whether and when idealisation is a clarification rather than obscuration of philosophical issues. I have been pressing on a different point, namely the the requirement that reasons for a belief should have content that is relevant to the truth of the content of the belief. I have drawn your attention to a specific argument about endurance versus perdurance which turns on the problem of temporary intrinsics. Attending to that argument shows why Lewis thinks temporary intrinsics means perdurance is true whilst Van Inwagen does not, and the reasons that Lewis adduces are other metaphysical doctrines, in particular, doctrines about what it is to be an intrinsic property. The belief in their disagreement has no content that bears. It is quite irrelevant. If you add it as a premiss to either of their arguments it sits as an idle cog. It can neither justify nor defeat any of the reasons they adduce in this dispute. So I have given an example of the way in which disagreement, at least prima facie, has no rational significance, and your reply is to say, well yes, but in my formal model with impossible possible worlds it does. Fine, say I, so much the worse for your model!

]]> By: Robin Hanson http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423996 Robin Hanson Sat, 09 Dec 2006 13:31:52 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423996 Nick S, I presume we agree that there is a true answer to the question of the "persistence of identity through time," as otherwise there is no point in disagreeing about that topic. Even if you thought that this claim was either true in all possible worlds, or false in all possible worlds, in the usual sense of a self-consistent "possible world," we can invoke the concept of an "impossible possible world" which need not be self-consistent. Using this concept of an impossible possible world, we can formally express the idea that as you proceed with error-prone analysis, you may mistakenly believe a claim that is not consistent, but that as you conduct more analysis you will make fewer such mistakes. In such a framework the fact that someone else proceeding similarly disagrees with you is a powerful relevant clue about the errors in your analysis, just as ordinary disagreement is a powerful clue about ordinary inference. Nick S, I presume we agree that there is a true answer to the question of the “persistence of identity through time,” as otherwise there is no point in disagreeing about that topic. Even if you thought that this claim was either true in all possible worlds, or false in all possible worlds, in the usual sense of a self-consistent “possible world,” we can invoke the concept of an “impossible possible world” which need not be self-consistent.

Using this concept of an impossible possible world, we can formally express the idea that as you proceed with error-prone analysis, you may mistakenly believe a claim that is not consistent, but that as you conduct more analysis you will make fewer such mistakes. In such a framework the fact that someone else proceeding similarly disagrees with you is a powerful relevant clue about the errors in your analysis, just as ordinary disagreement is a powerful clue about ordinary inference.

]]> By: Eliezer Yudkowsky http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423995 Eliezer Yudkowsky Sat, 09 Dec 2006 01:26:03 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423995 Nicholas, just to amplify on that, what I'm challenging you to do is justify the step from "philosophical disagreement is a fact of our lives" to "reasonable philosophical disagreement is a fact of our lives". That's the key step, and you can't just say you observed it - one wishes to know what justification for the term "reasonable" is strong enough to override normative Bayes. The fact that it *feels* very reasonable to you is not strong enough to override Bayes, because conjunction fallacies, and availability biases, and nine-tenths of the known fallacies in the lexicon, *feel* reasonable. Nicholas, just to amplify on that, what I’m challenging you to do is justify the step from “philosophical disagreement is a fact of our lives” to “reasonable philosophical disagreement is a fact of our lives”. That’s the key step, and you can’t just say you observed it – one wishes to know what justification for the term “reasonable” is strong enough to override normative Bayes. The fact that it *feels* very reasonable to you is not strong enough to override Bayes, because conjunction fallacies, and availability biases, and nine-tenths of the known fallacies in the lexicon, *feel* reasonable.

]]> By: Eliezer Yudkowsky http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423994 Eliezer Yudkowsky Sat, 09 Dec 2006 01:16:09 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423994 Heh. Nicholas, whilst I can see your inexorable chain of logic, one much wishes to know what possible desideratum - nay, not even the word of God spoken from out of the clouds - could possibly support "2. People do reasonably disagree" when no less a desideratum than NORMATIVE BAYESIANISM says otherwise. Heh. Nicholas, whilst I can see your inexorable chain of logic, one much wishes to know what possible desideratum – nay, not even the word of God spoken from out of the clouds – could possibly support “2. People do reasonably disagree” when no less a desideratum than NORMATIVE BAYESIANISM says otherwise.

]]> By: Nicholas Shackel http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423993 Nicholas Shackel Sat, 09 Dec 2006 00:45:09 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423993 Eleizer: the argument would be 1. Normative Bayesianism implies that people cannot reasonably disagree. 2. People do reasonably disagree. 3. What is actual is possible. 4. Therefore people can reasonably disagree. (2, 3) 5. Therefore normative Bayesianism is false. (1, 4 modus tollens) Robin: of course I grant that *any* kind of disagreement (not just ordinary ones) *might* be unreasonable. That’s not the issue. The point I was making was that in the case of beliefs for which we can give a causal truth tracking account, we can at least give some kind of account of why someone disagreeing with me could be *evidence* for me, and so count among possessed reasons I have that bear on the belief in question— whereas that is not the case with respect to a priori propositions in dispute between epistemic peers. I reiterate my earlier remark: the only way Van Inwagen can take cognisance of Lewis’s disagreement is by attending to what Lewis says about why he disagrees (etc.). The relevant sense of ‘can’ is rational epistemic possibility. You said this was clearly false, but it’s not. I did not say that the only way anyone could ever take cognisance was by attending to what is said, but I said it of *Van Inwagen* in relation to *Lewis* with respect to the *metaphysical issue of persistence of identity through time*. The crucial point here is that there comes a point at which other people’s reports of their disagreement may have no rational significance. They have to say something which bears on the issue in virtue of the relevance of the content of what they say to the issue. That there are other occasions in which I might, as you put it, ‘change my opinion based on the fact of your conclusion’ because ‘All I need is a belief that your conclusions tend to be well-founded.’ is irrelevant. Eleizer: the argument would be
1. Normative Bayesianism implies that people cannot reasonably disagree.
2. People do reasonably disagree.
3. What is actual is possible.
4. Therefore people can reasonably disagree. (2, 3)
5. Therefore normative Bayesianism is false. (1, 4 modus tollens)

Robin: of course I grant that *any* kind of disagreement (not just ordinary ones) *might* be unreasonable. That’s not the issue. The point I was making was that in the case of beliefs for which we can give a causal truth tracking account, we can at least give some kind of account of why someone disagreeing with me could be *evidence* for me, and so count among possessed reasons I have that bear on the belief in question— whereas that is not the case with respect to a priori propositions in dispute between epistemic peers. I reiterate my earlier remark: the only way Van Inwagen can take cognisance of Lewis’s disagreement is by attending to what Lewis says about why he disagrees (etc.). The relevant sense of ‘can’ is rational epistemic possibility. You said this was clearly false, but it’s not. I did not say that the only way anyone could ever take cognisance was by attending to what is said, but I said it of *Van Inwagen* in relation to *Lewis* with respect to the *metaphysical issue of persistence of identity through time*. The crucial point here is that there comes a point at which other people’s reports of their disagreement may have no rational significance. They have to say something which bears on the issue in virtue of the relevance of the content of what they say to the issue. That there are other occasions in which I might, as you put it, ‘change my opinion based on the fact of your conclusion’ because ‘All I need is a belief that your conclusions tend to be well-founded.’ is irrelevant.

]]> By: Robin Hanson http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423992 Robin Hanson Fri, 08 Dec 2006 01:00:49 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423992 Nick S, you seem willing to grant that ordinary disagreements may be unreasonable, but want to carve out an exception for disagreements about "philosophical doctrines and other a priori truth." I admit I've always had trouble understanding some of these distinctions. For most topics there is a sensible distinction between people with different basic evidence and people with different analysis of that evidence. The simplest Bayesian formulation does not allow for different analysis, but it can be straightforwardly generalized to allow for different analysis, and then the standard disagreement results remain. If the concept of a priori truths is fundamentally different from different analysis of the same evidence, then the question is whether we can make sense of "impossible possible worlds" in which these apriori questions have different answers. If we can, then the usual Bayesian results will hold. Nick S, you seem willing to grant that ordinary disagreements may be unreasonable, but want to carve out an exception for disagreements about “philosophical doctrines and other a priori truth.” I admit I’ve always had trouble understanding some of these distinctions.

For most topics there is a sensible distinction between people with different basic evidence and people with different analysis of that evidence. The simplest Bayesian formulation does not allow for different analysis, but it can be straightforwardly generalized to allow for different analysis, and then the standard disagreement results remain.

If the concept of a priori truths is fundamentally different from different analysis of the same evidence, then the question is whether we can make sense of “impossible possible worlds” in which these apriori questions have different answers. If we can, then the usual Bayesian results will hold.

]]> By: Eliezer Yudkowsky http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423991 Eliezer Yudkowsky Fri, 08 Dec 2006 01:00:03 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423991 "Furthermore, someone could argue that since reasonable disagreement is a fact of our lives, the very strong result about disagreement is a reductio of Bayesianism as a model for belief." Why isn't this pure naturalistic fallacy? Plenty of gross reasoning errors are facts of modern-day human life. If you mean that it's a reductio of Bayesianism as a *descriptive* model of human belief, no one sane uses it that way. “Furthermore, someone could argue that since reasonable disagreement is a fact of our lives, the very strong result about disagreement is a reductio of Bayesianism as a model for belief.”

Why isn’t this pure naturalistic fallacy? Plenty of gross reasoning errors are facts of modern-day human life. If you mean that it’s a reductio of Bayesianism as a *descriptive* model of human belief, no one sane uses it that way.

]]> By: Nicholas Shackel http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423990 Nicholas Shackel Thu, 07 Dec 2006 22:49:17 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423990 Robin: The formal results are certainly of considerable interest, but Bayesianism is a mathematical model of belief, and there naturally arises questions about the extent of its applicability. Of its nature Bayesianism dispenses with a great many distinctions between kinds of beliefs that have philosophical significance. The claims you are making about the weight of evidence given by disagreement are more appealing when applied to beliefs to which a notion of causal truth tracking makes sense. But when it comes to philosophical doctrines and other a priori truths, it makes less sense. Furthermore, someone could argue that since reasonable disagreement is a fact of our lives, the very strong result about disagreement is a reductio of Bayesianism as a model for belief. Robin: The formal results are certainly of considerable interest, but Bayesianism is a mathematical model of belief, and there naturally arises questions about the extent of its applicability. Of its nature Bayesianism dispenses with a great many distinctions between kinds of beliefs that have philosophical significance. The claims you are making about the weight of evidence given by disagreement are more appealing when applied to beliefs to which a notion of causal truth tracking makes sense. But when it comes to philosophical doctrines and other a priori truths, it makes less sense. Furthermore, someone could argue that since reasonable disagreement is a fact of our lives, the very strong result about disagreement is a reductio of Bayesianism as a model for belief.

]]> By: Robin Hanson http://www.overcomingbias.com/2006/12/reasonable_disa.html#comment-423989 Robin Hanson Thu, 07 Dec 2006 20:28:27 +0000 http://prod.ob.trike.com.au/2006/12/reasonable-disagreement.html#comment-423989 Guy, I fear I am reaching the limits of my ability to converse seriously with philosophers while trying to "pick up" their language via random readings. At least that interpretation seems preferable than what you seem to be saying, namely that we are each justified in giving ourselves the benefit of the doubt that our hunches are insight, in a way that we are not justified regarding others. In moral philosophy giving yourself more benefit of the doubt about the morality of your actions is considered a self-favoring bias. Why not here also? You and Nick S. both seem to say that philosophers consider disagreement to be only weak evidence, whereas I say that in detailed Bayesian models it is very strong evidence. Could it be that philosophers are not sufficiently aware of this result? Or would that still not be convincing to them? Guy, I fear I am reaching the limits of my ability to converse seriously with philosophers while trying to “pick up” their language via random readings. At least that interpretation seems preferable than what you seem to be saying, namely that we are each justified in giving ourselves the benefit of the doubt that our hunches are insight, in a way that we are not justified regarding others. In moral philosophy giving yourself more benefit of the doubt about the morality of your actions is considered a self-favoring bias. Why not here also?

You and Nick S. both seem to say that philosophers consider disagreement to be only weak evidence, whereas I say that in detailed Bayesian models it is very strong evidence. Could it be that philosophers are not sufficiently aware of this result? Or would that still not be convincing to them?

]]>