I've started to think about the abstract structure of the observation selection problem (OSP) with an eye towards the fine-tuning argument. The way the OSP works, at least in the simple cases, is that some claim is provided that undercuts one's inference from data to hypothesis. For instance, if one's data is that 75% of the polled group G report voting for the republican candidate then the inference to probably the republican candidate will win may be undercut by adding the claim that G is not a representative sample of the entire voting population (G is comprised of 80% republicans). The plausibility of this undercutting move seems to rely on some sort of sensitivity requirement: if the hypothesis were false then the data would indicate that. So far, so good. But when we come to the inference that the fine-tuning data is very surprising on a chance hypothesis, we are told that we should not find it surprising because of the OSP, and this because this is the only kind of universe we could observe. Let e = the fine-tuning data and N= the chance hypothesis, the claim that P(e/N) is very low should be undermined by some claim k that we add to N. What's that k? Suppose k = the only data we could possess is e. How is this supposed to undermine the claim that P(e/N) is very low? It seems like in this case we can still reason from the properties of the chance hypothesis to the improbability of e. It's true, of course, that something along the lines of k is true, but that doesn't affect the predictive power (or, in this case, the lack thereof) of the chance-hypothesis. Maybe, however, there's a better way to cast the OSP. Thoughts?
What is the observation selection problem?
2 TrackBacks
My apologies for the delay, but I am pleased to announce Philosophy Carnival XV, covering the best posts from around the philosophy blogosphere for the month of June (and a few from late May). This month’s carnival neatly divides into three cat... Read More
The 15th Philosophers' Carvnival is at The Buckingham Inquirer. Ted's post on fine-tuning represents our blog.... Read More
I don't see how the OSP comes up here. In the election case, there's real data that makes the results of the poll inaccurate. You've polled a group, and their answers aren't indicative of everyone that could have been polled. But the design argument isn't like that. It's true that the constants we're looking at might only hold in this universe, and if there are other universes they might have other constants. It's also true that you can come up with two alternative explanations of the constants' being in such a seemingly unlikely place. One is that they were designed. Another is that there are so ridiculously many universes that one of them had to be this way.
It seems funny to me to think of this as observation selection, though. The closest way it's like that is that we're not there to observe the other universes, but we're also not in places where unicorns, ghosts, and goblins manifest themselves, and no one claims observation selection bias in arguments against such things. Observation selection bias seems to me to apply only when you know what the bias is and you know that some set of facts has been ignored. We don't know any such thing in this case.
You might argue via Occam's Razor that these gajillions of other universes are the sort of thing we already believe in (though you have to admit that a designer is only one more thing as opposed to bazillions more, and you also need to propose a mechanism for generating universes, and we don't already believe in that). I don't think we can claim OSP as an argument for preferring lots of universes to a designer unless you already have an argument for lots of universes.
The OSP manifests itself in the election problem because you needn't have data that rebuts the data you in fact possess. All you need is that the method of data acquisition is strongly biased in favor of the republican candidate in such a way that the following counterfactual doesn't hold: if the candidate were to lose then your data would indicate that.
I agree that there's something funny with the OSP applied to the fine-tuning argument, but people offer it as a reason for rejecting the fine-tuning argument. For example, Steven Unwin in a recent book, _The Probability of God_, offers the OSP as the primary reason for rejecting the fine-tuning argument. Nick Bostrom, at www.anthropic-principle.com, seems to think that there are lessons to be learned about the OSP and the fine-tuning argument. Every now and then, I think I get a glimpse of a significant issue here, but as soon as it comes, it goes. When think about the overall structure of the OSP for the fine-tuning argument it either seems inapplicable or subject to counterexample (for instance Leslie's marksmen case or Swinburne's card shuffling case).
If there were invisible goblins living under bridges, would our data indicate it? If there were genuine ghosts haunting old houses, would our data indicate it? You can't apply this sort of argument across the board. There have got to be other considerations. The OSP in the political case makes that kind of poll pretty bad, but it doesn't make it illegitimate to claim that there aren't invisible goblins just because I haven't run into any. The mere falsity of the relevant counterfactual doesn't seem to me to be what's making OSP cases bad if there are analogous cases that are just as good with the same sort of false counterfactual.
Is Leslie's marksmen case the following? If a bunch of mean looking guys told to kill you in a point-blank firing squad all end up missing, you should think it rigged rather than thinking there are zillions of firing squads. If so, I've been trying to find out where it originally came from. I learned it from Roger White, but I lost the original source.
The marksmen case occurs in Leslie "Anthropic Principle, World Ensemble, Design" APQ 18 (1982): p. 150. He just briefly mentions it at the close of his article.
You're right about relying too much on the mere falsity of the relevant counterfactual. Dretske's painted mule case makes a similar point. I was wondering whether the OSP would turn out to rely on such a mistaken principle (or at least in its purported application to the fine-tuning argument).
Here's a reducto absurdum of the Anthoporic principle.
- NASA is cheap on funding lately so they cut safety costs, their new space craft has only a one in 1000000000000000000000000000000000000 chance of surviving the atomsphere. They force some poor unfourtnate to get aboard, he survives the descent. "Phew, I survived, what a coincidence." Then he thinks "No its not! If I wasn't here I wouldn't be here to observe it, therefore my survival was not vastly improbable.
The fact that something happens does not make it improbable. It makes it actual. Probability has to do with how likely something is given certain background conditions. Actuality has to do with whether it happens anyway. Lots of improbable things happen. What's interesting is when improbable things with respect to certain background conditions happen. That NASA case would be one. If something had that small a chance of surviving for all we know, and it survived, then it would indeed be grounds for figuring out what went wrong in the calculations and why it did survive. It's thus parallel to the design argument as you say, but it goes the other way. This is exactly the sort of case when we do want an explanation, so if you want to push it as parallel to the design argument then the cosmological constants also cry out for explanation.
I think Eliot Sober has an article (in 'The Blackwell Companion to the Philosophy of Religion') where he argues that there's an error in how most people interpret the firing squad.
On a slightly related matter, there's the Ikeda-Jeffries argument that fine tuning implies naturalism
http://quasar.as.utexas.edu/anthropic.html
which seems to hinge on the supporter of fine-tuning _also_ being a supporter of Intelligent Design (which may be more common in the US than it is in the UK), though it argues that supporting the former but not the latter leads to equal weight being given to design/naturalism. (There also seem to be some theologically dubious points being made therein, but that may just be me). I'm not enriely sure what one should make of this...
Thanks for the references. I’m aware of Sober’s discussion, but I haven’t found it persuasive. I’ll read the Ikeda-Jeffries article with interest. If, however, the argument turns on the assumption you mention then it’s mistaken. Trent Dougherty and I have a paper under review that argues that the FTA and BDA are clearly separable and, in fact, given some assumptions in conflict with each other. I’d be happy to send you a copy, if you’re interested.
Yes, that'd be great, thanks (though I'll come clean here and admit that I'm a physiscist, not a philosopher, so some of it may be a bit beyond me).
I think I've also slightly misrepresented the Ikeda-Jeffries argument in my summary - their position is that _if_ we have BD and FT, then a designer is improbable, however, with FT alone, we have the likelyhood of design and the likelyhood of chance being equal (and that hence the FTA fails).
I'd be interested in hearing what you make of it.