August 21, 2004

Worlds Without End, So It's Not Like Yours Is Anything Special, Amen

Lee Smolin gives a very nice over-view of his theory of "cosmological natural selection", and tears into the anthropic principle, in a recent pre-print:

Lee Smolin, "Scientific alternatives to the anthropic principle", hep-th/0407213
Abstract: It is explained in detail why the Anthropic Principle (AP) cannot yield any falsifiable predictions, and therefore cannot be a part of science. Cases which have been claimed as successful predictions from the AP are shown to be not that. Either they are uncontroversial applications of selection principles in one universe (as in Dicke's argument), or the predictions made do not actually logically depend on any assumption about life or intelligence, but instead depend only on arguments from observed facts (as in the case of arguments by Hoyle and Weinberg). The Principle of Mediocrity is also examined and shown to be unreliable, as arguments for factually true conclusions can easily be modified to lead to false conclusions by reasonable changes in the specification of the ensemble in which we are assumed to be typical.
We show however that it is still possible to make falsifiable predictions from theories of multiverses, if the ensemble predicted has certain properties specified here. An example of such a falsifiable multiverse theory is cosmological natural selection. It is reviewed here and it is argued that the theory remains unfalsified. But it is very vulnerable to falsification by current observations, which shows that it is a scientific theory.
The consequences for recent discussions of the AP in the context of string theory are discussed.

Quite remarkably, it contains no math, and fairly little jargon; I'd imagine it'd be quite accessible to anyone who'd normally read a popular science book on astronomy. While it makes me want to tackle Smolin's more technical works, this is probably not going to happen, since even pedagogical papers on loop quantum gravity remind me of why I left fundamental theory for statistical physics in the first place --- I'm just not smart enough for the former. (Update, 27 August 2004: I should mention an even more recent Smolin pre-print, "An Invitation to Loop Quantum Gravity", hep-th/0408048.) Having said that, I do have three comments.

First, I think Smolin's absolutely right about the anthropic principle. Take, for instance, the example he gives in section 5.1.3. Fred Hoyle once reasoned that carbon is necessary for life, that carbon must have been formed by stellar nucleosynthesis, and that this reaction could only have proceeded if carbon nuclei had certain properties, which experimentalists then proceeded to show they did have. Smolin fairly schematizes this as follows. (1) X is necessary for life (or intelligence, etc.). (2) X is, as it happens, true. (3) If X is true, and the laws of physics are Y, then Z must also be true. (4) Therefore Z.

We see clearly that the prediction of Z in no way depends on step 1. The argument has the same force if step 1 is removed. To see this ask what we would do were Z found not to be true. Our only option would be to question either Y or the deduction from the presently known laws of physics to Z. We might conclude that the deduction was wrong, for example if we made a mistake in a calculation. If no such option worked, we might have to conclude that the laws of physics might have to be modified. But we would never question 1, because, while a true fact, it plays no role in the logic of the argument leading to the prediction for Z.
Exactly so.

Second, which is a quibble, I think the idea of falsifiability, while still very valuable and basically right-headed, is somewhat more complicated than Smolin, following Popper, allows. This is especially true when statistical hypotheses are at issue. On this point, I recommend the work of Deborah Mayo, especially Error and the Growth of Experimental Knowledge. I don't think this upsets the points Smolin is making, however.

Thirdly, one of Smolin's statements, while correct, needs to be read with caution. Smolin imagines that there is a set of fundamental, dimensionless constants of physics, p, living in some space P, and that universes give birth to new universes, with a child's value of p differing slightly from that of its parent. (He has a particular mechanism in mind, involving quantum-gravitational effects smearing out the singularities of classical black holes.) For a given value of p, there will be a certain average number of descendants per universe, F(p). He then states his selection principle as follows (section 5.2):

If p is changed from the present value in any direction in P the first significant changes in F(p) encountered must be to decrease F(p).
Notice that this does not (as I first thought it did) commit Smolin to the belief that the fitness function has only isolated local maxima. This principle is entirely compatible with having a continuous extended set in P where F(p) has a constant value, and that value is a local maximum. This is what evolutionary biologists would call a "neutral network", where by changing several parameter simultaneously we can keep the fitness constant. (Such neutral networks play an important role in evolutionary dynamics, or so my friends tell me.) In fact, Smolin's selection principle is very carefully worded to allow us to be on a neutral network, provided that's a local maximum.

For more discussion, see Crumb Trail and Not Even Wrong; from the latter it appears that Smolin has rather annoyed some prominent string theorists.

(Via Milan Cirkovic, in e-mail.)


Posted by crshalizi at August 21, 2004 11:26 | permanent link

Three-Toed Sloth:   Hosted, but not endorsed, by the Center for the Study of Complex Systems