This week’s New Scientist has an article by Jim Baggott and Daniel Cossins entitled Beyond Experiment: Why the scientific method may be old hat, which deals with the recent controversy over attempts to excuse the failure of string theory by invoking the multiverse. The article (unfortunately behind a paywall) does a good job of describing the nature of the controversy: what do you do when it becomes clear your theory can’t be tested? Do you follow the conventional scientific norms, give up on it and work on something else, or do you try and find some kind of excuse, even if it means abandoning those norms?
Much of the article deals with the issues raised at the recent Munich conference (discussed here). Two of those quoted (Dawid and Gross) are not multiverse partisans, instead argue that the motivations that got people interested in string unification more than 30 years ago are good enough to justify indefinitely pursuing the theory, no matter how bad things look for prospects of connection to experiment. On the other hand:
Their enthusiasm is far from universal, and some physicists are downright alarmed. Woit warns that the need for empirical vindication could be pushed so far into the background as to be invisible. Carlo Rovelli, a theorist at the University of Aix-Marseille in France, believes that this scenario has already come to pass. Rovelli … argues that the last thing we need is a system that legitimises failed theories. “A theory is interesting when it teaches us something new about the real world,” he says. “Not when it becomes a house of cards that delivers nothing but university positions.”
On the question of the string theory multiverse as science, those gathered at the Munich conference were pretty uniformly hostile. As a proponent of this, the article quotes only one person, who wasn’t there:
Sean Carroll, a theorist at the Caltech Institute of Technology at Pasadena and a leading advocate of the multiverse, insists that if anyone is being unscientific, it is those physicists who seek to enforce outmoded philosophical principles and impossibly high standards. “People support these theories because they offer the best chance of providing a useful account of the data we actually do collect here in our universe.”
I’m not sure how the string theory multiverse provides an account of data we have collected that is “useful”, except in the sense of “useful to those who don’t want to give up on string theory.”
Carroll has explained his views in more detail here, arguing that falsifiability is an idea that needs to be retired, to be replaced by “empiricism”. “Empiricism” seems to mean “ability to account for the data”, with “the multiverse did it” an acceptable way to account for data, even if not falsifiable. He’ll be giving a talk on this at the American Astronomical Society meeting in San Diego this summer, with abstract:
A number of theories in contemporary physics and cosmology place an emphasis on features that are hard, and arguably impossible, to test. These include the cosmological multiverse as well as some approaches to quantum gravity. Worries have been raised that these models attempt to sidestep the purportedly crucial principle of falsifiability. Proponents of these theories sometimes suggest that we are seeing a new approach to science, while opponents fear that we are abandoning science altogether. I will argue that in fact these theories are straightforwardly scientific and can be evaluated in absolutely conventional ways, based on empiricism, abduction (inference to the best explanation), and Bayesian reasoning. The integrity of science remains intact.
Carroll’s argument seems to be that the conventional understanding of how science works that we teach students and use to explain the power of science has always been wrong. Falsifiability by experiment isn’t necessary, instead, what is the “absolutely conventional” way to do science is “empiricism, abduction (inference to the best explanation), and Bayesian reasoning”. I’d never heard of abduction as a basis of science before. If you believe Wikipedia, this goes back to Charles Sanders Peirce, whose view in later years was:
Abduction is guessing. It is “very little hampered” by rules of logic. Even a well-prepared mind’s individual guesses are more frequently wrong than right. But the success of our guesses far exceeds that of random luck and seems born of attunement to nature by instinct (some speak of intuition in such contexts).
As for “Bayesian reasoning”, I would have thought that Polchinski’s Bayesian calculation of an “94% chance” of a multiverse would have conclusively shown the absurdity of that.
Unifier,
I don’t dismiss the multiverse as a possible solution to the CC problem. Maybe that’s the way the world is. The problem is that “the multiverse did it” is a scientifically empty argument. The worst part of what is going on here is that this kind of empty argument is being made in order to justify refusal to admit failure.
But, ignoring that, what I mean when I say it’s empty is this: the string theory landscape does not allow you to calculate a real probability distribution of CCs, so what people do is choose a flat probability distribution over the anthropically allowed range, and then congratulate themselves that the observed CC is not probabilistically extremely unlikely (although how unlikely it is depends on all sorts of assumptions). I agree with Arkani-Hamed that there’s no good idea about what sets the CC, so I propose the Peter Woit CC theory, which is “I have no frigging idea why the CC is one thing rather than the other”. What are the observational predictions of the Peter Woit CC theory? Well, since there’s no reason for the CC to be one thing or another, my observational prediction is that it will be in the anthropic range (this is a tautology), with a flat probability distribution. Funny, but this is exactly the same prediction that you get from the string theory landscape. Now do you see why I say “the multiverse did it” explanation of the CC is empty of conventional scientific content?
Peter , OT. But from a recent preprint by ellis, it seems Lev Okun is no more. RIP.
In the rarefied atmosphere of Theoretical Physics I suspect that the elite 2% sometime forget that when Bayesian statistics or arguments on the principle of falsification are invoked, then this is an area where many can meet them on completely equal terms.
There is no intellectual argument that is so esoteric that it can transcend the more prosaic principles that these simple concepts demand.
So when an argument looks circular to the “ordinary guy” there is a pretty good chance that it is. You have to be able to justify yourself in ordinary terms. If you cannot then it is certain that your arguments are too weak, not too complex.
Pingback: Abduction, Deduction, Induction, Analogy, Inquiry : 7 | Inquiry Into Inquiry
“Carroll’s argument seems to be that the conventional understanding of how science works that we teach students and use to explain the power of science has always been wrong. ”
Well, I think that philosophers would argue that conventional descriptions of science taught to students in science courses vastly oversimplify how science works, or should work. A lot of science is not really about trying to falsify theories.
Jr,
I don’t recall as a student being taught any rigid notion of what was science and what wasn’t, other than the importance of being able to find out if your were wrong by doing an experiment. This is what Carroll is now campaigning against, and the goal is to have multiverse pseudo-science written into the textbooks. Do you really think that teaching students about the multiverse and explaining to them that this is “absolutely conventional” science is a good idea?
Pingback: Abduction, Deduction, Induction, Analogy, Inquiry : 8 | Inquiry Into Inquiry
Pingback: Peter Woit vs. Sean Carroll on bent rules | Uncommon Descent
No, I do not think teaching the multiverse theory seems like a good idea and it does strike me as seriously misguided. But independently of that, it may be a good idea to make the textbooks at little more nuanced on the topic of falsification. I do not want to go into to much discussion on the point since anything I could say has been said much better by philosophers already.
Re: Scott Church
Names are not important of course, except for the purpose of communication. The important thing is for us to distinguish hypothesis formation from hypothesis evaluation. Now, there happens to be a long tradition of using the word abduction to distinguish that former, most incipient stage of inquiry and I think it serves communication to preserve that tradition.
Concepts, hypotheses, and theories have to be formed, logically speaking, before they can be evaluated. In complex inquiries extending over long periods of time, formation, evaluation, and re-formation will of course proceed in cascades of parallel and series operations, but the analytic distinction between elements and mixtures is still worth its salt.
The role of ab-, de-, in-duction in the cycle of inquiry is discussed a bit further in the following article:
☞ InterSciWiki • Inquiry
Your mention of “Bayesian reasoning” in reference to Polchinski’s Bayesian calculation of an “94% chance” is worrisome.
There are books on gravitational waves or astrostatistics that include discussions on the so-called Bayesian and frequentist “viewpoints.” It is one thing to say which “mathematical formula” is applicable in which
situation. It is another thing to discuss philosophical issues which generate potential source of esoteric matter and confusion in physics teaching. It think it will be a great service to science and students of science if authorities sit together and set the meaning of the phrase “Bayesian” in gravitational physics and drop it if it is unnecessary.
Would “testable prediction” be the same as “falsifiable”? I think next to everybody can agree that a theory with no testable predictions is pure speculation. Now, what would be the distance between testability and falsifiability?
LM,
I don’t think the issue of how you test a scientific idea can be summarized in any one word, whether its “testable” or “falsifiable”. This is a complex question, and you have to look at exactly what is being claimed as a “test”.
The relevance of falsifiability is that there’s obviously something suspicious about a “test” that you can’t fail. But also obviously, that can be meaningless. String theorists are known to claim string theory is falsifiable because it is based on quantum mechanics and quantum mechanics is falsifiable, but claiming “my theory is falsifiable, because it says that “absurd thing X won’t happen” is obviously problematic. Any time spent thinking about this leads to the conclusion that this is a complicated issue, with bogus “tests” easy to come up with, meaningful ones not hard to recognize as such. My claim is just that string theory is not capable of giving a meaningful one, and to evaluate that claim you need to look at what supposed string theory “tests” are.
I agree that the distinction doesn’t answer all questions but the point is that no one would argue that to be science a theory must be tested. So it’s a matter of test. There has to be a definite and defined test, it has to be reproduceable, and yield definite results. It has to be new – in the sense that a new theory cannot be based exclusively on previous tests (it wouldn’t add any new knowledge – if I’m not mistaken, the Occam’s razor). And as long as theory aims to describe real world the test has to be empirical i.e based on observable reality. So wouldn’t these discussions better served defining testability – the notion that may indeed be evolved and extended with time?
LM,
The problem is that of defining what a good “test” is. You’ve given some criteria, but they don’t address the relevant problems that come up. For instance: some physicists like to claim that their multiverse theory is testable, because it is conceivable within their framework that certain distinctive patterns would appear in the CMB. So, there’s a test: does this pattern appear or not in the CMB? If it did, that would definitely be evidence for their theory. The problem with non-falsifiable tests appears though: if the pattern doesn’t appear, that isn’t evidence against their theory. If you look into their theory, you see that it’s so poorly defined that it says nothing at all about the likelihood of such patterns appearing. There’s obviously a problem here with claiming “my theory is testable”, when there is no possibility of failure. This is the main issue with “tests” of string theory or the multiverse.
Peter, I understand your concerns but maybe we are living in a different world today? For example, the Net is full of stuff of all kinds from obvious garbage to brilliant ideas and insights. It’s all there, nobody can possibly falsify it all because new stuff just comes out faster than any capacity for policing.
Where’s the answer? How do we deal with it? With openness and collective memory (e.g. rating). As long as test is clearly declared, and it fits the clear and precise scientific criteria a theory must declare its results, openly for everybody to see and make their own judgement. A theory with a hundred unconfirmed tests (and these need to be new tests, specific to purported new knowledge) would still have a rating of zero, and no talks and speculation could change that for anybody with a capacity for rational judgement. Could it be enough – I don’t know just thinking of some way forward. And for already mentioned reason, I wouldn’t even call an idea, concept, hypothesis a “theory” until it comes up with a set of tests that fit the criteria of science.
Few people would dispute that science needs to be testable in the sense of accounting for observation. That’s not the problem here. The real question is… to what degree do they do so? It’s not enough to come up with a model that explains things (what Sean Carroll and others have dubbed the new “Empiricism”), nor is it enough for it to be “elegant” or “beautiful”… the “my wife is hotter than yours” approach. To be testable in any sense that matters scientifically a theory must account for observation specifically, and uniquely.
General relativity predicted the bending of starlight in the gravitational field of the sun, and the perihelion precession of Mercury. These predictions were scientific because they were specific enough to be tested practically to great accuracy, but more importantly, because they were consequences of GR alone. Newtonian physics did not predict either, and could only account for them with ad-hoc contrivances ripe for Occam’s razor. It is true that for a while Mercury’s perihelion precession could be explained by the gravitation of an unseen planet called “Vulcan”, but once the solar system had been mapped thoroughly that explanation went away. There’s also the fact that the solar system actually could be mapped accurately enough to rule out the Vulcan hypothesis.
String theory and the multiverse are only “testable” in the weak sense of being able to loosely (very loosely) account for observation… the way the Vulcan hypothesis once did. What makes them unscientific is that their “predictions” are neither specific enough, nor unique enough to render them potentially favorable over their competitors by any foreseeable test. The ability to hit a barn at 5 paces with a sawed-off shotgun does not make one a marksman, nor does it earn one a sharpshooter medal if a hundred other folks can do the same thing blindfolded.
Pingback: Abduction, Deduction, Induction, Analogy, Inquiry : 14 | Inquiry Into Inquiry
Pingback: Links 3/17/16 | Mike the Mad Biologist