Multiverses and falsifiability

Adam Becker’s 2018 book What is Real? The Unfinished Quest for the Meaning of Quantum Physics is an excellent account of the longstanding and intractable controversy over how to interpret quantum mechanics.  One of the main themes of the book is how much the direction of twentieth-century physics was driven by personalities, political factors, career interests, and, not least, unexamined and woolly philosophical assumptions – something philosophers of science like Thomas Kuhn and Paul Feyerabend have shown has always been true of science historically.  The tendency of contemporary physicists, especially, to be both ignorant of and condescending toward philosophy comes in for special criticism.

In at least one case, though, Becker himself is a bit too dismissive of a philosophical line of argument.  Becker discusses how the notion of a multiverse has been defended by many physicists on the basis of several independent considerations, viz. Hugh Everett’s “many worlds” interpretation of quantum mechanics, inflationary cosmology, and string theory.  One objection raised against the notion in any of these versions is that it is unfalsifiable – that is to say, that it generates no predictions that could in principle be proven false by observation and experiment, in which case it is empirically untestable. 

Becker rightly notes that falsifiability, a theme made famous by Karl Popper, is not as straightforward a matter as popular presentations often suppose.  For one thing, a scientific theory is never tested in isolation, because it never generates predictions in isolation.  Rather, its predictions follow from the theory only together with various further assumptions of a theoretical or empirical kind. 

For example, suppose researchers working for a soap company want to determine whether the chemical ingredients in a new product they are developing really will, as they suppose, kill certain kinds of bacteria.  They put samples of the bacteria on a slide, apply the soap, and see what happens.  If the bacteria are not destroyed, has the theory been falsified?  Not necessarily.  For in making and testing the prediction that the soap will kill the bacteria, they are assuming that dead bacteria will have such-and-such an appearance under a microscope, that the slide on which they are put has been cleaned properly (and thus doesn’t have some residue of chemicals that might counteract the effect of the soap they are testing), that the standard theory about how microscopes work is correct, that the particular microscope being used is not malfunctioning, and so on.  And if the test does not come out as predicted, it could be that one of these background assumptions is false, rather than that the soap does not really kill such bacteria.

Of course, there may be very good reasons for judging that none of these assumptions is false, so that the reasonable conclusion to draw is that the soap is not in fact effective against the bacteria.  The point, though (famously emphasized by Pierre Duhem and W. V. Quine), is that testing a scientific claim is not a matter of carrying out a “crucial experiment” that might all by itself either falsify or vindicate the claim.  There is often a certain amount of wiggle room by which a theory might in principle be upheld in the face of apparent counterevidence, even if actually continuing to uphold it is not necessarily reasonable all things considered.

Becker discusses a famous pair of examples from the history of science that illustrate how complicated the matter of falsification actually is.  Newtonian physics was in general spectacularly successful in describing and predicting the observed motions of bodies, but there were exceptions.  The motions of Uranus and Mercury did not conform to the predictions of Newton’s laws, but for a very long time, this did not lead scientists to judge that Newton had been refuted.  After all, the theory worked for most observations, and there was at first nothing better to put in its place.  So, they looked for alternative explanations of the divergence between observation and theory.  In the case of Uranus, it turned out that its motion was being affected by the gravitational pull of another, heretofore unknown planet, Neptune.  That particular problem for Newton’s theory was thus solved.  But the conflict with the observed motion of Mercury resisted any similar solution, and it wasn’t until Einstein’s general theory of relativity appeared – and explained the motion of Mercury along with all the observational evidence that Newton could explain – that Newton’s theory was judged to have been falsified, and Einstein’s adopted in its place.

So far so good.  But Becker then fallaciously draws from these considerations the conclusion that “scientific theories don’t need to be falsifiable” (p. 264) so that:

Claiming, then, that multiverse theories are unscientific because they are unfalsifiable is to reject them simply because they do not live up to an arbitrary standard that no scientific theory of any kind has ever met.  Claiming that no data could ever force the rejection of a multiverse theory is merely stating that a multiverse theory is just like any other theory. (p. 263)

This is not true, and it certainly doesn’t follow.  To understand what is wrong with Becker’s position, we need to draw some distinctions.  First, Popper argued that falsifiability comes in degrees.  Some statements might have empirical consequences independently of any others, other statements might have empirical consequences only in conjunction with further statements, and yet other statements might have no empirical consequences at all.  The first sort of statement would be strongly falsifiable, the second weakly falsifiable, and the third utterly unfalsifiable. 

Now, even if the considerations raised by Becker show that a scientific theory need not be strongly falsifiable, it doesn’t follow that it can be altogether unfalsifiable.  It may, for all Becker has shown, still need to be at least weakly falsifiable.  Now, the critic of multiverse theories might argue that whereas Newton’s physics was weakly falsifiable, multiverse theories are altogether unfalsifiable, so that the parallel Becker wants to draw is bogus.  And in that case, Becker’s response does not suffice to save multiverse theories from the objection in question.  He would have to show, either that multiverse theories are at least weakly falsifiable, or that a scientific theory need not be even weakly falsifiable.  And he does not establish either of these claims.

But even if he were to take the second route and argue that scientific theories needn’t be even weakly falsifiable, there is a further problem, as can be seen by drawing some further distinctions.  For there are different ways in which a statement might be empirically unfalsifiable, some of them unproblematic but some of them problematic.

First, there are statements that are unfalsifiable in the way that mathematical and metaphysical truths can be.  For example, that 2 + 2 = 4 and that the fundamental constituents of reality are substances (as opposed to attributes, say, or events) are, I would argue, not empirically falsifiable.  That is not because they are less certain than empirical claims, but because (as I would also argue) they are more certain.  They are bedrock truths that pertain to any possible reality, to non-empirical immaterial reality no less than to the empirical, material world. 

Second, there are statements that are unfalsifiable in the way that truths of the philosophy of nature can be.  These are claims that, unlike those of the first category, do apply only to empirical reality yet are nevertheless certain.  For example, take the claim that change occurs.  We know this statement empirically, but it is not empirically falsifiable for the simple reason that to falsify something requires having a sequence of experiences (as happens when we set up an experiment, carry it out, and then record the results).  And having a sequence of experiences itself involves change.  To try empirically to falsify the claim that change occurs would thus be self-defeating.  That does not entail that the reality of change is less certain than other empirical truths, but rather that it too is more certain.

Third, there are statements that are unfalsifiable in the way that the most fundamental theses of modern empirical science arguably are.  For example, some have held that the principle of the conservation of energy and the second law of thermodynamics are unfalsifiable.  This is debatable, but it is certainly plausible to maintain that these ideas are so central to modern science’s picture of the universe that they are treated in practice as unfalsifiable, even if they are falsifiable in principle.  The idea is that giving them up would so radically undermine the rest of the modern scientific edifice that, if there ever appeared to be evidence that conflicted with them, scientists would judge that there must be something wrong with the evidence or with other parts of science, rather than that these fundamental principles themselves are false.

Fourth, there are statements that are unfalsifiable in the way that Popper famously took astrology, Marxism, and Freudianism to be.  These are statements that purport to be empirical rather than metaphysical, but are neither parts of the philosophy of nature nor central to the modern scientific picture of the world.  Because they do not fall into the first three categories I’ve just described, the reason they are unfalsifiable is not that they are necessary truths (the way mathematical and metaphysical truths are), or because denying them would be self-defeating (the way denying my example of a truth in the philosophy of nature would be), or because to deny them would take down the whole edifice of science (as the examples in the third category would).  So, they do not have the certainty that truths in these other categories have.  The reason they are unfalsifiable is instead that they make predictions that are too vague or open-ended to be crisply testable.

Now, it is unfalsifiability of this fourth kind that is the most problematic, and that Popper took to be paradigmatic of pseudo-science.  The first two kinds of unfalsifiable statement are, I would argue, unproblematic, and the third kind is at least arguably defensible.  Suppose multiverse theories are indeed unfalsifiable.  Which of these four classes would they fall in?

They don’t fall into the first category, because their description of the world is not true of metaphysical or arithmetical necessity.  That is why even defenders of multiverse theories typically allow that they might be wrong, and at least try to come up with ways of testing such theories empirically.  This would make no sense if the theories had the bedrock status that truths of mathematics and metaphysics are traditionally claimed to have.

They also don’t fall into the second category, because they aren’t fundamental truths about what any possible empirical world must be like, which it would be self-defeating to deny.  Again, even defenders of multiverse theories allow that they might be wrong, and certainly one can doubt such theories without being led into incoherence (by contrast with the attempt to deny the reality of change, which, I would argue, would be incoherent).

Nor do multiverse theories fall into the third category, because they are hardly fundamental to the modern scientific picture of the world in the way that the conservation of energy and the second law of thermodynamics are.  This is obvious just from the fact that they are highly controversial in a way that the fundamental scientific principles mentioned are not.

So, if multiverse theories really are not even weakly falsifiable, but altogether unfalsifiable, it looks like they will fall into the fourth and most problematic class of unfalsifiable theories, alongside astrology, Marxism, and Freudianism.   And in that case, Becker will not have succeeded in defending multiverse theories from the objection in question. 

Successfully to defend them against that objection would require either (a) showing that unfalsifiable statements even of the fourth category are scientifically respectable, (b) showing that multiverse theories are, appearances notwithstanding, unfalsifiable in the way that statements in one of the other three categories are, or (c) showing that multiverse theories are in fact falsifiable and open to empirical testing.  I don’t think that any of these routes is promising, but route (c) would certainly be the way to go if the defender of a multiverse theory wants to convince anyone that such theories are “scientific” in just the same sense that what Newton, Einstein, and the founders of quantum mechanics were up to was scientific.  To do that, however, would not be to sidestep the objection from falsifiability (as Becker wants to do), but precisely to meet the objection head on.