Philosophers of science generally assume that theorizing is boundless in principle but in fact constrained mostly by the phenomena. Theories that violate empirical constraints are weeded out in the scientific process. String theory research, which unifies the four fundamental forces in nature and postulates strings as the smallest physical unit, is different: in its current form, string theory makes no empirical predictions. Still, string theory research has enjoyed great popularity in the community of high-energy physics theoreticians for decades. In fact, string theory is more popular than any other approach that attempts to unify gravity with the principles of quantum mechanics. Given the lack of empirical constraints, this surely seems odd. One way of viewing the situation is it to hold on to the standard view of the relation between theorizing and empirical constraints. The popularity of string theory is then an unhealthy anomaly, which can be explained sociologically as resulting from the influence of certain individuals and herd behavior (for such views, see Penrose Reference Penrose2004; Smolin Reference Smolin2007; Woit Reference Woit2011). Richard Dawid, in his rich book String Theory and the Scientific Method, has set himself a much more demanding task. Instead of castigating string theory research, Dawid asks how we could perhaps extend our idea of the scientific method so as to incorporate string theory research.

Dawid’s book consists of three parts. In the first part, he argues that, even in the absence of any direct empirical tests, theorizing in the context of string theory is not as unconstrained as one might think. In the second part, Dawid suggests that string theory research is not so special after all: it can be viewed as the end point of a historical trend toward an ever greater distance between our observations and our theories. In the third part, he relates his discussion to the realism debate in the philosophy of science. In what follows I try to lead the reader through some of Dawid’s main arguments, focusing on the first and third parts of the book.

Throughout the book, Dawid frames the question of how theorizing gets constrained nonempirically in terms of the underdetermination of theories by the evidence. In the first part of the book, Dawid is concerned only with what he calls ‘scientific underdetermination’, which is better known as transient underdetermination (Stanford Reference Stanford2006): the evidence cannot tell two theories apart in the present but might do so in the future.Footnote ¹ When physicists opt for string theory, they “assess,” as Dawid puts it, whether there is, or rather how likely it is that there is, any transient underdetermination of string theory by the evidence. Dawid’s thought is that physicists are justified in embracing string theory when they can convince themselves that string theory is not underdetermined. Given that string theory has no testable consequences, this way of setting up the problem should strike one as rather awkward. Dawid, however, believes that it is warranted because string theory “conceptually relies on the given empirical data” (50). I am not entirely sure what he means by that, but I think it is clearly too thin for talk about potential underdetermination in any case. For present purposes, I have to let this pass though. Unless otherwise indicated, I will always refer to transient underdetermination.

According to Dawid, physicists possess three ways of “assessing” whether there is any underdetermination in the context of string theory: (i) even after trying very hard, they can find no alternatives to string theory; (ii) string theory has unexpected explanatory consequences; and (iii) the research program that string theory is part of has been very successful in predicting new phenomena in the past. Dawid refers to these arguments in favor of string theory as the no-alternatives argument (NAA), unexpected explanatory coherence (UEA), and the metainductive argument (MIA), respectively. Dawid purports that physicists are actually led by these arguments to believe in string theory. However, he provides little (if any) evidence for that.Footnote ² More importantly, I do not think that any of his arguments are compelling.

First NAA. The failure of theorists to come up with alternatives to string theory, Dawid points out, need not mean that we have not tried hard enough. It does not even have to mean that we are cognitively limited beings. It could also mean that there simply are no alternatives. In the latter case, physicists would have good reasons for believing in string theory despite the lack of evidence. But how can we even make sense of the idea that there are no theoretical alternatives to string theory? The underdetermination of theories by evidence is a logical thesis: no given set of evidence implies any one theory. In order for us to say that the number of theories is highly restricted (or even that there exists only one theory for the evidence), we need further criteria than fit with the evidence. Aesthetic criteria like simplicity and beauty, Dawid rejects explicitly (54–55). Curiously, unification is not even mentioned by Dawid. Instead he believes that MIA can do the constraining (see below). Then again one might wonder what NAA does in the first place. Indeed NAA seems to amount to no more than an assumption that our failure to conceive of alternatives is a good sign for the lack of alternatives or limitation of underdetermination. NAA, on its own, therefore simply begs the question.

Next UEA. In analogy to the rationale offered by some for the importance of novel success in theory confirmation (Worrall Reference Worrall, Gooding, Pinch and Schaffer1989), Dawid suggests that a theory’s explanations of phenomena count more in favor of the theory when the theory was not constructed in order to explain those phenomena (52). It is unexpected, Dawid alleges, that string theory not only unifies gravity and quantum principles within a gauge field theory framework but at the same time also explains the scale hierarchy between the Planck scale and the abundance of black hole entropy. Dawid claims that it is unlikely that there is more than one theory that explains all of those phenomena, rather than just one (in an unexpected way). Thus, UEA ‘constrains’ underdetermination. But why is it unlikely? Although it seems plausible that a theory is likely to fit the facts when it was constructed to fit the facts, the reverse does not follow. Moreover, Dawid should be aware that the use-novelty account he appeals to has been shown to be highly problematic (Musgrave Reference Musgrave1974). Suppose scientist S1, because of her ignorance of the theory itself or her ignorance of the extant phenomena, did not expect theory T to explain phenomenon P, whereas S2 did. Then, absurdly, T would receive more confirmation from P when considered by S1 than when considered by S2. Furthermore, although UEA seems very amenable to accounts that value the unifying power of theories in confirmation, Dawid, for some reason, does not make this connection.

Finally, MIA. MIA is a ‘metainduction’ over the past successful predictions of a research program to the likely future predictive success of the research program. Dawid takes his lead from the immense empirical success of the standard model, which correctly predicted a host of fundamental particles. Physicists were thus very confident that the standard model’s prediction of the Higgs boson was correct, even though they had to await its discovery for several decades (36–37). Dawid claims that string theory is part of the same research program as the standard model and concludes that therefore string theorists are entitled to trust in the future predictive success of string theory on the basis of the standard model’s past predictive success. But what are the identity conditions of a research program? What makes string theory part of the same research program as the standard model? Dawid has little to say about these basic questions.

Dawid makes another claim: the fewer the (existing) theoretical alternatives, the better one’s chances of choosing a theory in the present that will be predictively successful in the future. The track record of predictive success of a research program, accordingly, would indicate that “underdetermination is limited” (54–55). If that really were the case, NAA would be supported (see above). But again, I can detect little by way of argument for these claims. I cannot even see why this ought to be so, unless one supposes that a theory is more likely to be true. Then we would have good grounds for expecting our theory to perform well in the future too. But of course we need to be given reasons for this supposition. Dawid seeks to provide such reasons only later in the book. I will get to that in a moment.

The inference from the ensemble of NAA, UEA, and MIA to a limit on underdetermination, Dawid says, is an inference-to-the-best explanation (IBE). That is, the best explanation of the failure to find alternatives, the unexpected explanatory coherence of diverse phenomena, and the regular predictive success in the research program is that there exist few (if any) alternative theories in the relevant context. Dawid in fact holds that this ‘metalevel IBE’ underlies the standard IBE from empirical data to the truth of hypotheses: we must make sure that a likely hypothesis is among the ones considered (67). Dawid then provides a Bayesian analysis of his ‘metalevel IBE’, which he developed together with two coauthors elsewhere (Dawid, Hartmann, and Sprenger Reference Dawid, Hartmann and Sprenger2014). The upshot is that, even in the absence of direct evidence for H, the lack of alternatives to H confirms H; that is, P(T|F_A) > P(T), where F_A = scientists’ failure to find alternatives to T, under the assumption that the failure of scientists to find alternatives to T is not due to scientists’ limited cognitive capacities.Footnote ³ Dawid and his collaborators believe that MIA can compel us to accept this assumption in a research field with regular predictive success because, according to them, this is an indication of a lack of alternative theories. However, as I indicated above, MIA itself is questionable.

In the third part of the book, Dawid finally addresses a concern, which he flags early on in the book and which anyone familiar with the more recent scientific realism debate will have. Stanford (Reference Stanford2006) has argued very forcefully that our repeated failure in the past to conceive of better theories than we actually did suggests that we have little reason to believe that our current theories are likely to be true and that we are successful in getting at the theory that best explains our evidence. Stanford’s argument, also known as the problem of unconceived alternatives (PUA), directly threatens NAA and the very idea that our inability to conceive of alternatives actually is good evidence for the number of alternative theories being limited. Dawid’s response to this concern is that his strategies for transient underdetermination actually apply also to repeated transient, and even permanent, underdetermination because of the very nature of string theory (cf. 139 and 169). There are two properties in particular that Dawid highlights in this context: minimal length scale (133ff.) and structural uniqueness (141ff.). According to so-called T-dualities, and for reasons having to do with the compactification of dimensions, any distance measured below the string length can be translated back to the string length. Any information below the string length therefore becomes ‘redundant’. String theory is thus ‘final’ because it implies that no other theory will give us any greater knowledge about the world at the microphysical level. String theory is structurally unique in that, in contrast to many other physical theories, it does not possess any free parameters. Theories of gravitation, for example, have a gravitational constant that is determined experimentally. The standard model possesses more than 20 free parameters. The more free parameters, the easier it is to fit a theory as with the data. Many philosophers consider too much parameter freedom in a theory as a vice (Popper Reference Popper1959; Forster and Sober Reference Forster and Sober1994; Worrall Reference Gooding, Pinch, Schaffer and Gardenfors2002). Both the minimal length scale and the structural uniqueness of string theory, Dawid believes, justify ‘final theory claims’ about string theory: regardless of inductive risk, there is unlikely to be any theory in high-energy physics that will succeed string theory. I am not persuaded. At least one of these two ways of justifying final theory claims about string theory presupposes the truth of string theory itself. And the vast number of ground states of string theory (100⁵⁰⁰) render the theory extremely flexible with regard to a possible fit with the phenomena.Footnote ⁴ Indeed, as Dawid admits himself, it is even questionable whether string theory is a proper theory in the first place (175). Regardless, even if his arguments for the finality of string theory were successful, they would be very specific. But Dawid’s account is supposed to put limits on underdetermination not only in the assessment of string theory but also in many other contexts: predictive success and IBE (169–70), theory choice (in favor of string theory and to the disadvantage of cosmic inflation and quantum loop theory; 89–90), inferences to unobservables in physics and in paleontology (103ff.), the anthropic principle (149), and even data analysis (111). Dawid’s account is thus a bit of a theory of everything itself. However, in none of these contexts, it seems, are there any theoretical features comparable with string theory’s structural uniqueness or length scale. Dawid’s strategy against Stanford’s PUA therefore fails in all of these cases.

Although I am not convinced by many of Dawid’s arguments, I find his book valuable for a number of reasons. Dawid does a great job in laying out, in a comprehensive way, some of the core ideas of perhaps the most challenging theory that science knows to date and offers a thought-provoking, genuinely philosophical, perspective on it. I am highly sympathetic to his taking on what I described above as the hard explanatory task; sociological explanations fall too short in making sense of the popularity of string theory among theoretical high-energy physicists. I also think that he is spot on in assessing the at times acrimonious debate between friends and foes of string theory. It is one of the few instances, Dawid notes, in which the ‘expert principle’ is violated; that is, nonexpert scientists feel warranted to judge the soundness of an established research field on the basis of their intuitions about the scientific method, with the result that people are talking past each other. This does call for a debate of the scientific method, and Dawid deserves respect for accepting the challenge. I strongly recommend Dawid’s book for anybody interested in a highly accessible philosophical treatment of string theory and for anybody open to some fresh thoughts on the scientific method.