In response to “Physics on Edge” (Vol. 3, No. 2).
To the editors:
In his essay, George Ellis expresses substantial concerns about recent ideas on theory assessment in fundamental physics. As one of the exponents of those ideas mentioned, I would like to address some of the concerns and sketch my understanding of the situation.
Let me start with a point that no one disagrees with: we are not free to choose the world we live in. We may prefer living in a world where our most fundamental physical theories are empirically fully testable within a short time. But it so happens that this is not the case. Today, the most far-reaching hypotheses tend to remain empirically untested for more than a scientific generation. Even worse, some important hypotheses have implications which seem to a large extent empirically untestable in principle. Given this situation, it is an important question how to understand the epistemic status of empirically untested or partly untestable theories.
In a recent book and a number of articles, I have suggested that, even in the absence of empirical confirmation, very significant reasons may lead scientists to believe that a given theory is probably viable as a description of phenomena in the world. Those reasons, which are spelled out by Ellis in his essay, are not purely theoretical. They are based on observations about the research process. The no-alternatives argument is based on the observation that no functioning alternative to the given theory has been found. Another argument is based on the observation that the given theory turns out to explain things it was not developed to explain. Finally, the relevance of these two arguments is supported by the observation that they had been good indicators of a theory’s eventual predictive success in comparable contexts of scientific reasoning in the past. Whether or not one should rely on those arguments is an epistemic issue. It is not merely about justifying work on a given theory. It is about having genuine trust in the theory’s viability. In the language of Bayesian confirmation theory, the described arguments in conjunction would, if successful, amount to substantial theory confirmation; they would substantially increase the subjective probability that the theory under scrutiny is viable in a given—so far empirically untested—regime.
This does not mean that arguments of non-empirical theory confirmation, as I call them, can ever reach the same level of conclusiveness as empirical confirmation. In our world, there are very good reasons to assume that they cannot. Empirical confirmation will always remain the ultimate judge over a theory’s fate. Significant doubts about the viability of an empirically unconfirmed theory will remain even in the case of strong non-empirical confirmation. But there is a lot of room between saying that we have conclusively established a theory’s viability and saying that we have no basis for assessing a theory’s chances of being viable at all. I argue that declaring empirical confirmation to be the sole mechanism that can move a theory away from the lower end of that spectrum fails to provide an adequate representation of the scientific process. I claim that an undogmatic assessment of the scientific process suggests that non-empirical confirmation can—on its own—generate a fairly high degree of trust in a theory’s viability.
The idea that a theory can be trusted to be viable in the absence of empirical confirmation may seem implausible and even unscientific at first glance. A fairly straightforward consideration can enhance the idea’s plausibility, however. Arguments of the very same kind as those involved in non-empirical confirmation also play a crucial role in establishing the significance of empirical confirmation. A particularly nice example of this mechanism is the discovery of the accelerating expansion of the universe in the late 1990s.
Two experimental groups at the time measured significant patterns of luminosity/redshift dependence in distant supernovae of a specific type. That data suggested that the expansion rate of the universe increased with time. The measured luminosity/redshift data was beyond doubt and clearly significant. What was far less clear was whether the inference from the data to an accelerating expansion was unassailable. It took several years of ruling out or rendering unlikely alternative explanations of the data before a consensus emerged that the hypothesis of accelerating expansion was well-established. Even today, some physicists retain doubts regarding this conclusion. The emerging consensus on accelerating expansion builds on a widening spectrum of empirical data that has reduced the prospects for alternative explanations. Still, that consensus could not have been formed without reliance on arguments of the very kind I have described in the context of non-empirical confirmation. From the observation that no convincing alternative explanations are in sight, one needs to infer that no such alternatives in fact exist. This kind of inference has to be supported by the observation that similar strategies of ruling out the existence of unconceived alternatives have worked fairly well as indicators of a theory’s viability in the past. Empirical confirmation of physical hypotheses like the accelerating expansion of the universe would not get off the ground if these non-empirical forms of reasoning lacked epistemic power.
The core idea behind non-empirical confirmation is simple. If the described non-empirical forms of reasoning carry epistemic weight when assessing the significance of empirical confirmation, it is implausible to categorically deny their epistemic significance in contexts where empirical confirmation is not available. After all, they do exactly the same job in both cases; they provide an assessment of the probability of unconceived alternatives to the hypothesis at hand.
I therefore argue that one needs to take arguments of non-empirical confirmation very seriously in cases like string physics, where they are, implicitly or explicitly, endorsed by many of the theory’s exponents. Needless to say, the actual strength of those arguments in each specific case must be evaluated by an in-depth scientific analysis of that individual case.
Ellis suggests that non-empirical confirmation is a philosophical rather than a physical argument. Based on the line of reasoning laid out in the previous paragraphs, I disagree with this understanding. No one would doubt that forming a consensus on the high confirmation value of a given set of empirical data, such as the consensus on the accelerating expansion of the universe described above, constitutes genuine physical reasoning. As we have seen, the assessment of unconceived alternatives is an integral part of generating such a consensus. Assessing unconceived alternatives thus is acknowledged as a genuinely scientific element of reasoning in this context. If so, there is no reason to start calling it a philosophical argument once it leads to non-empirical confirmation.
Ellis’s main concern is not about string theory but the multiverse scenarios of cosmic inflation and Everettian quantum mechanics. It is of crucial importance to emphasize a fundamental difference between a case like string theory and the multiverse case. One assumes that string theory, if fully understood, will lead to predictions that someday—and this may be in the very far future—can be tested. This understanding implies that the strength of the method of non-empirical confirmation in a context like string physics can itself be tested. Whenever a theory that was first trusted based on non-empirical confirmation is empirically confirmed, this also increases trust in the method of non-empirical confirmation. If a theory is refuted, this at the same time reduces trust in the method of non-empirical confirmation that had suggested the theory’s viability before. The possibility of empirically testing the method of non-empirical confirmation makes the method scientifically legitimate.
In the case of multiverse theories, it is possible to test predictions of eternal inflation in our universe. But trust in the multiverse hypothesis must rely on the understanding that the multiverse is the only conception that provides a consistent theoretical description of the observed phenomena. This step is a pure case of a no-alternatives argument. Non-empirical confirmation therefore is essential for endorsing the multiverse hypothesis. There is a problem, however: by definition, it is impossible to empirically test the idea that non-empirical confirmation works also with respect to hypotheses that are empirically untestable in principle. Empirical testing of the method thus is fundamentally constrained in this case, just like empirical testing of the theory itself. This fact threatens the reliability of non-empirical reasoning in the multiverse case.
I can very well understand Ellis’s unease on that basis. My understanding of the mechanism of non-empirical confirmation does reflect the problem Ellis is pointing at in the given context. Based on what I have said in the previous paragraph, one could plausibly come to the conclusion that non-empirical confirmation is not applicable to cases like the multiverse where future empirical testing of the hypothesis under scrutiny seems impossible.
On the other hand, given that both the multiverse theory and the method of non-empirical confirmation that is deployed in its support can be partly tested (with respect to the way they play out in our universe), it may seem implausible to entirely suspend judgement on their viability. In case the theory can be fully and consistently developed and empirically confirmed in our universe—and non-empirical confirmation looks strong when judged based on observations in our universe as well—this may be seen as a sufficient basis for taking non-empirical confirmation seriously with respect to the multiverse claim. One might argue that predictions extracted from anthropic reasoning, if they can be made sufficiently stable, do extend the realm of predictive hypotheses to other universes and therefore weaken the fundamental distinction between observable and unobservable parts of the multiverse.
To conclude, I believe that non-empirical confirmation has an important role to play with respect to string theory and other theories where comparable evidential situations have arisen, or may yet arise. The multiverse is a much more complicated case. One might argue that the multiverse increases the epistemic risk involved in non-empirical confirmation but keeps the core of the argument intact. One could also contend that the control mechanisms for non-empirical confirmation are fundamentally flawed in the multiverse case and non-empirical confirmation therefore must not be used in that context. I have no clear suggestion as to which position to choose. I do think, though, that the novel character of the problem requires a careful reconsideration of the concept of theory confirmation that focuses on what is new about the situation we face. Thinking about confirmation must be allowed to move beyond the confines of an understanding that seemed adequate at earlier stages of physical inquiry.
Georges Ellis replies:
I like Richard Dawid’s defense of his position. He makes a good case that one can build up support for a theory on purely theoretical grounds without any observations having yet been made. The example of the existence of gravitational radiation comes to mind. In that particular case, eventually extraordinary technology has given us the needed observations. But as he says,
Non-empirical confirmation … is essential for endorsing the multiverse hypothesis. There is a problem, however: by definition, it is impossible to empirically test the idea that non-empirical confirmation works also with respect to hypotheses that are empirically untestable in principle.
Precisely so. It is a careful and reasoned analysis. I look forward to the further development of his ideas in this context.
Richard Dawid is a philosopher of science at Stockholm University and has a PhD in theoretical physics.
George Ellis is Emeritus Distinguished Professor of Complex Systems in the Department of Mathematics and Applied Mathematics at the University of Cape Town in South Africa.