Newly Published Articles

Some Counterarguments to the Pessimistic Meta-Induction

November 6, 2008

By Victor Bilan

The success of science is beyond dispute. Currently held scientific theories make novel predictions, subsume ever-widening spheres of diverse phenomena, and open new avenues of fruitful research through their implications. It is a separate issue whether scientific theories, however successful, tell us the literal truth concerning unobservable entities. One argument favoring an anti-realist view of science—that we should not believe, or look for, this literal truth—has been termed the pessimistic meta-induction (PMI), and denies, by way of an induction over the history of science, that empirical success entails literal truth of any kind:

(P1) Current scientific theories are successful.

(P2) Past scientific theories were successful, but false.

(C) Current scientific theories are probably false.

While a scientific realist believes that the success of scientific theories licenses belief in the literal truth of the statements they make about unobservable entities, the thrust of the PMI is to deny the connection between success and truth with historical counterexamples, a notion made famously explicit by Larry Laudan (1981). In the following remarks I will describe some counterarguments to the PMI, first noting two different perspectives on the preservation of the link between success and truth and then describing a potentially fatal paradox that may arise from the use of the PMI to refute first-order observational evidence.

One objection to the PMI is that while the induction assumes that past theories are identical to current theories in all respects, current theories are actually different from past theories, and thus no inductive argument can be made. Current theories are more successful than past theories; they are more rigorously tested, make more accurate predictions, and are applicable over a broader range of domains than the past theories they supplanted. On these grounds, inferences drawn from past theories cannot apply to current theories. Adherents to “convergent realism” extend this argument, reasoning that since the success of scientific theories scientific theories has increased over time, it is reasonable to think that scientific theories appear to be converging on the literal truth of the phenomena they describe. While current theories may not be absolutely true, this convergence towards truth indicated by increasing success makes it reasonable to assign current theories some level of “approximate truth.” While realism still awaits a cogent account of approximate truth, this notion of convergence pulls strongly on intuition.

The PMI may be formulated in terms of Bayes’ theorem, with the goal of showing that realists commit the base-rate fallacy when they attempt to draw a connection between success and truth. However, this result can be shown to be founded on a sampling error on the part of the anti-realist (Magnus and Callender 2004). Consider the following presentation of Bayes’ theorem:

                                  Pr(Sx|Tx) * Pr(Tx)
Pr(Tx|Sx)   =  ______________________________________
                        Pr(Sx|Tx) * Pr(Tx)  +  Pr(Sx| ~Tx) * Pr(~Tx)

Let Tx represent “a theory x is true”, and Sx represent “a theory x is successful.” While the outcome of any argument for scientific realism is the assignation of a high value to Pr(Tx|Sx), the anti-realist seeking to attack this connection may argue that the realist has neglected that true theories, if they exist, would be extraordinarily rare, forming a very small subset of the set of all theories; thus, no matter how low a value we assign to the “false positive” term Pr(Sx|~Tx), a given successful theory will always be overwhelmingly more likely to be false than true. Thus, success is not a reliable indicator of truth. This form of the PMI misses the point, however; if a given successful theory is likely to be false, it is not because of a missing connection between success and truth, but simply because the set of past theories is heavily populated with false theories. Our only candidates for true theories are currently held theories, and it is this critical difference that is again being missed by the anti-realist. The fact that we have a large group of false past theories from which to take a Bayesian sample is no indictment of the truth of current theories; they are fundamentally different.

The discussion so far has criticized the anti-realist’s use of the PMI to attack the link between empirical success and literal truth. Another objection to the PMI that is worth noting concerns contrasting evidence bases, as phrased by Jarrett Leplin (2004). The success of current theories provides first-order evidence for the truth of current theories. The PMI, an induction drawn from past theories, provides second-order evidence in opposition to our first-order evidence. It is not at all obvious that second-order evidence should trump our first-order evidence; and even if we grant this dubious assertion, the PMI may well run aground on a paradox, since the conjunction of our first- and second-order evidence, necessarily allowable by the principle of conjunction that is necessary to scientific inference, results in a contradiction. While possible resolutions to this predicament have been explored (Bilan 2008), none is definitive.

The PMI remains the premier argument of the scientific anti-realist, but in this discussion I have attempted to highlight some important counterarguments. The arguments I cite (1) show that the PMI does not dispose of the link between empirical success and metaphysical truth nearly so obviously as its adherents claim, and (2) raise important questions about the relationship of the PMI’s second-order evidence to the first-order evidence of the theories it concerns.


Bilan, V. (2008), “Evaluating an Objection to Anti-realist Historical Induction Arguments”, unpublished manuscript, retrieved from

Magnus, P. D. and Callender, C. (2004), “Realist Ennui and the Base Rate Fallacy.”, Philosophy of Science 71: 320-338.

Laudan, L. (1981), “A Confutation of Convergent Realism”, Philosophy of Science 48: 19-49.

Leplin, J. (2004), “A Theory’s Predictive Success can Warrant Belief in the Unobservable Entities it Postulates”, in C. Hitchcock (ed.), Contemporary Debates in Philosophy of Science. Malden, MA: Blackwell.