Mario Alai

A.I., Scientific Discovery and Realism

Epistemologists have debated at length whether scientific discovery is a rational and logical process. If it is, according to the Artificial Intelligence hypothesis, it should be possible to write computer programs able to discover laws or theories; and if such programs were written, this would definitely prove the existence of a logic of discovery. Attempts in this direction, however, have been unsuccessful: the programs written by Simons group, indeed, infer famous laws of physics and chemistry; but having found no new law, they cannot properly be considered discovery machines. The programs written in the ‘Turing tradition’, instead, produced new and useful empirical generalization, but no theoretical discovery, thus failing to prove the logical character of the most significant kind of discoveries. A new cognitivist and connectionist approach by Holland, Holyoak, Nisbett and Thagard, looks more promising. Reflection on their proposals helps to understand the complex character of discovery processes, the abandonment of belief in the logic of discovery by logical positivists, and the necessity of a realist interpretation of scientific research.

Resisting the historical objections to realism: Is Doppelt’s a viable solution?

There are two possible realist defense strategies against the pessimistic meta-induction and Laudan’s meta-modus tollens: the selective strategy, claiming that discarded theories are partially true, and the discontinuity strategy, denying that pessimism about past theories can be extended to current ones. A radical version of discontinuity realism is proposed by Gerald Doppelt: rather than discriminating between true and false components within theories, he holds that superseded theories cannot be shown to be even partially true (except insofar they agree with current ones), while present best theories are demonstrably completely true. I argue that this position, running counter both the cumulativity of science and fallibilism, is untenable; it cannot account for the success of past theories, nor for the failures of current theories, and rather than shutting the door to the pessimistic historical objections it opens it wide. The best strategy, instead, joins the selective idea there was both some truth and some falsity in discarded theories, like in current ones, with the moderate discontinuity idea that the truth rate in present best theories is much greater than in past ones.

The Debates on Scientific Realism Today: Knowledge and Objectivity in Science

Debates on realism in science concern two main questions: whether theoretical knowledge is possible, and whether it is objective. Today, as in the past, the possibility of theoretical knowledge is often denied because of the empirical underdetermination of theories. Realists rely on explanatory power, theoretical virtues, and instrumental but theory-free observation to solve this problem. Besides, they use the no miracles argument for the truth of successful theories. Antirealists, however, deny that explanation is either necessary or possible, and that is a cue to truth. Moreover, they reject realism and the cogency of the no miracles argument by the pessimistic induction from the falsity of past successful theories. Some realists reply that there is a radical discontinuity between past science (largely off-track) and current science (basically sound). But this reply is at best insufficient, and most realists prefer to restrict their commitment to selected parts or features of theories, both past and present. Forms of “selective realism” are entity realism, structural realism, deployment realism and semirealism, but also the verisimilitude research program and the restricted-domain approach. Realists need criteria to identify the true components of theories, and a noteworthy candidate is essential involvement in functionally novel and surprising predictions. The second main question is a special instance of the old debate between realists and relativists or idealists: according to antirealists science cannot be objective, because of its inherently “perspectival” nature, characterized by a priori and subjective factors. On the contrary, perspectival realists argue that the specific “viewpoints” within which scientists must work do not prevent them to discover objective features of reality.

The Issue of Scientific Realism

Agazzi stages a complete, very detailed and overall convincing defence of scientific realism, its presuppositions and corollaries (mind-independence of reality, referentiality of theories, truth as correspondence, knowledge as the goal of science, justifiability of beliefs in unobservables through abductive arguments). But his claims that truth is relative to a circumscribed domain, not “pictorial” and not pertaining to theories, and that scientific objects certainly exist because they are nothing but abstract bundles of properties, are potentially ambiguous. Moreover, to the antirealist objections based on radical theory-change he replies that pre- and post-revolutionary theories do not contradict each other and are equally true, because each one deals with a different domain of objects of its own making: but this reply (apparently a legacy of neopositivistic operationalism) risks to make theories analytic, so slipping into conventionalism, or to reduce their content to observable phenomena, thus giving into antirealism.

The No Miracle Argument and Strong Predictivism Versus Barnes

Strong predictivism, the idea that novel predictions per se confirm theories more than accommodations, is based on a “no miracle” argument from novel predictions to the truth of theories (NMAT). Eric Barnes rejects both: he reconstructs the NMAT as seeking an explanation for the entailment relation between a theory and its novel consequences, and argues that it involves a fallacious application of Occam’s razor. However, he accepts a no miracle argument for the truth of background beliefs (NMABB): scientists endorsed a successful theory because they were guided by largely true background beliefs. This in turn raises the probability that the theory is true; so Barnes embraces a form of weak predictivism, according to which predictions are only indirectly relevant to confirmation. To Barnes I reply that we should also explain how the successful theory was constructed, not just endorsed; background beliefs are not enough to explain success, scientific method must also be considered; Barnes can account for some measure of confirmation of our theories, but not for the practical certainty conferred to them by some astonishing predictions; true background beliefs and reliability by themselves cannot explain novel success, the truth of theories is also required. Hence, the NMAT is sound, and strong predictivism is right. In fact, Barnes misinterprets the NMAT, which does not involve Occam’s razor, takes as explanandum the building of a theory which turned out to predict surprising facts, and successfully concludes that the theory is true. This accounts for the practically certain confirmation of our most successful theories, in accordance with strong predictivism.