Chance, Experimental Reproducibility, and Mechanistic Regularity

Abstract

Examples from the sciences showing that mechanisms do not always succeed in producing the phenomena for which they are responsible have led some authors to conclude that the regularity requirement can be eliminated from characterizations of mechanisms. In this article, I challenge this conclusion and argue that a minimal form of regularity is inextricably embedded in examples of elucidated mechanisms that have been shown to be causally responsible for phenomena. Examples of mechanistic explanations from the sciences involve mechanisms that have been shown to produce phenomena with a reproducible rate of success. By contrast, if phenomena are infrequent to the point that they amount to irreproducible observations and experimental results, they are indistinguishable from the background noise of accidental happenings. The inability to detect or measure the phenomenon of interest against the background noise of accidental correlations makes it impossible to elucidate a mechanism by experimental means, to demonstrate that a proposed mechanism actually produces the phenomenon, and ultimately to justify why a hypothetical scenario involving an irregular mechanism should be preferred over attributing irreproducible happenings to chance.

Acknowledgements

This work was supported by a generous fellowship from the Konrad Lorenz Institute for Evolution and Cognition Research. In particular, I would like to thank Stuart Glennan for useful discussion and comments on earlier drafts of this article. I would also like to thank the editor, James W. McAllister, and two unnamed referees for their excellent comments and suggestions.

Notes

[1] ‘Mechanisms are entities and activities organized such that they are productive of regular changes from start or set-up to finish or termination conditions’ (Machamer, Darden, and Craver 2000, 3). Alternatively, a mechanism is ‘a complex system that produces that behavior by the interaction of a number of parts, where the interactions among parts can be characterized by direct, invariant, change-relating generalizations’ (Glennan 2002).

[2] Conversely, the fact that a mechanism reliably succeeds in producing a phenomenon (regularity i) does not automatically entail that the mechanism always functions in the same manner (regularity ii). This might be particularly relevant in molecular biology, where biological phenomena are produced as a result of many copies the same molecular mechanism functioning at the same time (Bogen 2005, 414n22). Thus far, mathematical models have shown that the same molecular network may be characterized by more than one stable state, and it has been hypothesized that such states may underlie developmentally differentiated cell types (Kauffman 2004) or physiological cell states (e.g. proliferating vs. apoptotic cells; Huang 1999). According to these hypotheses, the same molecular network may function in two or more significantly different ways, although there is nothing here to suggest that each copy of a molecular mechanism functions in a singularly different way.

[3] ‘We have studied the mechanism of voltage-dependent gating using biochemical, X-ray crystallographic and electrophysiological methods’ (Jiang, Lee, Chen, et al. 2003, 34). All the items in the list are experimental techniques routinely used in neuroscience, and, with the exception of electrophysiological measurements, in molecular biology in general.

[4] This applies both to the ‘ontic’ and the ‘epistemic’ views of mechanistic explanations (Illari and Williamson 2011; 2012). According to the ontic view, explanations are objective features of the world (Salmon 1984; Craver 2007). On the epistemic view, they are descriptions of the causal-mechanistic structure of the world (Bechtel 2008).

[5] For instance, mathematical models revealed a wealth of surprising properties of molecular mechanisms which previously escaped our intuitions (Baetu forthcoming).

[6] Theoretical approaches proved extremely fruitful, and there cannot be any doubt that they play an important role in scientific discovery. For example, it seems highly unlikely that entities that are not directly observable, such as atoms or DNA, would have ever been discovered, have they not been previously postulated by theories aiming to explain puzzling phenomena, systematize our knowledge, or probe the limits of that which is theoretically possible.

[7] For example, possible mechanisms of DNA replication were devised only after a detailed knowledge of the chemical structure of the DNA double-helix became available.

[8] An organism's defense reactions to pathogens and potentially dangerous chemicals. The first evidence for the cellular basis of immunity comes from the work of Elie Metchnikoff on phagocytosis.

[9] Conversely, interventions that do not result in changes may be interpreted as possibly targeting causally irrelevant factors, although it should be noted that negative results do not preclude the possibility of nonlinear causal contributions or non-modular causal interactions in which factors that must act in concert with other factors in order to contribute to a change in the phenomenon of interest.

[10] In parallel, another breakthrough resulting in the elaboration of a more precisely circumscribed experimental set-up and the elucidation of the genetic basis of transfusion reactions, was the realization that immune rejection is less likely to happen if donor and recipient are genetically related, followed by the discovery that the probability of immune reactions associated with transfusions follows consistently reproducible probabilistic patterns that vary depending on the degree of genetic relatedness between donor and recipient. For a brief overview of the discovery of the ABO blood types and their genetic basis, see Crow (1993).

[11] Examples of mechanisms and their discovery are discussed in Bechtel (2006), Darden (2006), Craver (2007), Bechtel and Richardson (2010), and Baetu (2012).

[12] At the same time, it is also reasonable to assume that the ability to conduct experiments cannot increase indefinitely and that, in the end, no amount of experimentation can compensate for a complete lack of mechanistic regularity.