The unreasonable effectiveness of traces

Adrian Currie (Cambridge University)

Abstract

Obdurodon tharalkooschild was an enormous platypus living in Australia around fifteen million years ago. The lineage’s existence only became known in the last decade, and only due to an entirely to a single molar tooth. Most philosophers and scientists, when reflecting on historical reconstruction, provide trace-centric accounts of how we access the past. In this paper I’ll present a problem for these accounts, before resolving and drawing lessons about how philosophers should think about evidence. Traces are the downstream remains of past events: fossils (of molars, for instance) and trackways in paleobiology, middens and ruins for archaeologists, background radiation for cosmologists, and so on. By tracing how such remains form, historical scientists infer from remains to the past. An account of methods in historical science is ‘trace-centric’ to the extent that it construes inferences from traces to the past as the main business of those methods.

In this paper, I’ll suggest that traces, as understood by trace-centric accounts, are unreasonably effective. A charge of ‘unreasonable effectiveness’, most common in mathematics, suggests that something works better than it ought to given our conception of it. The charge suggests some explanatory failure on the part of our understanding of that thing. In cases like O. tharalkooschild, highly stable, well-supported and rich reconstructions of the past are made on the basis of very small data-sets. On the face of it, if historical reconstruction is simply a matter of the relationship between traces and the past, these cases are problematic: surprising claims need a lot of evidence, and these claims apparently only require a few scrappy traces.

I’ll resolve this puzzle by emphasizing the role of middle-range theory, our knowledge of how traces form, which are required for taking traces as evidence in the first place. When we appeal to the fine-grained details of these—the amount of knowledge we have about how mammalian molars form, for instance—how so much can be done with so little is no longer puzzling. Traces are not unreasonably effective after all, but to understand them we need to attend to the local details. I’ll consider what upshots this has for abstract philosophical accounts of evidential reasoning, suggesting that their very abstractness obscures the local, explanatory detail.