Amir Motesharei

Philosophy of machine learning frequently frames what machine learning models do in science in target-oriented terms. This paper argues that this framing misses a significant modeling function: across many stages of scientific practice, machine learning models are used primarily for exploration, with target-oriented, representational assessment deferred. I develop a framework for exploratory machine learning models, characterizing them by their functional role in carving and exploring spaces of theoretical/experimental possibilities under target indeterminacy. I then propose robustness, calibration, and downstream investigative yield as primary evaluation criteria. I further argue that such workflows can warrant theoretically constrained modal claims and support how-possible inferences.

Having offered the “argument from bad lot”, Bas van Fraassen has raised an important criticism on the Inference to the Best Explanation (IBE) and challenged scientific realism. Regardless of scientific realism, and given the reliance of today's scientific methods on IBE, this criticism poses a deeper threat to the validity of scientific enquiry. In this article, we will try to provide a way to deal with this threat, according to the structure of the scientific community. Some structural features of the contemporary scientific community have caused the scientific enquiry to be limited to the science derived from the Western tradition and deprived it of the theoretical opportunities of various traditions. The main claim of this article is that paying attention to different traditions together to produce science is a good way to expand scientific enquiry and deal with the mentioned threat. In the first section, we will depict how the argument from bad lot threatens the validity of scientific enquiry. In the second section, we discuss the close relationship between science and tradition, and then examine the status of this relationship in the contemporary scientific society. The final section is devoted to proposing the idea of pluralism regarding the use of the theoretical opportunities of different traditions in scientific enquiry, as a proposed solution to counter this threat.