Abstract

Abstract: In this talk, I analyze how different modeling strategies have been put to work in the computational models used to handle very complex targets in astrophysics and astrochemistry. The field of computational astronomy, for instance,  has been shaped by a diversity of approaches, from first very simple structure simulations in the 1970's, numerical experiments performed by Efstathiou or Mellott , idealized scenario or semi-analytic models, to high resolution hydrodynamical simulations.

My goal in this talk is to track the epistemological commitments and constraints of each of this strategy, and to show how the race to ever-more comprehensive models, in part encouraged by the funding landscape, has tend to blur what kind of questions can be addressed by each of these strategies. On this basis, I will distinguish predictive approaches from exploratory approaches for modeling complex systems and argue that many disagreements about how to solve discrepancies between observations and model's outputs actually stems from misinterpreting the kind of conclusions that the most complex and realistic simulations allow to draw. 

I will wrap up this presentation by highlighting how these considerations can inform the current debate about  machine-learning techniques and their legitimate use.