The Problem of Reverse Inference: Philosophical Models and Neuroscientifc Methods

Reverse inference is a crucial inferential strategy employed by neuroscientists to infer the engagement of a cognitive pro- cess from observed brain activation in functional magnetic resonance imaging studies. Despite its widespread use, in recent years reverse inference faced increasing skepticism, par- ticularly after Russell Poldrack authoritatively criticized its uncontrolled use in his influential 2006 paper. The general aim of the thesis is to provide an assessment of the current debate about reverse inference at the interface between neuroscience and the philosophy of science, by evaluating the models and methods proposed to improve the theory and practice of reverse inference. To do this, I offer three main contributions to the debate. In the first part of the thesis, I offer a comprehensive and updated overview of the problem of reverse inference in cur- rent cognitive neuroscience, emphasizing both technical and methodological issues. In the second part, I present and discuss the different philo- sophical models of reverse inference proposed in the literature, assessing their relative advantages and limitations. I also dis- cuss two novel models of reverse inference, one based on Bayesian confirmation theory and one based on the idea of inference to the best explanation. In the third part, I present the main results of a systematic review of the literature, aiming at evaluating the impact of NeuroSynth, the most widely employed meta-analytical soft- ware for performing reverse inference, on current practice in neuroscientific research.