At the 2024 meeting of the Southern Society for Philosophy and Psychology I participated in an author-meets-critics roundtable with Cameron Buckner on his book, From Deep Learning to Rational Machines. The other presenters were Raphaël Millière and Kathleen Creel. It was a good discussion but I doubt I will ever work these comments up for publication, so I’m posting them here. Nothing that has happened in the space of machine learning models recently seems to me to have substantially undermined any of these points. Even quite “sophisticated” deep network models, such as ChatGPT-4o, struggle to overcome the challenges that plagued earlier architectures. Having learned nothing, we seem grimly fated to re-wage the connectionism-classicism wars of the ’90s. However heated that past rhetoric might have been, it was still to some degree an academic debate. No longer. This time the arguments are amplified by the vast amounts of capital (and, therefore, political power) staked on the success of ever larger-scale DNNs. As I briefly suggest at the end of these comments, philosophic discussions of machine learning systems should acknowledge that they are taking place in a new epistemic and rhetorical environment and take as their point of departure the material conditions and economic interests that determine how and where these systems are deployed.

Continue reading →