A Neuro-symbolic Approach to the Logic of Scientific Discovery

Scientific discovery is a neglected topic in the philosophy of science. Since around the middle of the last century, the received view has been that discovery is not governed by logic or, more generally, by rationality, but is a largely elusive and inscrutable process. Thankfully, not everyone has been persuaded by this mystical view. Given the recent cascade of developments in automation and AI, this means that now more than ever we need to carefully re-evaluate our attitude towards this view. This paper aims to do precisely that by exploring how such developments are reinvigorating the search for a logic of scientific discovery. Neuro-symbolic approaches to AI, it will be argued, offer hope in the reinstatement of the rationalist model, and the automation, of scientific discovery. A neuro-symbolic system is proposed that seeks to integrate several identified desiderata like the ability to detect patterns and to reason. On this proposal, both neural and symbolic methods are employed in the service of automating scientific discovery, but the former are conceived of as subservient to the latter. The rationale for this subservience relation is that any discoveries made need to be integrated into an accessible body of knowledge, and that can more easily be achieved with symbolic methods at the helm.