Phillip Kieval

State-of-the-art machine learning systems now routinely exceed benchmarks once thought beyond the ken of artificial intelligence (AI). Often these systems accomplish tasks through novel, insightful processes that remain inscrutable to even their human designers. Taking AlphaGo’s 2016 victory over Lee Sedol as a case study, this paper argues that such accomplishments manifest the essential features of achievements as laid out in Bradford’s 2015 book Achievement. Achievements like these are directly attributable to AI systems themselves. They are artificial achievements. This opens the door to a challenge that calls out for further inquiry. Since Bradford grounds the intrinsic value of achievements in the exercise of distinctively human perfectionist capacities, the existence of artificial achievements raises the possibility that some achievements might be valueless.

We claim that scientists working with deep learning (DL) models exhibit a form of pragmatic understanding that is not reducible to or dependent on explanation. This pragmatic understanding comprises a set of learned methodological principles that underlie DL model design-choices and secure their reliability. We illustrate this action-oriented pragmatic understanding with a case study of AlphaFold2, highlighting the interplay between background knowledge of a problem and methodological choices involving techniques for constraining how a model learns from data. Building successful models requires pragmatic understanding to apply modelling strategies that encourage the model to learn data patterns that will facilitate reliable generalisation.

The predominance of machine learning based techniques in cognitive neuroscience raises a host of philosophical and methodological concerns. Given the messiness of neural activity, modellers must make choices about how to structure their raw data to make inferences about encoded representations. This leads to a set of standard methodological assumptions about when abstraction is appropriate in neuroscientific practice. Yet, when made uncritically these choices threaten to bias conclusions about phenomena drawn from data. Contact between the practices of multivariate pattern analysis and philosophy of science can help to illuminate the conditions under which we can use artificial neural networks to better understand neural mechanisms. This paper considers a specific technique for MVPA called representational similarity analysis. I develop a theoretically-informed account of RSA that draws on early connectionist research and work on idealization in the philosophy of science. By bringing a philosophical account of cognitive modelling in conversation with RSA, this paper clarifies the practices of neuroscientists and provides a generalizable framework for using artificial neural networks to study neural mechanisms in the brain.

According to doxastic involuntarism, we cannot believe at will. In this paper, I argue that permissivism, the view that, at times, there is more than one way to respond rationally to a given body of evidence, is consistent with doxastic involuntarism. Rober :837–859, 2019a, Philos Phenom Res 1–17, 2019b) argues that, since permissive situations are possible, cognitively healthy agents can believe at will. However, Roeber fails to distinguish between two different arguments for voluntarism, both of which can be shown to fail by proper attention to different accounts of permissivism. Roeber considers a generic treatment of permissivism, but key premises in both arguments depend on different, more particular notions of permissivism. Attending to the distinction between single-agent and inter-subjective versions of permissivism reveals that the inference from permissivism to voluntarism is unwarranted.