28 Association of Research Libraries Research Library Issues 299 2019 Explainable Artificial Intelligence Michael Ridley, Librarian Emeritus, University of Guelph PhD Candidate, Western University and Postgraduate Affiliate, Vector Institute Introduction Algorithmic decision-making, enabled by machine learning, is ubiquitous, powerful, often opaque, sometimes invisible, and, most importantly, consequential. Machine learning is embedded in many information tools and systems, central to numerous research methods, and pervasive in the applications of everyday life. Safiya Noble emphasizes the critical nature of artificial intelligence (AI) by observing that it will become “a major human rights issue in the twenty-first century.”1 As with nearly all aspects of contemporary life, AI is having a profound influence on research libraries, scholarly communication, and key functions of the academy. Because “authority is increasingly expressed algorithmically,”2 it is crucial that this authority be interrogated and assessed with the same rigor and appropriate methods relevant to all aspects of the academic mission. Machine learning and deep learning are potent technologies that will be utilized to great advantage. However, “the danger is not so much in delegating cognitive tasks, but in distancing ourselves from– or in not knowing about–the nature and precise mechanisms of that delegation.”3 Hence the critical importance of “explainable artificial intelligence” (XAI) and its two pillars: trust and accountability. XAI is a diverse set of strategies, techniques, and processes that render AI systems interpretable and accountable. While some XAI approaches are highly technical, involving the perturbation of individual features in multi-layer neural network models, others are broad social and political policies enacted through regulation or legislation. Whatever the approach, XAI emphasizes explainability as an essential requirement for a technology that has for too long been defined by its opacity and what Frank Pasquale calls “remediable incomprehensibility.”4
Previous Page Next Page