Heather Douglas: Bases for trust in AI

The rise of AI has tremendous promise for helping to assist with judgments in complex situations.  However, automated expert systems must be open to evaluation as sources for trusted judgment in the same ways that human experts are.  In this talk, I will describe three bases for evaluating expert systems and describe how they apply to computer-based expert systems.  The three bases are 1) a demonstration of expertise, either in terms of obvious success (including error rates) or explanation of judgment; 2) grounding of expertise in an expert community that functions well and exhibits the requisite diversity; 3) a display of the required values, including social and ethical values.  Currently, AI is limited to trust on the basis of obvious success.  Building the other bases of trust is essential for the promise of AI to be fulfilled.

Eric Horvitz: AI, people, and society: Rising questions and directions

Sabina Leonelli: Big data analysis and the human face of automated systems

Within the world of research and knowledge production, the current emphasis on big data is closely associated with the opportunities offered by the increasing digitalisation and automation of processes of discovery. Using computational tools to efficiently mine large datasets for new insights is often portrayed as an alternative - indeed, a replacement - for labor-intensive methods of data analysis requiring extensive human intervention and decision-making. Building on extensive empirical research and philosophical analysis of data processing practices across the biological and biomedical sciences, I shall take issue with this narrative. In its place, I shall argue that human judgement and extensive manual labor are a necessary component of any reliable computational system for data analysis. Automation is not an alternative to human intervention: rather, the automation of research requires new forms of human mediation and decision-making, geared towards ensuring the trustworthiness, reliability and fairness of the processes through which data are collected, analysed and interpreted. Within this approach to the use of algorithms in data analysis, ethics plays a crucial role not only in fostering socially meaningful discovery, but also in promoting sustainable and robust inferential reasoning.