These aren’t the droids you’re looking for: a sociotechnical systems perspective on AI safety

Abstract

As artificial intelligence (AI) becomes more advanced and ubiquitous across society, concerns are increasing around potential risks. Though various AI safety principles and guardrails have been proposed, there is little guidance to assist stakeholders in understanding and preventing such risks. Further, though highly relevant, there are few sociotechnical systems perspectives on AI safety. This presentation provides an overview of the findings from a program of research applying Human Factors and Ergonomics (HFE) and sociotechnical systems methods and principles to a. identify the risks associated with future advanced AI technologies, and b. identify the controls that are required to support the design, implementation, and use of safe, ethical, and beneficial AI. Eight categories of AI risks are discussed along with examples from two case studies in road transport and defence. The SECURE (Safe, Ethical, Controllable, Usable, Responsible, and Effective) network of sociotechnical system controls is then outlined with the aim of supporting the effective management of the risks throughout the AI technology lifecycle.

Bio

Paul is a professor of Human Factors and creator of the Centre for Human Factors and Sociotechnical Systems at the University of the Sunshine Coast. He has almost 25 years’ experience of applied Human Factors and systems science research in a diverse set of domains. His current research interests are focused on the application of systems science to manage global level risks. Paul has received several prestigious awards from the International, Australian, US, and UK Human Factors and Ergonomics societies, and for the past 5 years The Australian has identified him as Australia’s field leader in the area of quality and reliability.