The explainable AI (XAI) movement arose in 2016 from a problem identified by the US government research funder DARPA that warfighters would have difficulty using autonomous weapons if they couldn’t explain why these weapons do what they do. Since then, tens of thousands of papers have been written making suggestions about how AI decisions could be explained. Most of these are based on “post-hoc”, after the event, descriptive method. This research argues that post-hoc reasoning doesn’t meet the common meaning of “explain”.

Further, based on methods from the philosophy of technology, it argues that in many cases the action of an AI system may not be explainable in the deterministic sense generally found in technology. Instead, the researcher argues that the behaviour of AI is more like a natural phenomenon than a digital process, and should be approached using the tools of science for explaining natural phenomena.

Speaker bio

Dr Greg Adamson is an honorary Associate Professor at the University of Melbourne School of Population and Global Health. He was an editor of the IEEE document on ethics and AI systems, Ethically Aligned Design, and is Technical Activities Vice President of the IEEE Society on Social Implications of Technology. He is a leading expert on the work of Norbert Wiener, founder of cybernetics. His day job is cybersecurity practitioner in the transport industry.