I will consider Decision Basis Evaluation: Toward Safe Machine Intelligence a success if it puts decision basis evaluation on the ‘radar screen’ of machine intelligence engineers developing autonomous machine decision makers (AMDMs). Whether these engineers actually incorporate decision basis evaluation and decision repair into their machines is out of my control, but I hope that they will at least make a conscious decision to include or exclude these algorithms.
Another potential positive impact of the book is in the area of regulation. If AMDMs lacking decision basis evaluation and decision repair are deemed a safety hazard, the deployment of such machines could be restricted. Inferring the presence or absence of decision basis evaluation and decision repair algorithms from an AMDM’s overt behavior may be difficult, so enforcing regulations restricting AMDM deployment could be challenging.
A third way in which the book would be a success is if it attracts engineers and researchers interested in artificial intelligence safety to decision basis evaluation as an area of study. The book is almost 200 pages, yet there are many interesting avenues for further exploration and development.