I am exceedingly happy to report the release of my first academic paper in this field, “Classification Schemas for Artificial Intelligence Failures.” It is on the arXiv site here.
Last year I reached out to professor Roman Yampolskiy, one of the endorsers of Crisis of Control, a prolific author and speaker on the topic of AI Safety (and who coined that term), asking whether there was anything I could do to assist, and he suggested writing a paper on AI failures, which he would supply data for.
Roman wrote a 2018 paper cataloguing several dozen failures of artificial intelligence and other cybernetic systems, raising important questions about where this trend might go as AIs become more complex and powerful. My paper extends that analysis to further questions and a classification schema that facilitates the categorization of AI failures.
I’m excited at how it turned out and I look forward to it being picked up in a suitable journal!