The Sotif Meta-Algorithm: Quantiative Analyses of the Safety of AI
Michael Wagner, Noah Carlson, Chris Dwyer · SAE technical papers on CD-ROM/SAE technical paper series · 2025
This paper presents updates to a “meta-algorithm” for achieving safer AI driven systems by integrating systems theoretic process analysis, quantitative fault tree analysis, structured generation of safety metrics, and statistical hypothesis testing of metrics between simulation and reality. This paper presents updates to the meta-algorithm after its application in use cases involving commercial autonomous vehicle deployment.