A framework for testimony-infused automated adjudicative dynamic multi-agent reasoning in ethically charged scenarios
Brandon Rozek, Michael Giancola, Selmer Bringsjord, Naveen Sundar Govindarajulu · 2022
In "high stakes" multi-agent decision-making under uncertainty, testimonial evidence flows from "witness" agents to "adjudicator" agents, where the latter must rationally fix belief and knowledge, and act accordingly.The testimonies provided may be incomplete or even deceptive, and in many domains are offered in a context that includes other kinds of evidence, some of which may be incompatible with these testimonies.Therefore, before believing a testimony and on that basis moving forward, the adjudicator must systematically reason to suitable strength of belief, in a manner that takes account of said context, and globally judges the core issue at hand.To further complicate matters, since the relevant information perceived by the adjudicator changes over time, adjudication is a nonmonontonic/defeasible affair: adjudicators must dynamically strengthen, weaken, defeat, and reinstate belief and knowledge.Toward the engineering of artificial agents capable of handling these representation-and-reasoning demands arising from testimonial evidence in multi-agent decision-making, we explore herein extensions to one of our prior cognitive calculi: the Inductive Cognitive Event Calculus (IDCEC).We ground these extensions in a recent, tragic drone-strike scenario that unfolded in Kabul, Afghanistan, in the hope that use by humans of our brand of logic-based AI in future such scenarios will save human lives.