A US intelligence-funded CS group reformulated Heuer's ACH as Bayesian networks, not as rule-chaining

A University of South Carolina computer-science group (Marco Valtorta, Jiangbo Dang, Hrishikesh Goradia, Jingshan Huang, and Michael Huhns), working under a US intelligence-community research program (ARDA/NIMD), built software to translate Analysis of Competing Hypotheses matrices into Bayesian networks and compared the two formalisms' expressiveness: "We implemented a software program to translate analytic problems represented as ACH matrices into Bayesian networks and compare the result with that using the ACH method."

Their treatment of ACH's own logic — including "draw tentative conclusions... by trying to disprove the hypotheses instead of proving them" — is framed entirely in probabilistic and graphical-model terms: bipartite graphs, conditional probability tables, and diagnosticity recast as sensitivity/specificity. The paper never characterizes ACH's elimination logic as chaining, rule-based, or expert-system reasoning of any kind. This makes it the one AI-adjacent formalization of ACH surfaced in this search, and it points at a different cross-domain bridge than the backward-chaining echo: ACH ↔ probabilistic graphical models. That the same intelligence-community program later funded a PARC software build of ACH suggests the formalization of ACH ran through Bayesian/decision-analytic machinery, not production-rule systems. It is one supporting strand in the finding that no source draws an explicit ACH ↔ backward-chaining comparison.