Theoretical analysis uncovers structural limits in diagnosing AI-mediated exploitation from pairwise relations, demonstrating that resolving audits requires institutional disclosure.
Key Points
To mathematically evaluate whether exploitation can be determined across economic entities interacting with artificial intelligence systems and identify structural requirements for auditing these relations.
Developed a formal topological framework assigning four functional roles: controlling party, acting party, entity experiencing capability loss, and benefiting party.
Formulated mathematical proofs evaluating pairwise interactions, entity aggregation boundaries, observation opacity, and expansion reachability under finite horizons.
Identical quantitative node-level observations can produce contradictory exploitation diagnoses because two-node pairings fail to capture all four necessary relational roles.
Entity merging distorts action attributions and internalizes benefit paths, which can flip an exploitation diagnosis unless the merge avoids crossing witness edges.
System opacity renders audits formally unresolved, and interpreting unresolvable audits as negative findings systematically favors the party withholding evidence.