Evidence Sufficiency
A stopping-condition principle in AI agent design, stating that an agent should halt when it has gathered sufficiently reliable information such that further action would not change the decision, or when remaining uncertainty requires human intervention. The concept draws from the cognitive science principle of satisficing.
Timeline (last 90 days)
Evidence sufficiency as a stopping condition should trigger in one of three situations: sufficient reliable information has been obtained, further action is unlikely to change the decision, or the remaining uncertainty requires human intervention.
Evidence sufficiency is not a fixed threshold; it is highly dependent on the risk level of the task, and the required level of evidence sufficiency for medical diagnosis tasks is far higher than for everyday information retrieval.
All Facts (2)
Evidence sufficiency as a stopping condition should trigger in one of three situations: sufficient reliable information has been obtained, further action is unlikely to change the decision, or the remaining uncertainty requires human intervention.
50%UnverifiedEvidence sufficiency is not a fixed threshold; it is highly dependent on the risk level of the task, and the required level of evidence sufficiency for medical diagnosis tasks is far higher than for everyday information retrieval.
50%