Towards a Framework for Detecting Deceptive Contextual and Behavioural Signals

Like "every contact" in forensics, deception leaves a trace: not fingerprints or DNA but—in many cases—evidence of conduct that exploits the belief-formation and reasoning processes of the deceived. Every deception takes advantage of particularities in the model that the deceiver has constructed of its target, whether that target be human, an autonomous system, or a human-AI team. This paper focuses on the former: insights from psychology and sociology used to fool people. Drawing on the work of professional deceivers from disciplines such as defence, cybersecurity and stage magic, we examine the factors in play and propose a framework within which those factors may be operationalised as an AI system that will be capable of detecting deception because it is capable of being deceived.