Dr Peta Masters - King’s College London

To counter deception, it is first necessary to recognise that it is occurring. Goal recognition (GR) is a much-studied problem in AI whereby an agent’s goal is determined from its behaviour. Dr Peta Masters’ work will use a combination of novel and well-established GR techniques to screen out deceptive behaviours to reveal a deceptive agent’s underlying motivation. Setting aside inadvertent and ‘benevolent’ deception (so called white lies), her focus is strategic deception: deliberately deceptive behaviour used as the means to an end. The programme has three aims: (1) Identify and formalise mechanisms used to deceive humans and verify their effectiveness. Extending her own work and that of others, this will draw on psychological principles exploited (so known to work) by professional deceivers such as military strategists, advertisers, and stage magicians. (2) Develop and verify a human-like GR system that, when it is exposed to deceptive behaviour, ‘believes’ what a human would believe. (3) Develop a behavioural deception filter (BDF). The BDF runs the GR system, then interrogates it to understand how it arrived at its conclusions. This enables the BDF to recognise and filter out mechanisms that might have been used to manipulate the human-like system's beliefs. It thereby seeks to reveal any underlying motivations that may have been obscured. The purpose of the research is not only to reveal the deceptive agent’s underlying motivation but also to provide an understanding (through the human-like GR system) of what they hope – and expect – to be believed. So armed, intelligence, surveillance, and reconnaissance (ISR) can develop a plan that allows an adversarial agent to think their deception has worked when it has not. Thus, this programme of research will take an important step towards counter deception not only in the sense of deception detection but in the stricter sense that it provides a basis for deceiving the deceiver.