Machine Impostors Avoid Human Detection by Interrupting the Formation of Stable Conventions: A Minimal Turing Test
Thomas Franz Eisenmann, Levin Brinkmann, James Winters, Niccolò Pescetelli · 2022
Humans are increasingly interacting with algorithms, and these algorithms do not necessarily disclose their identity. The classic approach to humans’ ability to recognize bot impostors, known as the “Turing test”, is focused on natural language. In the current study, we avoid natural language in a minimal Turing test setup, opening up space to study the foundations of human communication. In particular, we are interested in the roles of emerging conventions and reciprocal interaction for successful communication. Our experiment asked participants to distinguish between a human partner and a bot impostor in online interactions relying only on virtual movements in a 2D space. The main hypothesis was that access to the interaction history of a pair would make a bot impostor more deceptive because it can interrupt the formation of novel conventions between the human participants. By comparing bots that imitate behavior from the same or a different dyad, we find that impostors are more deceptive when they copy the participants’ own partners and that this leads to less conventional interactions. We also show that reciprocity is beneficial for success when the bot impostor prevents conventionality. We conclude that machine impostors can impede their detection and the formation of stable conventions by imitating past interactions, and that both reciprocity and conventionality are adaptive strategies under the right circumstances. Our results provide new insights into the emergence of communication and imply an increased risk through bot impostors online when they can access personal information, e.g. on social media.