Common knowledge: Mental modeling across minds • Josh Lovejoy, AUTONOMOUS Summit 2025 скачать в хорошем качестве

Common knowledge: Mental modeling across minds • Josh Lovejoy, AUTONOMOUS Summit 2025

Human beings can’t help but project their theory of mind onto any sufficiently complex system. If they can’t describe how something works in simple cause-and-effect terms, they’ll often resort to using anthropomorphisms instead: • The toaster oven is “acting up again”. • My car “doesn’t like” driving in this weather. • This cold is “tenacious”. • The city feels “alive” tonight. This is an adaptation, mind you, not a flaw. Theory of mind is 200,000+ year old firmware that’s served our species exceptionally well. It’s enabled homo sapiens to build relational mental models with nearly any agent, regardless of whether that agent is organic or artificial, animated or inert, embodied as a singular entity or as a collective. We effortlessly use our own internal cognitive processes as a lens through which the invisible workings of others can be made more legible. But none of this is unique to AI. What is unique to AI is that it appears to be capable of constructing its own theory of mind about us. Specifically, the human agent on the other side of the screen. This is because the single most defining condition in an AI’s programming is that its outputs must be considered “believable” by its recipients; which requires a form of perspective-taking in order to pull off with any kind of consistency. As such, a Human:AI relationship has the potential to take on a dynamic that heretofore was presumed solely the domain of Human:Human relationships: The ability to identify, disambiguate, form, and refine “common knowledge” through an ongoing multi-turn dialogue. In this brief talk, we'll define common knowledge, construct a model together that roughly represents its shape and dynamics, and then apply that model as a lens for Human:AI interaction design.