(PS10-B43) Can Chatgpt Have the Theory of Mind Abilities Across a Lifespan, and What Does That Mean for Psychological Therapies?

Theory of mind (ToM) is the ability to infer not only our own mental states but also those of other people. It is an essential social-cognitive skill that proves to be key in psychological therapies, in order to successfully communicate and understand what support clients require. ToM also has large overlaps with cognitive empathy. In an age where psychological therapies can be increasingly digitized or supported by AI chatbots (e.g., Woebot), it is key to understand how we could evaluate empathy in interactive artificial agents, and assess whether this has an impact on how AI Chatbots apply psychological therapies. A recent study showed that a November 2022 version of GPT-3 (davinici-003) was able to perform comparably to a nine-year-old child on ToM tasks, which seems to have spontaneously emerged as a by-product from having better language skills. Our study extends this work by investigating the extent of ToM tasks the newest version of ChatGPT (dated Feb 13) would be able to complete. Specifically, we created a standardized battery of ToM tasks that captures ToM across a lifespan and tested it on ChatGPT. The battery created was designed to be a standardized test of ToM for ChatGPT and other interactive artificial agents as they improve over time. We also describe what implications ToM scores have for psychological therapies in interactive artificial agents.