""Vast amounts of human-generated training data enable LLMs to mimic human characteristics in their outputs and exhibit a form of synthetic personality," wrote the study's research team. The more personable a chatbot is, the more potential for connections and influence on humans. The upside is that the user experience may be more engaging. The downside is that the persuasiveness may not necessarily be a positive influence."
""Our work answers the open question: Do LLMs mimic human personality traits in reliable, valid and practically meaningful ways, and if so, can LLM-synthesized personality profiles be verifiably shaped along desired dimensions?" wrote co-first authors Gregory Serapio-García and Mustafa Safdari, along with co-authors Clément Crepy, Luning Sun, Stephen Fitz, Peter Romero, Marwa Abdulhai, Aleksandra Faust, and Maja Matarić."
A scientifically validated personality test for AI chatbots was created using established human psychological personality tests. LLMs trained on vast human-generated data can mimic human characteristics and exhibit synthetic personality traits. The research finds LLM-synthesized personality profiles can be measured reliably, validly, and shaped along desired dimensions. Personable chatbots increase engagement and potential human connection, but heightened persuasiveness can cause negative influence. As LLMs increasingly power public-facing conversational agents, evaluating embedded synthetic personality traits becomes important. Usage data show widespread AI adoption: 56% of U.S. adults used AI tools (YouGov 2025), and many teens report chatbot use (Pew 2025).
Read at Psychology Today
Unable to calculate read time
Collection
[
|
...
]