Browsing by Author "De Freitas, Julian"

Now showing 1 - 2 of 2

Open Access
Chatbots and mental health: Insights into the safety of generative AI
(John Wiley & Sons Ltd., 2023-10-26) De Freitas, Julian; Uğuralp, Ahmet Kaan; Oğuz-Uğuralp, Zeliha; Puntoni, Stefano
Chatbots are now able to engage in sophisticated conversations with consumers. Due to the “black box” nature of the algorithms, it is impossible to predict in advance how these conversations will unfold. Behavioral research provides little insight into potential safety issues emerging from the current rapid deployment of this technology at scale. We begin to address this urgent question by focusing on the context of mental health and “companion AI”: Applications designed to provide consumers with synthetic interaction partners. Studies 1a and 1b present field evidence: Actual consumer interactions with two different companion AIs. Study 2 reports an extensive performance test of several commercially available companion AIs. Study 3 is an experiment testing consumer reaction to risky and unhelpful chatbot responses. The findings show that (1) mental health crises are apparent in a nonnegligible minority of conversations with users; (2) companion AIs are often unable to recognize, and respond appropriately to, signs of distress; and (3) consumers display negative reactions to unhelpful and risky chatbot responses, highlighting emerging reputational risks for generative AI companies.
Open Access
Self-orienting in human and machine learning
(Nature Research, 2023-08-31) De Freitas, Julian; Uğuralp, Ahmet Kaan; Oğuz-Uğuralp, Zeliha; Paul L.A.; Tenenbaum, Joshua; Ullman, Tomer D.
A current proposal for a computational notion of self is a representation of one’s body in a specific time and place, which includes the recognition of that representation as the agent. This turns self-representation into a process of self-orientation, a challenging computational problem for any human-like agent. Here, to examine this process, we created several ‘self-finding’ tasks based on simple video games, in which players (N = 124) had to identify themselves out of a set of candidates in order to play effectively. Quantitative and qualitative testing showed that human players are nearly optimal at self-orienting. In contrast, well-known deep reinforcement learning algorithms, which excel at learning much more complex video games, are far from optimal. We suggest that self-orienting allows humans to flexibly navigate new settings.