A Safety Framework for AI in Caregiving: What Standard Evaluations Miss
A pilot study of how AI handles caregiving context and when its responses go beyond the evidence available.
Ali Madad · GiveCare ·
AI can lose relevant caregiving context or recommend an intervention the evidence does not support. This poster examines how to match a response to the evidence available in a conversation.
A pilot using four scripted conversations traces individual judgments to saved answers and specific turns. Its findings are limited to those cases. Further study is needed to assess performance across caregiving situations.