Seeing Is Not Knowing: The Epistemic Limits of LLM-Based Mental Health Inference from Youth Social Media Conversations

arXiv:2611.08565v1 Announce Type: cross Abstract: Adolescents increasingly use social media for private conversations, sharing experiences and seeking support. These interactions generate digital traces, prompting efforts to infer mental health and psychosocial states. As large language models become more capable of reasoning about conversations and interpersonal dynamics, a fundamental question remains nonetheless: what can models actually "know" about an individual's psychological experience from what they are able to "see"? We evaluated ten reasoning LLMs using donated Instagram Direct Messages from 80 adolescents with a history of depression or suicidal thoughts and behaviors, comparing predictions with self-reports and analyzing 204 reasoning traces. Models performed better when inferring more observable states than participants' subjective appraisals of social experience, such as perceived social support. Beyond prediction accuracy, our findings highlight the epistemic limits of digital traces for AI design and evaluation: what can be observed, what can reasonably be inferred, and what requires people's own accounts.

LLMs have epistemic limits in inferring mental health from social media conversations.

The study highlights that LLMs can infer some observable social states from social media but have limits in accurately understanding subjective mental health experiences, indicating potential risks in misinterpretation and moderate utility in adjunctive support. The overall impact is mixed, with cautious implications for AI's role in mental health inference.

This is a cautionary academic study showing LLMs have real but limited ability to infer mental states from private youth messages, correctly flagging epistemic limits rather than hyping capability. The implication is mixed: it tempers overconfident AI deployment in sensitive domains while surfacing real capability and privacy concerns.

The study directly addresses the capabilities and limitations of AI (LLMs) in a highly sensitive domain: inferring adolescent mental health from private social media data. Its findings indicate that while some observable states can be inferred, subjective experiences remain largely beyond AI's grasp, underscoring significant privacy and accuracy concerns if these tools are misapplied. This suggests a material negative implication for human flourishing if AI-based mental health inferences are used without understanding their epistemic limits.

The text discusses limitations in AI's ability to infer mental health from social media. It implies that AI may misinterpret or oversimplify complex human experiences, leading to potential misdiagnosis or misjudgment. The study found that AI performed better on observable states than subjective experiences, highlighting a risk of oversimplification.

SOURCE     arXiv cs.CY (Computers and Society) (tier 1)
   ↓
DOCUMENT   25b49daf-6850-46c8-bec9-0fa747e89212
           https://arxiv.org/abs/2611.08565
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  5 runs, methodology 1.0.1
   ↓
SCORE      -58  (Adverse)
   ↓
CONFIDENCE 79%