← News

Agents & MCPSources: arXiv · 18 Nov 2026

OAK-Mind infers mental health changes from wearable and phone data without training

A zero-training, multi-agent LLM pipeline reads passive sensing data and rates mental health changes against a person's baseline, with written rationale.

In plain words

Researchers describe OAK-Mind, a system that uses large language models to interpret everyday sensor data from wearables and smartphones without any task-specific training. It organizes signals like sleep and activity, has specialized agents interpret them using mental health literature, and produces human-readable symptom ratings. The authors report performance comparable to or better than supervised models that needed labeled data.

What was announced

The pipeline, described in an arXiv paper abstract, arranges sensing data into behavioral dimensions including sleep, physical activity, location mobility, and smartphone use. Specialized agents grounded in mental health literature interpret longitudinal behavioral changes, and the system synthesizes their assessments with questionnaire-based symptom rating to infer changes relative to an individual's baseline assessment. Each inference comes with a human-readable rationale tied to observed behavioral patterns and domain knowledge.

According to the authors, across datasets and prediction tasks OAK-Mind achieves comparable or better performance than the most predictive supervised baselines trained on additional labeled samples, despite using no task-specific training. It also shows consistent improvement over direct prompting across four tested LLM backbones. Ablation analyses support combining behavioral interpretation with symptom rating, distributing analysis across domain-specific agents, and incorporating domain knowledge. The evidence is limited to the paper's abstract, so dataset details, participant populations, and validation methods are not described here.

What it does not show

  • The evidence is the paper's abstract only; full methodology, datasets, and peer-review status are not covered.

Sources

Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 19 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.