ComBodiedAgents: New Paradigm Shifts AI Focus from Tasks to Human Long-term State
机器之心 · wechat · 2026-08-25
Researchers from 16 institutions including Mila, Tsinghua, and Oxford propose ComBodiedAgents, a new paradigm where agents optimize for human long-term state trajectories and agency, not just external task completion.
Core Shift
Current Digital and Embodied Agents focus on "doing things for people," which risks atrophying user capabilities. ComBodiedAgents aim for "growing with people," prioritizing what happens to the human after the task is done.
Technical Framework
The paper outlines a four-layer closed-loop system:
- Human-State Perception: Reconstructing event evidence from multimodal data with provenance and confidence.
- Longitudinal Memory: Maintaining memory of state trajectories and intervention feedback.
- Personal World Model (PWM): Predicting future impacts of different interventions (bounded, not a full digital twin).
- Intervention Planning: Deciding actions within constrained permissions, where "not intervening" is a valid action.
Deployment & Evaluation
The research advocates for an "edge-native" architecture where the user holds authority over the model. Evaluation expands beyond single-task success to long-term trajectories and agency preservation (e.g., user control, skill retention), classifying "irreversible overreach" as a critical failure.
More from AGI Musings
- AI Flooded Marketing With Content — Only Scarcity Will Work Now — nikvassev · 2026-08-25
- 11 configurations of firm/worker/agent require different designs, shifting focus from automation to collaboration — random_walker · 2026-08-25
- Best-case tech adjustment: incumbents keep jobs, youth enter new roles — soumitrashukla9 · 2026-08-25
- Nicolas Cole: 100x AI output comes from explicit process documentation — cen6wkf · 2026-08-25
- Hinton: No Sharp Line Between Tool and Subject, Agency Emerges — MacrinePhD · 2026-08-25
- The AI Disillusionment Law: Frontier Models Feel Godlike for Days, Then Dumb — D3VAUX · 2026-08-25