arXiv:2606.27826v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) are increasingly deployed as embodied planners in egocentric environments, where task success requires not only achieving instructed goals but also acting in socially appropriate ways. While explicit goals may render certain actions optimal, implicit social norms often impose hidden constraints. Existing evaluations typically focus on explicit goal achievement or direct norm knowledge, seldom assessing whether planners can infer and apply these hidden constraints within action sequences. We introduce NormA
The increasing deployment of MLLMs as embodied planners necessitates robust evaluation benchmarks that account for the complexities of real-world social interaction beyond explicit task completion.
Achieving socially appropriate behavior in embodied AI is crucial for their widespread adoption and integration into human environments, moving beyond basic task execution to nuanced interaction.
The introduction of NormAct provides a standardized method to evaluate MLLMs' ability to infer and comply with hidden social norms, pushing the frontier of AI capabilities beyond just explicit goals.
- · AI developers
- · Robotics companies
- · Social AI researchers
- · Developers of socially inept AI
- · Ethical AI frameworks lacking social nuance
Embodied MLLMs will begin incorporating social norm compliance metrics into development.
Public acceptance and trust in AI-powered robots and agents will increase as their behavior becomes more human-aligned.
The definition of 'intelligence' in AI will expand to explicitly include social and ethical reasoning as core components.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI