Inaugural
Embodied AI Meta-alignment Workshop @ AAAI-27
When, How, and With Whom should AI systems Align?
News
- The paper submission deadline has been extended to November 24, 2026 AoE.
- The Call for Papers is open. Submissions are due November 24, 2026 AoE, at the submission site on HotCRP.
- The workshop website is live. The Call for Papers will open on or after October 16, 2026.
- The Meta-Alignment workshop has been accepted at AAAI-27.
About the Workshop
We aim to bring together a community of researchers at the intersection of robotics, AI, and Human-Robot Interaction to study the embodied meta-alignment problem. At its core, meta-alignment is concerned with an AI system’s reasoning about the alignment problem itself: what should constitute the alignment parameters in the first place? Embodied agents present a unique challenge. Unlike pure software systems, embodied agents operate in shared spaces where actions can have immediate, irreversible consequences that directly impact nearby people. This includes what or whom it should align with, how competing alignment targets should be represented and reconciled, which alignment mechanisms are appropriate in a given context, and how alignment should balance competing objectives stemming from people’s perceptions of key factors such as safety, rights, norms, or institutional authority. We will explore how, in physical environments, AI agents can go beyond satisfying human preferences to actively influence human trust, expectations, and future preferences through their actions. A key outcome of the workshop is to identify methodologies that ensure the advancement of AI capabilities while understanding and limiting their negative influence.
Call for Papers
We welcome participation from researchers within the AI community, particularly at the intersection of robotics, AI, and Human-Robot Interaction (HRI), who are interested in embodied meta-alignment.
We solicit 2-4 page position papers and 6-8 page regular papers in AAAI format. Submissions will undergo double-blind review, with acceptance based on quality and relevance. Submission length will not be considered as a factor when evaluating submissions, as long as it is under the maximum page length. Accepted papers are non-archival and will be posted on the workshop website.
Topics of Interest include, but are not limited to:
Robotic Foundation Models and Embodied Agents
Architectures, pre-training, and representations for physical reasoning and embodied decision-making.
Human-Robot Interaction and Shared Autonomy
Collaborative control, physical interaction paradigms, and shared decision-making.
Value-aligned learning for physical systems
Reinforcement learning, inverse RL, and preference learning under physical and environmental constraints.
Dynamic Meta-Alignment
Formal models for dynamically selecting alignment parameters in multi-user physical systems.
Safety vs. Preference Tradeoffs
Optimizing for user satisfaction while bounding physical influence and enforcing physical safety constraints.
Perceiving Latent States
Measurement of human internal states such as trust, intent, and expectation via multi-modal perception.
Formal representations of alignment targets
preferences, intentions, values, norms, constraints, rights, roles, laws, and institutional objectives; alternatives to scalar reward/value representations.
Influence, persuasion, and manipulation
Distinguishing legitimate assistance and deliberative support from framing, strategic information selection, manipulation, and deception.
Multi-party and pluralistic alignment
reconciling conflicting preferences, roles and authority; social choice, negotiation, and group decision making in shared physical environments.
Evaluation of meta-alignment
Benchmarks, testbeds, metrics, causal interventions, and longitudinal evaluations for determining whether systems appropriately choose and revise alignment strategies.
Oversight and correction
Mechanisms through which people can understand, challenge, override, and revise an embodied agent’s alignment decisions.
Important Dates
All deadlines are 23:59 AoE.
-
Tue, Nov 24, 2026
Mon, Nov 16, 2026Paper Submission Deadline -
Wed, Dec 2, 2026Acceptance Notification
-
Wed, Dec 16, 2026Camera-ready Deadline
-
Monday, February 22, 2027, or Tuesday, February 23, 2027Workshop DateIn person at AAAI-27
Submission Instructions
- Papers must use the AAAI-27 author kit and be anonymized for double-blind review.
- References and appendices do not count toward the page limit.
- Reviewing is double-blind; each paper receives at least two reviews.
- Accepted papers will be posted on this website and presented as a poster or contributed talk.
Deadline: Tue, Nov 24, 2026 (AoE)
Submit via HotCRPInvited Speakers
Schedule
Tentative; all times local to the venue.
Opening and Framing (15 minutes)
Invited Talk I: Alignment Theory / Preference Learning (45 min)
Speaker One
Contributed Orals I: 4 x 15 minutes (60 minutes)
Break
Poster Session A: First Half of Accepted Papers (45 minutes)
Lunch
Invited Talk II: Cognitive Science / Behavioral Perspective (45 minutes)
Speaker Two
Contributed Orals II: 4 x 15 minutes (60 minutes)
Break
Poster Session B: Second Half of Accepted Papers (45 minutes)
Panel: Where is the line between calibration and manipulation?
Community Roadmap Breakout and Closing (30 minutes)
Accepted Papers
To be announced







