The 41st Annual AAAI Conference on Artificial Intelligence
Understanding how repeated, adaptive, and personalized AI interventions affect human decisions, behavior, trust, autonomy, and long-term outcomes.
Workshop scope
AI systems increasingly operate not only as predictors, but also as interventions that can influence human decisions, preferences, trust, behavior, and long-term outcomes.
Foundation models, AI agents, recommender systems, conversational assistants, digital health systems, educational technologies, and decision-support tools increasingly interact with users repeatedly and adaptively.
Traditional evaluation based primarily on predictive accuracy, offline metrics, short-term engagement, or one-time user studies is often insufficient for understanding these systems.
CALE-AI focuses on how AI interventions can be rigorously evaluated in terms of their causal, longitudinal, behavioral, and human-centered effects.
“A user-facing AI action or policy—such as a recommendation, explanation, nudge, warning, conversational response, or agent action—that changes the information, options, timing, or action environment presented to a person.”
The workshop considers both intended behavior-change interventions and incidental effects of AI systems on preferences, trust, reliance, autonomy, and well-being.
Main workshop themes
Randomized controlled experiments, A/B tests, micro-randomized trials, sequential multiple-assignment randomized trials, quasi-experimental designs, interrupted time series, difference-in-differences, causal machine learning, counterfactual evaluation, and off-policy evaluation.
Focus on separating intervention effects from self-selection, context, and system adaptation.
Repeated exposure, cumulative effects, delayed effects, sustained effects, carryover effects, habituation, reactance, preference formation, time-varying confounding, non-stationarity, human–AI co-adaptation, and feedback loops.
User-endorsed goals, autonomy, informed consent, transparency, fairness, privacy, reversibility, contestability, non-manipulation, and unintended harms.
Metrics beyond accuracy, engagement, and satisfaction, including behavioral outcomes, cognitive outcomes, health, sustainability, trust, appropriate reliance, user agency, well-being, durability, and distributional effects.
Reporting frameworks, intervention taxonomies, evaluation cards, datasets, benchmarks, simulation environments, reproducibility, cross-domain comparison, and standardized protocols.
Research agenda
How should causal estimands and study designs be defined when interventions are repeated, users and systems co-adapt, and earlier responses determine later exposures?
How should immediate, sustained, cumulative, delayed, and unintended human effects be measured across application-specific time horizons?
Who determines desirable behavioral outcomes, and how can AI interventions remain effective without undermining autonomy, consent, fairness, or non-manipulation?
What shared reporting standards, datasets, metrics, and evaluation infrastructure are needed for responsible and reproducible comparison?
Domain coverage
Call for submissions
CALE-AI invites contributions investigating how user-facing AI interventions affect human decisions, behavior, preferences, trust, reliance, autonomy, well-being, and other human-centered outcomes.
The workshop welcomes interdisciplinary contributions from artificial intelligence, machine learning, recommender systems, information retrieval, human–computer interaction, behavioral science, causal inference, digital health, education, computational social science, and responsible AI.
Submissions should clearly specify where relevant:
Review: “Each submission will receive at least two reviews.”
Publication: “Accepted contributions will be made available as non-archival workshop papers on the workshop website, subject to author permission.”
Critical milestones
Deadlines use the AAAI-relevant timezone if later confirmed.
Workshop structure
CALE-AI is planned as a one-day, in-person workshop. The program will combine contributed paper presentations, invited talks, lightning talks, an interactive poster session, a moderated panel, structured breakout discussions, and a final plenary discussion.
Approximately half of the workshop time is dedicated to presenting and discussing submitted research.
Expected outputs
A shared taxonomy categorizing interventions based on mechanism, adaptation level, context, intended outcome, and possible unintended effects.
A structured reporting framework covering the intervention, comparator, causal estimand, assumptions, target population, adaptation, time horizon, outcomes, uncertainty, subgroup effects, autonomy, and possible harms.
A community-developed agenda for datasets, benchmarks, longitudinal study designs, causal evaluation, reproducible protocols, and future research infrastructure.
Workshop participants may also be invited to contribute to a collaborative position paper or public post-workshop report.
Program
This indicative agenda is subject to change and does not imply that confirmed speakers or final times are available yet.
Featured contributors
Invited speakers will be announced soon.
Organizing team
Postdoctoral Researcher
University of Oulu, Finland
Responsible AI, recommender systems, longitudinal user behavior, health-aware personalization, human-centered AI evaluation.
Email: Mehrdad.Rostami@oulu.fi
Professor
University of Bergen, Norway
Director, SFI MediaFutures
Responsible AI, behavior-aware recommender systems, personalization, food and health technologies.
Email: Christoph.Trattner@uib.no
Professor
National University of Singapore
Behavioral science, computational social science, causal machine learning, experimental methods.
Email: gaveltri@nus.edu.sg
Professor
University of Oulu, Finland
Artificial intelligence, behavioral analytics, interdisciplinary AI applications.
Email: Mourad.Oussalah@oulu.fi
Review process
Program committee members will be announced soon.
Participation
The workshop is designed for approximately 40–60 participants and welcomes researchers and practitioners interested in evaluating the human effects of AI interventions.
Priority may be given to authors of accepted contributions and participants whose expertise contributes to interdisciplinary discussion.
Presentations must be delivered in person according to AAAI policy. Registered participants may have access to virtual attendance if provided by AAAI. The organizers will use AAAI-provided conferencing infrastructure.
Submit your work
Contact