AMAI

Villum Experiment · 2026–2028

Less food waste.
More thoughtful AI.

We explore how multilingual AI assistants can help households waste less food and make everyday choices that reflect their sustainability goals.

Explore the research Our publications

AMAI · Aligning Multi-Agent Interactions
for Sustainable Food Behavior

Illustration of a mobile food assistant offering ideas beside an open fridge
Everyday choices. Long-term values.
Research setting
University of Copenhagen
Programme
Villum Experiment
Project period
2026–2028
Focus
AI & sustainable food behaviour

01 / The research

Better choices start
with better support.

Awareness of food waste does not always lead to change. AMAI investigates how AI can turn sustainability goals into practical support for everyday life.

The project combines large language models, environmental tools and behaviour modelling to identify food waste risks and suggest personalised actions, from using up ingredients to finding a recipe for what is already in the fridge.

Align with values

Can AI help people reduce food waste by aligning with their long-term values, beyond immediate preferences?

Respect autonomy

How can interactive systems balance persuasive communication with people’s freedom to make their own choices?

Study everyday impact

What is the real-world impact of goal-aligned AI on household behaviour, emissions and user trust?

02 / Our approach

From household context
to meaningful action.

Connecting language technology, environmental impact and human behaviour.

  1. Understand

    Start with the household

    Combine food inventory tools and environmental impact calculators with user-defined values and behaviour modelling.

  2. Support

    Make the next step easier

    Develop a multilingual mobile assistant with daily recipe and action suggestions, using narratives tailored to users’ goals.

  3. Evaluate

    Learn from everyday use

    Plan a longitudinal study with 20–30 households to evaluate behaviour change, CO₂ reduction and user trust.

What we aim to share

  • Open-source assistant prototype
  • Dataset from the year-long study
  • Research on behaviour-aligned AI
  • Design guidelines for multilingual food-tech interfaces

03 / People

Meet the team.

Based at the Department of Computer Science, University of Copenhagen.

Postdoctoral researcher

Yuchen Yang

Postdoc

MSc student

Ruimin Huang

Master’s research

Visiting MSc student

Eijiro Mochida

Visiting master’s research

04 / Research outputs

Publications & presentations.

  1. 2026Workshop paper

    Diagnosing Empirical Target Retention in Multi-Turn LLM Scenario Construction for Household Food Waste

    Yuchen Yang, Amogh Raina, Eijiro Mochida, and Daniel Hershcovich

    NeurIPS 2026 Workshop on Grounded User Simulation for Model Evaluation and Training: Diversity, Fidelity, and Validity

    OpenReview Paper PDF
  2. 2026Poster presentation

    Realistic Interactive Consumer Simulation Grounded in Household Evidence for Food-Waste Behavior-Change Assistance

    Ruimin Huang, Yuchen Yang, and Daniel Hershcovich

    CS2Nordics 2026

    Poster PDF Conference programme

Let’s connect

Interested in the research?

For collaboration and project enquiries, get in touch with Daniel Hershcovich.

dh@di.ku.dk