AMAI: Aligning Multi-Agent Interactions for Sustainable Food Behavior

Villum Experiment, 2026–2028

logo

Project Summary

One-third of all food produced is wasted—driving unnecessary emissions and climate harm. Despite widespread awareness, simply tracking food waste rarely leads to change. AMAI rethinks how AI can support more sustainable everyday choices.

The project develops a multilingual assistant that uses large language models to detect food waste risks (e.g., expiring high-emission foods) and suggest personalized, persuasive solutions—like recipe ideas tailored to what’s in your fridge. It integrates environmental tools and behavior modeling to go beyond convenience and align with user-defined sustainability goals.

Objectives

Can AI help people reduce food waste by aligning with their long-term values—not just preferences?
How can interactive systems balance persuasive communication with user autonomy?
What is the real-world impact of goal-aligned AI on household behavior?

Approach

A mobile app where users receive daily recipe and action suggestions
Integration of LLMs with environmental impact calculators and food inventory tools
Narrative generation based on user values and behavior predictions
Longitudinal user study with 20–30 households
Evaluation of behavior change, CO₂ reduction, and user trust

Team

Principal Investigator

Daniel Hershcovich
Tenure-Track Assistant Professor, Department of Computer Science, University of Copenhagen

Deliverables

Open-source prototype of the AMAI assistant
Dataset from the year-long behavioral study
Research publications on behavior-aligned AI and sustainable HCI
Design guidelines for persuasive, multilingual food-tech interfaces

Contact

For collaboration or inquiries:
📧 dh@di.ku.dk logo