Adaptive Purposeful Communication Planning for Collaborative AI

Scientific and societal actors agree that the future of AI and robotics lies in building collaborative AI systems.

Communication is a key enabler for such collaboration: future collaborative AI systems must communicate efficiently to different human users, adapting to their skills, preferences, and knowledge.

This capability, called adaptive purposeful communication planning, is still beyond what is currently possible with intelligent agents and is the primary focus of this project.In this 4-year project, we will address the problem of adaptive communication planning by developing a novel framework that models several users, complete with a set of algorithms to create communication strategies adapted to each user.

The project´s first year will be dedicated to developing models of the human users to ensure a minimum viable product at the end of year 1.

During years 2 and 3, we will incrementally extend this architecture with more complex representations and develop algorithms for adaptive communication planning.

We will evaluate our project formally and empirically on the Hanabi Benchmark, a well-established benchmark for cooperative decision-making where communication planning is essential.This project will lead to a general adaptive communication planning framework that can be implemented on any AI system.

It will significantly advance the state of the art in user-modeling and automated planning and reasoning, and push forward the research frontier in collaborative AI.