Skip to main content

ERC Starter grant to fund development of foundational machine learning algorithms for human-AI coordination

Research by Professor Jakob Foerster aims to consolidate Oxford as a key player in developing AI systems that interact smoothly with human users in complex settings

Professor Jakob Foerster, who joined the Department in Autumn 2021, has been awarded an ERC starting grant for the development of novel machine learning algorithms for human-AI coordination. This builds on Professor Foerster’s previous work addressing how Artificial Intelligence (AI) agents can learn to cooperate and communicate with other agents. Most recently Professor Foerster carried out pioneering work in the area of “zero-shot coordination” which lays the foundation for the grant.

The European Research Council (ERC) has yearly calls for proposals covering all scientific fields, which are evaluated by international peer reviewers on the sole basis of excellence. Research funded by the ERC is expected to lead to advances at the frontiers of knowledge and to set a clear and inspirational target for frontier research across Europe.

The 2.3m Euro, 5-year grant will assist Professor Foerster in the development of new foundational machine learning algorithms for human-AI coordination, aiming to set the foundations for AI systems that interact smoothly with human users in complex settings such as mixed-autonomy teams or traffic situations. This could have important applications in situations where humans and robots work alongside each other, such as in warehouses or service settings.

Given the rapid development of AI in recent years, this is a crucial step towards AI technology which augments humans, rather than replacing them in the workplace.

Professor Foerster, pictured left, says, “Current machine learning algorithms are really good at using large scale computing power to beat humans at games, such as poker, chess, or Go. Wouldn’t it be great if we could also use it to help and support humans or more generally allow AI agents to work in teams with humans in complex settings?”

Rather than methods that rely on large datasets, the research addresses this goal from ‘first principles’, by developing novel algorithms which are fundamentally ‘coordination aware’, i.e. resulting in agents that are easy to interpret for human users and are able to interpret humans in turn.

He adds, “This grant will allow me to radically accelerate my research into scalable methods for human-AI coordination. It has the potential to turn Oxford into a key player in this crucial area.”