In this video Michael describes how intelligent agents are computational systems that reason, plan, learn and interact with each other. This video summarises research done in the Agents Group in Informatics.
At the Agents Group in the School of Informatics, we investigate agent and multiagent technologies that allow us to build intelligent software components that can reason, plan, learn and interact with each other. Examples for such systems are programs that bid on behalf of human users on eBay, robots that play football together (and against each other), or Web sites that mutually enhance their contents by exchanging information with each other.
In our research, we are particularly interested in reasoning about interaction, i.e. deciding how to behave toward others, how to model each other’s point of view, and how to communicate. As an example, consider two robots that disagree on how to perform a task together. How would they try to convince each other and/or to reach a compromise? Or a system of autonomously operating trains which need to avoid collisions while sharing the rail network – what rules should apply when they enter the same stretch of track coming from opposite directions? Or a group of travellers trying to optimise a plan for sharing means of transportation – how can they ensure they don’t go too much out of their way to travel together, while also suggesting a joint plan that keeps others happy?
Our group has developed algorithms for solving such problems, with a particular focus on realism and scalability. For this, we use methods from Artificial Intelligence, such as automated planning and machine learning, but also take inspiration from Economics and other social sciences. Currently, we are looking to expand the range of applications we are interested in to transportation and logistics, social computing, and large-scale distributed machine learning systems.
Find out more: