David Isaksson
WASP has recruited Hector Geffner as Wallenberg Guest Professor in AI. He is currently a researcher at The Catalan Institution for Research and Advanced Studies, ICREA, and a professor at the Department of Information and Communications Technologies (DTIC), Universitat Pompeu Fabra, Barcelona, Spain.
“There are many things about WASP that I find attractive. It’s a generous program that can give your research a larger impact, and it also creates collaboration opportunities that can enrich your research as well”, said professor Geffner.
Professor Geffner will spend part of his time at Linköping University.
The main topic of his research is model-based AI and in particular, planning.
“Planning is thinking before acting. It is how to organise behaviour in order to reach certain goals. In artificial intelligence, this means making a machine autonomously achieve given goals.”
He sees that AI research is now divided between model-based and model-free approaches.
“The excitement in AI over the last few years has been about machine learning and model-free methods that derive behaviour from data. They are not about using predictions to figure out what to do next, but about learning a reactive policy. One big disadvantage of model-free, data-based approaches is that it’s hard to understand their decisions. Data-based systems are basically black boxes.”
Model-based AI approaches, on the other hand, try to “think” and predict what will happen next in order to make decisions. Their shortcoming, in turn, is that the models have been built mostly by hand, something that does not scale up. On the other hand, they offer clear benefits, professor Geffner explains.
“They are more transparent, and we can understand their decisions. They are also more flexible and can be reused for a broad range of goals”.
Professor Geffner considers the integration of data-based and model-based approaches as one of the most central problems in AI.
“Data-based learners and model-based reasoners do not talk to each other right now. A challenge that I’m particularly interested in is closing this gap. One of the most promising ways of doing that is by learning meaningful symbolic models from data. This is what I have been pursuing for the last few years and what I will be pursuing in the near future”.
At the Universitat Pompeu Fabra, Professor Geffner teaches courses on AI and Logic and a more recent course on “Social and Technological Change (STC)”.
“I’m interested in the political and economic context of the developments in AI and technology in general. The central question in my STC course is: ‘How to make technology work for us’. Has technology worked for us in the past? And if so, how?”
His interest in AI started while he was completing his degree in electrical engineering in the early 80s.
“I got interested in chess-playing programs and how they work. I was attracted not so much to the practical uses of artificial intelligence; I’m not very practical myself. Instead, I was fascinated by AI from a philosophical point of view. I wanted and still want to understand human intelligence from a computational point of view”, said professor Geffner.
Published: October 21st, 2019