Young ITU researcher receives EliteForsk travel grant for teaching computers to think strategically

Niels Justesen is awarded a travel grant of DKK 200,000, and plans to travel to both New York and France, to work with some of the leading researchers in computer games and robots.

Digital Design DepartmentResearchartificial intelligencecomputer gamesawards

Thursday, 1 March, Ph.D. student at the IT University, Niels Justesen, will receive a travel grant of DKK 200,000 from EliteForsk, at an award ceremony in the Copenhagen Opera House. His project investigates how you can develop self-learning algorithms that can learn to play complicated strategic computer games, such as StarCraft.

»

Naturally, I am extremely pleased that I have been awarded an EliteForsk travel grant. Both because it is a pat on the back for the work I have already done, but also because it adds motivation, moving forward.

«
“Naturally, I am extremely pleased that I have been awarded an EliteForsk travel grant. Both because it is a pat on the back for the work I have already done, but also because it adds motivation, moving forward,” says Niels Justesen.

Evolutionary algorithms adapt to human actions
Besides ensuring his attendance at a number of international conferences concerning artificial intelligence, Niels Justesen will also use the grant to finance two journeys abroad: a stay in New York, in March, where he will work with a Game Innovation Lab, and a trip to France in the autumn, where he will study how robots that are able to adapt themselves to new environments might combine with the methods used in his own project.

Algorithms already exist, which can learn to master arcade games, by letting the computer play against itself. Part of the self-learning consists of the computer being able to adapt its moves, based on whether previous moves have yielded good or bad results in the game. Yet in complicated games such as StarCraft, a combination of various moves is sometimes needed before an effect is seen – a challenge, which might be solved by making the algorithms less targeted, and more curious to explore the game. Niels Justesen has, for instance, developed evolutionary algorithms, which can plan a combination of moves within the game, and at the same time continuously adapt their strategy based on the moves of the opponent.

The many possibilities, and the uncertainty surrounding the opponent’s moves, makes it difficult for an algorithm to learn the entirety of a game such as StarCraft by, for example, letting the computer play against itself. Instead, one of Niels Justesen’s approaches has been to make the computer imitate real players’ actions, based on a dataset that shows how other players have played in almost 800,000 different game situations. Afterwards, the computer can recreate overall game strategies.

Significance outside the game world
On the face of it, a Ph.D. project about a specific computer game might seem quite specialised, and without any great significance or interest to anyone who does not play the game. Yet it is not unheard of for algorithms originally developed for playing in games to find a use for something very different – such as lowering the power consumption of Google’s data centres. Both Facebook and the Google-owned company DeepMind are interested in the challenges presented by StarCraft. For instance, Facebook has recently won a big artificial intelligence competition, in which computer programs play against each other in StarCraft.

It is clear, in any event, that a future solution of the StarCraft issue will have significance outside of the game world, and will most likely be a result of a cooperation between the researchers and the large companies, which can combine the different aspects that each of them has explored.

“Solving StarCraft would be a milestone a bit like the moon landing. You may not get that much out of being on the moon, but the technology you have developed in the process can be used in many other areas,” says Niels Justesen.