Thomas Schön: Now machines are learning to contribute to a better world

His office, housed in one of Uppsala’s former regiment buildings, is anything but futuristic. A computer in the corner is the only indication that we have entered the 21st century. Still, we are at the centre of one of the world’s leading research environments in machine learning. Systems developed here are rapidly bringing science closer to intergenerational utopias such as self-driving cars and computer-controlled cancer operations.

“We work with algorithms. Great and pretty from an engineering standpoint, but not intrinsically smart in any way,” notes Thomas Schön, professor of automatic control at Uppsala University. “They can be compared with food recipes, a set of ingredients that, in the right combination, generate a specific result – a cinnamon bun or a vehicle capable of recognising pedestrians. As I see it, the artificial intelligence of our time, AI, is not even related to traditional intelligence and the ability to achieve complex goals. But having said that, we have the privilege of working in a fantastic field where development continues to accelerate.”

His words reflect a serenity that not everyone shares. Under the headline “Swedish working group against killer robots”, a nationwide morning newspaper recently reported that two Swedish cabinet ministers want to establish a group to review the development of artificially intelligent machines that can make their own decisions to kill people in conflicts. All colourfully illustrated with two heavily armed “robots with the right to kill” borrowed from the Terminator science fiction film. An effective framing of the article, but how relevant is it really as a future scenario?

“We humans tend to be afraid of the unknown, which leads to polarization concerning AI, which if we believe the media, will either save or destroy the world. The film industry, for its part, gladly plays on our fears about a form of intelligence that we do not fully understand. Of course, we should respect this new technology and be extra vigilant about certain fields of application, such as autonomous weapons systems, but I cannot possibly summon any concern that machines would gain power over people.”

Since 2017 Swedish basic research in artificial intelligence and autonomous systems has been brought together in Sweden’s largest individual research programme ever, the billion-kronor Wallenberg AI, Autonomous Systems and Software Program (WASP). The situation has changed radically since September 2013, when Thomas Schön arrived at Uppsala University. His contract included funding for two doctoral students, and aside from them, Schön remained an exotic feature in the national arena.

“At conferences abroad I could run into an occasional Swede in American corporate exile. Here at home I was extremely isolated before things finally fell into place. Today I head a research team with 20 members whose results are in demand around the world, which also represents the scientific contribution I am most proud of. Now, the time feels right to set our sights on our own turf, and right now we are putting the finishing touches on an initiative that I am convinced will take Uppsala University’s already strong position further in the right direction.”

The multidisciplinary initiative is called AI for the Sciences. Researchers at all of Uppsala University’s faculties will be invited to meet for a year in a common arena focused on artificial intelligence. The goal is to draw from the University’s collective knowledge and perspective to generate ideas about new areas where machine learning can produce added value in the future.

“The most inspiring aspects of my work include the freedom to talk to whomever I want and to focus on challenges with great social relevance. Safer vehicles and better health care already save lives, but in several fields we have barely begun to reflect on the opportunities technology offers. If everything goes according to plan, we will launch the project in 2020, which feels extremely exciting. Our research presupposes work without limiting assumptions, and we already see several interesting areas before us where machine learning can contribute to a better world.”