“We have to decide how we will let AI impact our lives”
12 May 2020
Artificial intelligence is here to stay, and its impact on our lives will continue to expand. With the five-year AI for Research initiative, Uppsala University is taking an interdisciplinary approach to developing this new technology, writes Thomas Schön, professor of control engineering at Uppsala University.
Today there is a lot of discussion about Artificial Intelligence (AI) and machine learning, which are both exciting and positive challenges for those of us working to develop this technology. The reason I began 10 years ago to research machine learning, an important part of AI, was because I simply could not resist. I saw a new field where I could use the mathematics I like so much to solve many more problems than what I previously had access to. The feeling, which has only grown stronger, is that there is an unlimited number of exciting problems to tackle.
When we want to use a computer to do a job, the classic way is to program the computer with defined rules. With machine learning, we let the computer draw its own conclusions, for example, based on what the data looks like or how others have solved the problem. This can require large amounts of data, but the payback is enormous. There are many examples of how machine learning has found new patterns and solved tasks more accurately than humans and traditionally programed computers have been able to do. And machine learning is in no way magical. Under the bonnet it uses learning algorithms that adapt flexible mathematical models to data.
This is a technology that impacts us more than we think, and its impact will likely only increase. Like all technology we develop, it brings with it both good and bad sides. That’s why it is necessary that we grasp the full extent of what we create while we use it to solve the important challenges we are facing.
It is against this background that, after much consideration, I took on the assignment nearly two years ago to consider what Uppsala University should focus on within AI. The assignment was both challenging and inspiring, and after many interdisciplinary discussions, we will soon be launching AI for Research, a five-year initiative where researchers and doctoral students from the entire University will come together in new facilities that are just being completed at Carolina Rediviva.
The first thing many people think about when they hear AI for Research is the application of AI within other fields. And certainly, we will pick some low hanging fruit like that. But the backbone of our strategy is a structured and productive interaction between basic research and applied research.
The key is identifying parallel projects in different teams that share the sense that the project is genuinely relevant and that share the commitment to realise the project. The groups able to identify these projects are well positioned to help each other move forward. Subsequently, we need links between the projects and the groups, something we will create through physically moving one or more individuals between the teams.
A concrete example is the collaboration we have created in the last two years with cardiologists in Brazil. Researchers like myself in machine learning are passionate about new mathematical models and algorithms that increase the ability of computers to handle processes that change over time, like our heart. The cardiologists, on the other hand, are passionate about creating better care and increasing medical knowledge about the heart using unique data they have collected over the last 10 years. The collaboration was possible because a Brazilian doctoral student spent about a year in our group in Uppsala, which revealed natural and exciting connections between our separate groups. This has also led to interesting and productive collaboration with heart researchers here in Uppsala.
My vision is that AI for Research will contribute to parallel and collaborative projects that continue to develop at the department and revitalise both research and teaching at Uppsala University. This assumes mutual trust and understanding, which is one of our important tasks. As a wise person once said, I believe that if humans spend more time trying to understand each other instead of trying to be understood, then we will make much more progress in a much more pleasant way. Something that likely applies to research within AI as much as it does to all types of human interaction.
My hope is that AI for Research will help us to reflect wisely over how this technology impacts us as a species today, how we want it to impact us in the future and to contribute to creating the right type of technology. There is no doubt that what we develop will have an impact on us. The question is simply how we want that to take place.
Professor of Automatic Control
Department of Information Technology
For more information