Artificial intelligence in support of humankind

The computer systems can serve as a support for us humans, since they are so good at handling large amounts of data and recognising patterns.

The computer systems can serve as a support for us humans, since they are so good at handling large amounts of data and recognising patterns.

Artificial intelligence is on the way in many areas. Specially trained computer systems are used in cars, for example, and as support for doctors making a diagnosis. These machines are especially good at pattern recognition – but we don’t yet have to worry about them becoming smarter than us.

Thomas Schön, Professor of Systems and Control, conducts research on machine learning, a field in rapid development in recent years. Today, computers or machines can automatically learn to handle a situation without a person having programmed the computer for this specific task.

“This has an enormous number of areas of use. In the auto industry, there is a lot already in production, such as adaptive cruise control and recognising pedestrians in the dark. There is a lot already being used today and the future looks bright.”
There are many possibilities, but we hardly need to worry about computers taking over in the immediate future, according to Schön.
“When we look at a car that autonomously handles a situation, it may seem smart in some sense, but if we break it down and look at the components, there is nothing I would exactly call intelligent in it. In terms of engineering it’s very elegant to build these systems, but it is not any kind of intelligence in the sense that we humans use the word,” says Schön.

However, the computer systems can serve as a support for us humans, since they are so good at handling large amounts of data and recognising patterns.

Recently, Schön has been working on medical applications, for example in cooperation with a hospital in Brazil that has many patients in areas that are geographically difficult to access. The researchers developed mathematical models, algorithms, to interpret ECG results.

“We have preliminary evaluations that look promising, where our algorithm provides a prediction of six of the most common problems that is on a par with the doctors we compared with.”

To train it, the algorithm got to see 2.3 million ECG measurements together with information from doctors about how they can be interpreted. Then the computer built up a model in which it can be fed an ECG and provide an assessment.

“I am convinced that this technology will help out in healthcare, but it will in no way replace doctors, not in the next 200 years. But if we can support doctors and remove certain cognitive elements, where machines have proven to work well, I think it’s excellent. By using this technology, we can free up time for other tasks.”

Thomas Schön’s research focus is algorithms for systems that can handle uncertainty.
Photo: Mikael Wallerstedt

Another example of when machine learning can save lives is a collaboration with a Swedish company (Elekta) that builds radiotherapy equipment for cancer treatment. They will initially build 15 machines around the world, of which one is here in Uppsala, where the doctor can use MRI to look at the tumour while it is being irradiated.

“Just being able to look at the tumour paves the way for even greater precision. What we are doing here is automating a part, a kind of self-driving radiation treatment, where the computer works out the radiation dose and changes direction based on what it sees in the MRI. The hope is that we will be able to focus more on the tumour and kill significantly less of the healthy tissue surrounding it.”

In this case, the algorithm needs to be able to locate the tumour and follow it. For example, we move when we breathe, even if we try to lie still. The algorithm can compensate for this so that the radiation can follow our breathing and maintain full focus on the tumour.

“This is very new for us, we have just got started. It feels very exciting and very rewarding to try to use my knowledge to fight cancer.”

Schön’s research focus is mathematical models, or algorithms, for systems that can handle uncertainty.

“We humans handle uncertainty to varying degrees when making all of our decisions. It’s the same in an advanced technical system. The measurements it gets are uncertain. If one can work with that uncertainty mathematically, a better decision or higher performance can be achieved.”

In cars, increasing numbers of systems are being developed that help drivers in various ways, such as adaptive cruise control or systems that warn of pedestrians when it is dark out. The technology advances further year by year, with more and more functions, until we might reach the self-driving car. People’s acceptance of this is growing bit by bit, according to Schön.

“These systems are already very reliable today. Nobody would even consider buying a car without an ABS system, which has been on the market for nearly 40 years now. But when they were being introduced, there was a similar discussion as today – ‘how will this work?’ The insurance companies’ statistics show how many lives this has saved, and how many accidents have been prevented.”

Annica Hulth

Subscribe to the Uppsala University newsletter