UT professor sees opportunities and limitations for Artificial Intelligence

Show a child a dog three times and a cat four times, and it will remember what they are for the rest of their lives. A computer is not that smart yet. Before it can make the distinction, it will need thousands of pictures of dogs and cats. “If you want to teach a computer to make a distinction between one object and another, you have to feed it with many data. The distinction between a dog or a cat is not too relevant. Still, if you want to teach a computer to drive a self-driving car, the distinction between a child or a rabbit is of vital importance. Our brains are much more efficient in image recognition. For example, a computer gets confused when there are multiple objects on a picture; our brains automatically filter that information.” 

The doom scenario of computers taking over humanity is, therefore, out of the question right now. Schmidt-Hieber even calls it “nonsense”. Artificial Intelligence has already beaten the best Go-player in the world, a game played by thousands of people worldwide, also professionally. The game involves a degree of intuition, so that is smarter than expected. A student from my group had a computer come up with new dishes, based on many existing recipes. Computers can make paintings based on the style of certain painters. But it is always a derivative of what already exists. Inventing an own style is still a bridge too far.”

However, the professor certainly does not want to be derogatory about the importance of AI. “Much data is available at hospitals. With Artificial Intelligence, you would be able to link that data and compare all experiences from the past. An example could be the treatment of cancer, choosing the most successful combination of, for example, medication and radiation. Another example: radiologists now decide whether someone has cancer, based on a scan. The computer can perhaps provide a second opinion. This would exclude the chance that a tumour is overlooked.”

Schmidt-Hieber is considered an expert in the field of AI, and especially about the mathematical side. “Computer science has been working on this for a lot longer, especially on applications. From the mathematical side, and my field of statistics, I focus on how we can interpret the results. With new mathematical techniques, you can analyse which methods work, and which do not, and why. How large should the neural network be to ensure the best functioning of the method? We have not yet discovered how the human brain works. It is easier to discover the functioning of computers. It has long been an underexposed aspect; everything was focused on applications. To move forward, we will also have to dive into the operations mathematically.

The young professor gives workshops and conferences, especially for academics. Not only the academic world is interested, so are the research departments of Google and Microsoft and large companies like Toyota. Given the stormy developments, he is an advocate of having students become acquainted with AI and “deep learning” early on in education.  “They are the ones that have to work with it later, so it must be integrated in education early on.” 

The University of Twente is happy to work with (regional) businesses on this topic. “There is much expertise in the Data Science Lab. We would like to get into contact with companies that want to get started with his, but naturally do not have the facilities that we can offer. By sharing their problems, we can help think of a solution. On the other hand, this gives us a good insight into the struggles of daily practice, which in turn provides input for research.”