Houben: ‘AI is going to change design processes’

Speech recognition, diagnostic systems, route planning – there is hardly an area left where artificial intelligence (AI) has not made its appearance. At the end of last year, TU Delft appointed a new Professor of Web Information Systems. Professor Geert-Jan Houben was tasked with developing AI at TU Delft as widely as possible. Delta asked him about his approach.

You started working as a ‘Pro Vice Rector artificial intelligence, data and digitalisation’ on 1 December last year. Is that a very different job from the Professor of Information Systems and Director of Education you did before then?
“Yes, it is. As Director of Education you are within one faculty. This new role is for the whole university. I now try to coordinate things for both education and research, and connect people. In that sense, it is different from the administrative position I held before. Besides that, I was and still am a professor.”

Do you still work as a Professor of Web Information Systems?
“Yes, I will continue to do so one day a week. I also have a research group that I’m responsible for. However, I will not be a Director of Education for long. This role takes two days of my time.”

Do you spend all day Zooming now?
“Yes, I do. I think most people do. My job does not entail spending many hours actually doing research myself. Instead, I spend a lot of time being in contact with people, trying to discover what they do, what others do, and making connections. Pre-corona, that would mean a lot of cups of coffee, but nowadays everything happens online.”

A word about the job title, ‘Pro Vice Rector’. ‘Vice Rector’ I understand, but what does ‘Pro’ mean?
“The name comes from Rector Tim van der Hagen. ‘Pro’ stands for the fact that I am not in the Executive Board, but am mandated by the Executive Board. I act on behalf of the Rector and lead his activities. Pro, in short, means for and on behalf of the Rector.”

I read that your task is to meet TU Delft’s goals in the field of artificial intelligence. I quote: ‘groundbreaking solutions for healthcare, energy transition and security’. What are these goals? Does this job feel like a heavy responsibility?
“Our main goal is to move AI, data, digitalisation in education and research forward. AI is not something you only consider in study rooms, but often by working on solutions from different disciplines. This means that you want to work with others more than before. That is how the area is developing. It means that we want to make sure that we involve the specialists that we have in-house in collaborations.”

Are groundbreaking solutions immediately guaranteed?
“I think our new innovations will come step by step. But my role is mainly to bring people together to make an even greater scientific and social impact with the things we are already doing.

At TU Delft, we distinguish between research in AI and with AI. If you want to design solutions for water management or transport with AI, then it is good to cooperate with people in AI in such a way that you both develop new solutions which have a social impact now, and which will generate new insights and issues for the AI scientists. For example, they may discover along the way that their system is too laborious, too slow or outdated. So, they go back to the lab to see what they can do about it.”

Back to basics for the moment. The term AI, artificial intelligence or machine learning. What exactly do we mean by it?
“You can pick up a lot of different definitions just within TU Delft. An important reason that AI is attracting societal attention is that we can encode a picture of reality in data, and that we now have software that can learn from that data. We can use the insights arising from that data in reality. This is often done through ‘machine learning’.

We used to have very small components in AI, but I think today’s challenge is that you make larger systems with those components, and those systems function within a certain context.”

Is this happening in the 24 AI labs? I got the impression that AI is a panacea that can be applied to all fields. Is that true?
“What does apply to all fields is possible changes in the design process. Traditionally, you start by considering how it could be done, and then you start making something, perhaps as a prototype. Then you assess the characteristics of the prototype and, if necessary, you go back.

What data-driven AI enables us to do is to play that process differently as it allows you to simulate rather than test a prototype. Without having to make a real prototype, you can learn from the simulation whether the desired properties will come out of it or not.”

Can AI replace the role of a physical test?
“It often can. You can use simulation as an alternative to the traditional design process. This phenomenon is worth testing in different places to find out the conditions under which it does and does not work. So, in that sense, it’s interesting for all our engineering disciplines to take an open look at whether it’s realistic to expect AI to be deployable in civil engineering, aerospace engineering or construction.”

Will that mean an acceleration of the design cycle?
“That could certainly mean an acceleration, but there are other implications as well. Designing could become cheaper. You may also come up with information that you wouldn’t have come up with otherwise. And you can collaborate in other ways with prospective users.”

How would that work?
“It could be valuable if a simulation would enable you to enter into discussions with future users. You could quickly show what a building would be like, for example. That could potentially be one outcome of a DAI lab.” (Delft AI lab, Eds.)

Does AI make simulations based on examples?
“The challenge here is to check that you haven’t forgotten anything when putting together the examples. You also make assumptions when putting together a traditional prototype or model, and unexpected things can happen during testing. This is also the case with AI techniques, it’s just that it happens in a different way. That is what we want to test in the DAI labs.”

Does AI have any downsides that we should be wary of?
“I think the questions with this kind of design process are always how I can make reliable designs? How can I carry out the process reliably, and how do I know that the design on the table is all right? These were always the main questions, but we had covered them in our process. This new approach puts these questions back on the table. We need to find out how to create reliability when designing with AI techniques like machine learning.”

I saw that 24 AI labs have been announced, for which 96 PhD students are being recruited. Are there any vacancies?
“We have done the set-up in three tranches of eight AI labs each. The first tranche is as good as finished and all the vacancies have been filled. I expect that we will start recruiting for the second tranche soon, and the third later this calendar year.”

Is there a lot of interest?
“Yes, researchers see AI Labs as a great opportunity to bridge different fields in science and to contribute to developing artificial intelligence in design practice. More and more people think that this is the way to shape AI research. That’s great.”

Is the high level of interest in AI labs the reason that they were created?
“Yes, and there are more applications than places. In that sense, I would also like to see more labs. You can see that the idea is catching on all over TU Delft. There is a lot of interdisciplinary interest in AI, as well as from all the faculties. So, let’s turn that interest into action.”