Floris Hoogenboom, Head of AI at Royal Schiphol Group, and his team have the goal to make the airport autonomous by 2050. Predicting lines at the gates and security checks, self-driving passenger busses and automated passenger bridges are all part of this project. AI has a key role in this project by forming a bridge between the operation today and what they envision in the coming years.
What is artificial intelligence, according to you?
People will immediately think about predictive technology when you talk about artificial intelligence. At Schiphol, this is only a tiny part of the entire spectrum in which we use AI. For example, we use predictive technology to know when a flight arrives, and it helps us run the operations at the airport. As a next step, we need to make decisions in our airport operations, and there is a role for AI in decision making. This means that the spectrum of prediction models, simulation reinforcement learning and technology that is needed to collect information is all being used to summarize what AI means for us at Schiphol.
How is AI being used at Schiphol?
A good example is our Turnaround Insights project if you’re talking about the insights part. We use a camera at the aircraft stand and the aircraft parking lot at the gate to gain insights into what is happening around the ground processes. That plane lands and needs to be refuelled and restacked with resources like food and drinks. That was always a black-box process. We saw the plane landing, but what happened in the meantime was a bit unclear. What was clear is that 40% of the delays occurred in this time frame. We are now using AI to use computer vision to analyze those camera images and map out what is happening around the aircraft in real-time to act on that and learn from those insights for optimization in the future.
How is the predictive role of AI being used at the airport?
We have the vision to become an autonomous airport by 2050. It is nice to have this vision, but it does not mean that we will turn a switch at the end of 2049 and that the entire airport will be autonomous. We see a step-by-step plan in which we build towards this with AI. The first step is insights, what do we need to achieve this, and how busy will it be at the gate and the departure hall? Another step that we call augmented decision-making (ADM) is about how you will act on these insights. What will you do to make real choices within your operations with this insight? That is an important bridge to eventually achieving to desired autonomy.
The focus in recent years has been on how we can gain insights into what is happening in different fields of the airport. Schiphol is a company that has been running for a long time., and planes landed here long before data was a thing. You can see that data is becoming increasingly important to gain more insights into our processes. That lays a foundation for using those insights to help the decision-makers in operation by making the right choices. Making the right choices is not only about looking at where and when it will be busy at the airport and arise, but also the options to take measures if necessary.
There is a vast amount of data available at Schiphol; how do you decide which data you want to use and which data not to use?
We have three focus areas; our operations are the most interesting because it contributes to the goal of being an autonomous airport by 2050. In addition, we have a lot of assets and infrastructure we need to keep our airport. Finally, we have our passengers and a commercial area for which we use AI. We try to use several data sources because both Schiphol and the airlines use data. Many of the data sources and streams we use do not necessarily belong to Schiphol itself.
In terms of privacy, we must comply with the GDPR just like everyone else. We are, of course, an airport that takes the safety aspect very seriously. However, the element of privacy differs significantly per domain. In the commercial field, we are very aware of the enormous amount of data that we have at our disposal. Still, we realize that we are responsible for what we do and do not want to use commercially.
How do you choose where you will work with machine learning or new algorithms?
In the operations domain, our primary goal is to be autonomous by 2050. We have to run our operations, and in the meantime, we would like to make the step toward augmented decision making. In doing so, we constantly keep in mind that our choices are essential for several airport processes. We use these insights to determine where we will use a particular algorithm, for example, by predicting bottlenecks at the security check. That does not change the fact that you can still have capacity problems, for instance, during the strike of KLM staff a few weeks ago, but sometimes we can predict the effects of these future problems. This is part of our airport operations plan, an integral program in which we use AI to identify, predict and solve bottlenecks in our operations.
What’s your view on the European AI act? Does it offer opportunities for innovation, or is it instead of a barrier?
What is always tricky is that it’s all principle-based legislation. On the one hand, that’s very good, because it achieves a lot and is very broad. On the other hand, it’s sometimes unclear what things exactly mean. We always choose the safe side, but it’s never entirely clear to what extent such legislation gives you the space to implement something. With this act, it’s something that has yet to form itself.
What does your team look like, and what requirements do they have to meet?
We consist of about 45 people working on AI, and we have three types of roles on which we focus. On the one hand, the data engineers are the ones who make the infrastructure to run our models. In addition, we have data scientists and analysts who make the models and can measure whether we add the value that we want to add. The third role we identify is the decision scientist, which is a role you hear less often. With this role, we want to make the step to augmented decision-making to apply AI more easily and successfully. We do this by using that role not always to build very large tooling but by having a role that can be placed close to the operations, and that can translate how the technology can be translated into the processes of the operations.
How can the AI community help you?
We see that the AI community is always very focused on the technical aspect. A few years ago, you had the hype of analytics translator that came up, and I think there’s a big gap in the whole community. We are very good at building models and are also getting better at bringing these models into production. However, little thought has been given to developing these models and how the process transformation should be put together. Consideration should also be given to what types of roles should be developed. As a community, we should look at training technically-minded people and look at the business and implementation side of it. The roles that can combine all this will generate the most value for this sector.
Read more articles on leading Dutch AI experts such as
- How LUMO Labs investor Andy Lürling helps AI-startups and tries to make the world a better place
- How Jim Stolze tracks down bias in algorithms with his new company, de Keuringsdienst van Data
- How Buster Franken educates AI-engineers to solve humanity’s biggest challenges
