In an interview, artificial intelligence specialist Alexander Motzek explains how machine learning can reduce companies’ workloads, why AI still has a long way to go when it comes to understanding images and what magic has to do with it all.
Mr. Motzek, “artificial intelligence specialist” sounds futuristic. Is that also part of the appeal?
Yes, there are new research findings in AI almost every day that can be immediately tested and applied. In no other field has real-world application ever been as closely related to research, and in no other have problems – some of them long-standing – been as fast and easy to solve.
Could you give us an example?
Customer support teams receive thousands of e-mails every day that employees then have to manually forward to the proper contact person. That takes a lot of time. AI is now capable of fully performing this task. Artificial intelligence reduces the amount of work people have, allowing employees to focus on more important things. It’s exciting to see how my work makes processes more efficient. It’s also a lot of fun to show others who only know AI from the media what it is already capable of today. AI is a bit like magic, but with mathematics rather than tricks.
That’s a nice way of putting it. Still, both things are impossible for some people to understand. Your job isn’t over once you’ve implemented AI at a company, right?
No. As artificial intelligence specialists at Lufthansa Industry Solutions, we not only use artificial intelligence in a wide range of sectors and at various different companies, we also constantly adapt and train AI. Ultimately, our aim is to consistently improve processes and constantly make them leaner.
“How do I stay up to date? AI provides me with the latest research findings, of course.”Alexander Motzek, artificial intelligence specialist
That sounds as if the point of AI were to transform manual processes into digital ones. Is that the core?
Yes, I would describe the use of AI as a transition from manual processes to methods based on bits and bytes. One term that comes to mind is “machine learning.” As a result of machine learning, people no longer have to spend time and effort reading data from machines and devices in industrial environments, for example. Measuring instruments independently transmit data to an IoT platform, where they are automatically analyzed by a software program and compiled into status reports. Machine learning allows companies to proactively maintain equipment, thereby preventing them from breaking down. Thanks to digital transformation and all its new technologies, such as artificial intelligence, many decisions will no longer be made intuitively and subjectively by people, but rather on the basis of objective methods.
And what does that mean for consumers?
Artificial intelligence is behind the digital equivalent of a concierge, for example. The digital version is there for you 24 hours a day, seven days a week, and provides support for a wide range of needs, such as booking tickets or canceling subscriptions, saving you the trouble of having to write a separate e-mail. The concierge provides a digital service that does not require the service provider to have to rely on a live human being.
Do scenarios such as this illustrate the development potential of AI in the next five years?
You can already do a lot with AI today. Artificial intelligence can already be used to precisely categorize an image and interpret it with regard to highly complex information in medical imaging to help spot tumors early, for example, or in construction to support quality assurance or material testing. Until now, you needed specialists with years of experience to do that. AI makes many things easier and faster, which is good for everyone involved.
However, AI still has a long way to go when it comes to understanding images in general. When a person sees a picture of a tree, he or she can then identify other trees as such based on this one example. AI can’t do that yet. Instead, it needs a large number of examples. There’s still a lot of potential for development in image research.
So we’re still going to be needed for a while?
Yes, most definitely.
As an artificial intelligence specialist, what is your exact approach to tackling a problem? Do you have any favorite algorithms?
The main thing we do is to solve the underlying business problem. AI specialists can make use of a dauntingly wide range of options, of course, but that’s part of the appeal.
In our day-to-day work, we use existing possibilities depending on which ones do the best job of solving the problem. My favorite in image processing is transfer learning from pretrained models, such as from a ResNet trained on ImageNet that is easy to apply using Keras and PyTorch. In text processing, simple model assumptions such as traditional probabilistic graphical models can have a big impact. But that may not be enough in some cases. Then you have to perform intensive research by looking through recent papers and journals and reading up on the latest research findings.
How do you maintain an overview with the constant flow of new things to read?
It makes things simpler that research and industry are more closely connected in AI than in any other field. I naturally keep a close eye on the latest developments. Major global conferences such as the International Joint Conference on Artificial Intelligence or the Conference on Neural Information Processing Systems play a role here.
The sheer volume of papers is overwhelming indeed. But AI helps me select the right papers, blogs and websites, of course. My Google Newsfeed provides me with the latest and most important articles on machine learning and deep learning on a regular basis.
And what kind of role does exchange with colleagues play?
We regularly discuss the latest trends at our AI Lab so as to improve our ability to evaluate research findings. The lab is a company-wide community of colleagues who are interested in AI and who all have both theoretical and practical experience. We artificial intelligence specialists benefit from this network, as do all employees – and ultimately our customers. That’s because our job is to solve their problems through AI and machine learning.
About Alexander Motzek
Alexander Motzek joined Lufthansa Industry Solutions in Norderstedt as an artificial intelligence specialist in February 2017. After completing a cooperative course of studies at Hamburg University of Applied Sciences culminating in qualification as an electronics technician for automation technology and a bachelor of engineering in computer science and electrical engineering, Motzek earned a master’s degree in computer science and engineering at the Hamburg University of Technology before writing his doctoral dissertation on probabilistic graphical models at the Universität zu Lübeck.