Siemens Worldwide

Pictures of the Future



Mr. Sebastian Webel
Mr. Sebastian Webel


Mobile: +49 172-7169762

Werner-von-Siemens-Straße 1
80333 Munich

Pictures of the Future
The Magazine for Research and Innovation

Autonomous Systems

Artificial Intelligence: “We Have Verifiable Successes”

Michael May is expert for artificial intelligence at Corporate Technology.

Artificial intelligence is currently one of those topics generating a lot of hype, especially ever since Google software beat one of the best Go players. Siemens has been very active in the area of neural networks for more than 30 years and is already using this technology to autonomously optimize plants and system in industrial practice. Our interview with Michael May of Corporate Technology makes it clear why Siemens certainly hold its own with Google.

Artificial intelligence is like nuclear fusion in the sense that, since the 1960s, people have been predicting that we’ll have a working reactor in 40 years. Except that artificial intelligence is still 15 years away. Is this still true?

May: That figure may still apply to nuclear fusion, but when it comes to artificial intelligence, the circumstances have changed dramatically in recent years. We now have verifiable successes. Systems that used to exist only as prototypes are now being used in production. Take image recognition, for example: In some areas, the recognition rates have reached 98 percent or more. This is comparable to human capabilities.

What is driving this progress?

May: There are several different forces behind these advances. Computing power has multiplied so that neural networks, for example, are much bigger today and can model problems more accurately. The algorithms have also improved so much that they often perform complex data preprocessing work at the same time. In addition, when it comes to processing images, the Internet already offers a huge number of images that can be used to train the software.

AlphaGo from Google is a milestone in the history of machine learning. In March 2016, AlphaGo achieved success, beating the world’s best Go player Lee Sedol.

How would you actually define artificial intelligence?

May: We use this term to refer to algorithm-based computer systems that perform functions for which humans need to use their intelligence. Examples include recognizing texts, driving cars, recognizing images and playing chess. But a machine does not have to think like a human but only perform in similar ways.

How does a system like that work?

May:  There are different approaches to machine learning. We call this deep learning, for example, or reinforcement learning. Without going into too much detail: a neural network, which has links similar to those of the new cells in the brain, is fed a large amount of data and then adapts these links or their strength so that it learns the correct responses to the sample data. This information is transferred to new, unknown data; the system learns to interpret the data and make decisions. Therefore, nothing is permanently programmed. Our deep learning processes work with thousands upon thousands of simulated neurons.

Customers can already receive a service in which artificial intelligence automatically optimizes the operation of gas turbines.

Where does Siemens use artificial intelligence?

May: Our often complex systems have many applications. We are most advanced with gas turbines. Customers can already receive a service in which artificial intelligence automatically optimizes the operation of gas turbines. We are the global leader in this area. After the Google software, Deep Mind, beat a champion in the board game of Go last year, experts have explained that we first need to see whether this can be done outside of a game environment in industrial practice. And we have this already!

Can you give us any other examples?

May: Configuring interlocking systems for new train stations. This is complex because so many different possibilities exist. Our system automatically creates an architecture for the hardware and software that verifiably complies with all safety conditions. Healthineers has a number of developments that give doctors real-time assistance during operations by providing images, for example when inserting heart valves. The software also automatically recognizes all ribs in a CT scan of the thorax and clearly organizes them in a two-dimensional representation.

Even tiny suspicious lesions are revealed by syngo.CT Bone Reading and can be further analyzed in conventional orhogonal display.

What is Siemens Corporate Technology’s role in all this?

May: We are the place where all expertise in the area of artificial intelligence comes together. Our job is to work with the operating units to develop specific business opportunities. We have been active in this field for more than 30 years, and we’ve already accomplished a great deal of optimization work with artificial intelligence: for steelworks and in paper production, to name just a few early successes.

Where do you see the biggest opportunities for artificial intelligence?

May: It is already involved in a great many areas. AI can be used in many places where humans currently contribute their experience and expertise. I already mentioned medical diagnostics, where artificial intelligence helps doctors evaluate thousands of X-ray images. The software saves time because it automatically points out anomalies that the doctor can then examine more closely. Artificial intelligence is a prerequisite for self-driving cars.

How will it change our lives?

May: We are already seeing it in everyday life with our mobile phones. They recognize our voices and then perform tasks. This works pretty well. These kinds of assistants will be everywhere in the future. In private life as well as on the job. Another development will be far closer collaboration with computers and machines.

The idea of artificial intelligence scares some people.

May: Machines will always think differently than do people, and so I see them as assistants and not as a threat. In many of these areas, AI enhances our intelligence without making people superfluous. Also, I do not believe there will be a single machine in the foreseeable future that will combine all the cognitive capabilities of a human being in one place. Indeed, this flexibility in solving problem is what makes humans unique. I believe that AI will enrich our work because it can automate tasks that people find less interesting or which are error-prone, thus giving us more creative freedom.

Interview by Norbert Aschenbrenner
Picture credits: from top: 2. picture Google