Artificial Intelligence: The Context Revolution
Artificial intelligence has reached the point where it may be set to trigger the next wave of technological disruption. But the deep learning techniques it often relies on are limited by their inability to provide context. Knowledge graphs – graphic representations of relationships – fill the gap. A world of new applications may be the result.
by Sandra Zistl
All of us use search engines every day. Such systems answer questions instantaneously, freeing us from consulting dictionaries, encyclopedias and libraries. In other words, digital technology brings the knowledge of the world to us wherever we are. We’ve also gotten used to the fact that this technology is becoming increasingly multifaceted. For example, it’s completely normal to receive different types of information as search results, and in different formats – i.e. as text, images, and video.
Extracting Data from Silos
Consider the following example: You search the term “Mona Lisa” with Google. The results tell you that the Mona Lisa is a painting by Leonardo da Vinci that is on display in the Louvre Museum, which is in Paris. You also learn that the French call Mona Lisa “La Joconde.” In addition, photos of the painting and the artist are displayed, as are results for videos (“The Secrets of the Mona Lisa,” for example), and a Google Maps window shows you hotels in your area that are called “Mona Lisa.” In other words, the results of your search don’t just include content; increasingly you’re also being provided with context.
There’s no magic behind this. Instead, it’s all made possible by digital knowledge graphs. A graph is defined as a graphic representation of relationships, for example, in the form of marked vertices (also called nodes or points) and connecting lines (also called edges). Digital knowledge graphs work exactly the same way: The graphs extract data from their data silos – i.e. various sources such as 3D models, blueprints, and histories (of machine lifecycles etc.) – and then identify interrelationships within the data. This is what enables them to provide the types of answers that artificial intelligence systems with learning neural networks (deep learning) were previously incapable of delivering.
The Next Wave of Technological Disruption
“Artificial intelligence has reached the point where it may be set to trigger the next wave of technological disruption – regardless of the business sector,” says Michael May, Head of Company Core Technology (CCT), Data Analysis and Artificial Intelligence at Siemens. The Group has completed a digital revolution that now makes it one of the top ten software companies in the world. Siemens currently employs some 200 data scientists and AI experts, who work at nine locations worldwide. May believes that deep learning “is still at the high point of the hype cycle,” but that “it fails when it comes to context. Knowledge graphs can handle context and enable us to address things that deep learning cannot address on its own.”
All of this amounts to a quantum leap for many different types of applications – for example, flexible manufacturing; facility and equipment maintenance; supply chain management as well as the area of advanced diagnostics. The use of knowledge graphs also makes it possible to align services with customer benefits more precisely than ever before.
Tapping Domain Knowledge
Knowledge graphs can be used for just about anything. Siemens, for its part, specializes in industrial knowledge graphs, which are taking Industry 4.0 to the next level. “The immense domain knowledge we have at Siemens is our foundation in this area,” says May. “We’re now tapping into this knowledge step by step and using knowledge graphs to digitally automate expert knowledge.” Siemens researchers have made a name for themselves here in the international research community as well, having, for example, received several awards at the international Semantic Web Conference 2017.
AI has reached the point where it may be set to trigger the next wave of technological disruption - regardless of the business sector.
Among other things, Siemens is using digital knowledge graphs for energy sector applications – more specifically, with gas turbines. AI algorithms are capable of using big data analytics to recognize rules for such applications – e.g. that certain components fail after a certain period of time. However, a system like the knowledge graph makes predictions using information such as the location where the component was manufactured, the temperatures at which it’s used and – particularly important with offshore oil platforms, for example – humidity and salt content in the air. All of this enables more precise predictions to be made. Such systems can be used, for example, to produce valid risk analyses in situations where only generalized calculations could be made in the past – and they also help shorten development times.
New Services for all Business Sectors
This type of data integration and analysis enables the establishment of a huge range of new services in all business sectors. Factories operated by AI systems, autonomous trains, the intelligent management of cities and their energy needs, and cooperation between humans and robots are all on their way to becoming reality. Siemens’ building optimization operations are another example of the great variety of ways in which knowledge graphs can be used. Siemens’ Building Technologies (BT) division is currently using a graph developed in cooperation with Corporate Technology to manage systems at Group headquarters at Wittelsbacherplatz in Munich. The building, which opened two years ago, can be described as very “smart” in terms of its architecture, systems, and energy consumption. It’s equipped with 50 types of sensors.
“Digital knowledge graphs are now extracting information that was previously completely inaccessible,” says Markus Winterholer, who is managing the Wittelsbacherplatz project for BT. For example, it’s now possible to define in a much more precise manner than was previously possible the ways in which individual rooms are used. What this amounts to is knowledge integration at the highest level, which in turn is made possible by the integration of three digital twins into one knowledge graph. One twin has knowledge about the sensors – i.e. the number of sensors and their units of measurement. The second knows the six-story building’s blueprints. The third twin sends live sensor data using the MindSphere cloud platform from Siemens. “This combination of historical and current data gives us an overview of building operations that was impossible to obtain in the past,” Winterholer explains.
These forms of knowledge are valuable for building users, landlords, and facility managers, as they allow them to minimize energy use as well as save and earn money. One example that Winterholer points to here involves new pay-per-use models for cleaning services and climate control systems in buildings. “This knowledge is extremely important for evacuation planning,” he adds. The digital twins make it possible to take such factors into account during the building planning stage – which is actually a new service that BT is offering its customers.
Powerful New Tools
Knowledge graphs are powerful new tools for human users. Their intelligence exceeds that of humans in that they can integrate and evaluate in fractions of a second knowledge that no human brain can possibly contain. Nevertheless, they are designed to assist people. Steffen Lamparter, who manages an AI research group at Siemens CT, says the most important thing to consider when creating a graph is “how the people who want to use it see the world around them. What are their questions, what are their sources of data and what type of experience do they have as experts?” Such an approach enables everyone to learn from the memories of others, as such memories then become retrievable.
“Our goal is to create a digital companion that will help people make better decisions,” says CCT’s May. “Nevertheless, it is still people themselves who will ultimately make the decisions.”
Picture credits: from above: 1. and 3. Picture gettyimage
Mobility, healthcare, energy consumption, logistics – there’s room for improvement in all of these areas. In this interview, Gerhard Weikum, Director at the Max Planck Institute for Informatics, explains how knowledge graphs can help us to make our lives more pleasant.
Pictures of the Future (PoF) You’re doing research in the area of knowledge derived from “unstructured data” – data that do not exist in a database or any other organized structures. What kind of added value can such data provide?
Gerhard Weikum: Unstructured data primarily include texts, ranging from scientific publications to social media. And they are in fact an extremely important source of knowledge for many applications – for example in transportation, logistics, energy, and healthcare. In all of these areas, there are optimization problems at the everyday level. If our goal is to make our lives better, we can’t afford to depend only on the data of machines and systems. Instead, we have to deal with the collection of the data that we human beings are generating. And as we do so, we have to focus more on the background knowledge – in other words, on the context. We’ve reached a point where we depend on such data in order to take the next optimization step in the fields I’ve mentioned.
PoF: Why do we need unstructured data for that?
Weikum: Let me give you an example from the field of energy. It’s not enough to install smart meters in every house. They don’t tell us when people are ready to change their consumption behavior. The same goes for local public transportation. Simply recording traffic volume will not enable me to know under what conditions people will be willing to use carpooling or which route they would travel by bike. In order to know these things, we need background knowledge: How does a person interpret this situation on the basis of its context? When does he or she make decisions that influence his or her behavior? Properly connecting this knowledge with the data we have gained from the systems, and drawing the right conclusions on this basis – that’s the task of knowledge graphs. They put data in context, and they’re already very good at supporting people by offering recommendations. That can be in the form of customized electricity rates or a traffic system that is popular with most of the residents of a city. The potential of knowledge graphs is tremendous. And that’s what we’re working on.
PoF: Supporting people is only one aspect. On the other hand, whenever we optimize processes, jobs are lost.
Weikum: For me there’s no doubt that digitalization is optimizing jobs away, so to speak. However, it also creates new jobs that are more demanding and that require capabilities that a computer doesn’t have. Of course it’s not possible for everyone, at any age, to retrain for every kind of new job. That’s why restructuring our society in this way has to be a long-term process. But after all, this has happened again and again throughout history. Humankind has also coped with the previous industrial revolutions. The reason why we’re talking about Industry 4.0 is that there have already been three industrial revolutions. We’re going to have Industry 5.0 as well, and we’re going to structure it in a positive way. It’s up to us to make sure that artificial intelligence makes our lives better.
Interview: Sandra Zistl
Prof. Gerhard Weikum (53) is a German computer science expert and one of the five directors at the Max Planck Institute for Informatics in Saarbrücken. There he heads the Department of Databases and Information Systems. His fields of research include distributed computer systems, the integration of data and texts, and the automatic creation of knowledge graphs. Gerhard Weikum is one of the originators of the YAGO knowledge base. He has been honored for his contributions to science with a number of awards, including the ACM SIGMOD Edgar F. Codd Innovations Award.
Subscribe to our Newsletter
Stay up to date at all times: everything you need to know about electrification, automation, and digitalization.