April 24, 2018

“AI: We are only at the beginning of the development"


				
					Niklaus H. Waser

Niklaus H. Waser, Head of Watson IoT Global Ecosystem at IBM, explains why predictive maintenance is a great application for the use of artificial intelligence (AI), where the AI journey is still going and why lawmakers are still required.
Source: IBM

What are the first AI fruits that industrial companies can reap?

Predictive maintenance is without a doubt one of them. In the context of industrial production, it generally helps to link unstructured data, such as images or noises, with structured data from machines. It's precisely this combination that will result in new insights.

How far is the industry already on the topic of predictive maintenance?

Predictive maintenance has still not really arrived in factory halls. Currently, it is a question of condition-based maintenance in most companies, i.e., condition-dependent maintenance based on the evaluation of machine data. Consequently, predictive maintenance would be the next step.

But you are already talking about prescriptive maintenance at IBM. What is meant by that?

Prescriptive maintenance goes even further. The idea behind it is much more complex. In the case of prescriptive maintenance, it is no longer only a question of in which time period maintenance is required or something has to be replaced, but also a precise calculation of which time window is the optimum economic one. An example: Does it make more sense when there are problems with a ship’s engine to repair it? Or should you go easy on it and sail more slowly? Or is it better to get the last possible use out of the engine to have the goods arrive on time at the destination? All three options are potentially related with costs and risks; the question is which option makes the most sense in the specific situation.

Apart from predictive maintenance: Where are the main potential uses for AI?

Many areas have enormous potential. We are only at the beginning of the development anyway. But one thing is clear: The core competencies of AI technologies – or better of cognitive systems – are speech recognition and dialog capability. In addition, there are possibilities to process enormous amounts of structured and unstructured data, thereby recognizing patterns and producing correlations. The AI systems learn from interactions and experiences and are continually improving. Based on these properties, these technologies can provide guidance for optimizing production processes or maintenance; they can find errors and strike an alarm when specific machine problems occur.

Is AI an integral part of machines and robots?

In our opinion, yes. There are several reasons for this: On the one hand, AI is becoming affordable as a cloud service, and the quality of machine learning (ML) algorithms is improving rapidly. In addition, data availability has increased enormously. It is also noticeable that many systems are provided with standardized interfaces today and consequently can also be combined considerably more easily with intelligent services and applications.

What about legal issues? Who is liable when a learning robot that has taught himself an activity causes damage?

In the case of liability for the faulty behavior of an intelligent robot or vehicle, there is still a need for industry-overlapping action. The robot or AI does not have a legal personality pursuant to current law and therefore cannot participate independently in legal proceedings. In the same way that a robot cannot conclude effective contracts, it cannot be liable for damages that it causes. In the case of damage, it's complicated.

Why?

For example, the injured party could assert claims against the manufacturer of the robot or the AI software. Likewise, however, he could also assert claims against the owner of the robot or those who were responsible for the operation in the specific case – not to forget those who provided the information by means of which the robot taught itself the activity. The search for the cause and consequently the person responsible would often be very difficult and lengthy in such situations – to the annoyance of all parties involved. Therefore, it is the responsibility of lawmakers to create a timely and practicable solution here.

IBM is advancing the topic of artificial intelligence as Watson IoT. Watson became known as a Jeopardy-playing supercomputer in 2011. What exactly is Watson?

Watson is not a supercomputer, not an omniscient superbrain, but a modularly structured cloud platform equipped with diverse AI-based services. Companies, organizations and individual users can access these services from the cloud via interfaces. With these APIs, services can be integrated into their own systems without great effort to solve individual tasks. Among other things, these are speech, image or text analyses, translation services or conversation aids. These diverse services are based on algorithms that are cognitive, i.e., are capable of learning in the broadest sense, and which are individually trained in the context of their use.

What role does Watson play in IBM's IoT offer?

The Watson IoT platform uses various Watson services to process unstructured data, which – besides structured (machine) data – can be handled by the Internet of Things. For example, these are images or noises which are converted into structured data with the aid of artificial neural networks or deep learning and thus enable additional insights. These can then be used to better control processes or optimize maintenance, keyword predictive maintenance.

Tip: The topic of artificial intelligence will also play a major role at the automatica forum, especially on Thursday, June 21, 2018: In addition to a panel discussion (“Artificial Intelligence: from Service Robotics to Smart Production. How Intelligent Can Machines Become?"), there will exciting talks. These include Professor Dr. Sabina Jesschke, Executive Board of German Railway for Digitalization and Technology (“From the Internet of Production to Rail 4.0“), and Günter Kruth, Division Head of Industry 4.0 Smart Operations, Daimler AG (“AI in Automobile Production”). In addition, Eduard Saller, Data Scientist Innovation Lab, BMW Group, and Kim Dressendörfer, Watson Cognitive Solution Architect, IBM Watson IoT Center, will talk about “Cognitive Intelligence in Automobile Production”.