“From an AI perspective, the assembly line is a microcosm.”

With Matthias Schulz, Manager Collaboration and Talent Solutions at IBM, we talked about AI’s understanding of ethics and the question of how to really talk to an AI at all. In the second part of the interview, we accompany IBM Watson on the Industry 4.0 floor of the Future Factory.

WYZE: Mr. Schulz, in the industry, the term “Right Data” is circulating – that is, the desire to obtain really important information at the right time using software tools from the mass of available data. Does an AI like Watson simplify this process?

Matthias Schulz: Of course, especially since developments in the Internet of Things and in the industry 4.0 and digitalization are fueling each other on when it comes to the use of cognitive components. At agricultural machinery manufacturer John Deere, for example, the company manufactures tractors in a “Future Factory” using a Smart Manufacturing Platform (SMP) that extends over three levels – Edge, Plant and Cloud.

The Watson API (Application Programming Interface) now takes over the “cognitive” part. For example, it is trained to automatically detect components in the shop floor – in a wide variety of lighting and work situations. With this know-how, Watson assists employees in their work steps and draws their attention to material or work errors, for example. This ‘trains’ the employee’s skills in various tasks and, at the same time, reduces the scrap and error rate of the entire production.

WYZE: But to what extent does the software act “intelligently” here? The sequence you describe sounds like a finer, digital variant of automation. But surely an AI-Assistant has to do even more?

Matthias Schulz: In fact, this is exactly what distinguishes automation from AI: the communicative interaction between humans and software. Imagine, for example, the situation where a machine suddenly stops. Today, the employee usually checks the error code on a screen or manually; if the problem cannot be solved, another colleague is involved later on.

WYZE: How does ‘Colleague AI’ help in such a standard situation?

Matthias Schulz: Our idea is that a chatbot as an assistant is the first point of contact. “Watson, why has the machine stopped?” – John Deere’s Future Factory employee starts the conversation with Watson so easily today. However, AI will not only answer with a list of first standard solution proposals, but offers to “look at” the problem itself. This means that the employee can send a picture of the machine to Watson via smartphone.

WYZE: Does Watson look at photos?

Matthias Schulz: If you like, yes. And then it compares these photos with the image information it “knows” from the working machine and answers its human colleague accordingly: “It looks as if the rotating arm is heavily dirty – can you clean it and then restart the machine? Depending on the fault analysis, AI provides the appropriate repair instructions and recommendations so that the problem does not occur again in the future.

In the context of industry 4.0, this is advantageous in two aspects: first, man and machine learn from each other, identify and reduce standard situations where the processes in the value stream fail or are too complicated. And, secondly, in combination with predictive maintenance, of course, the point of time shifts further and further back, where a machine problem binds multiple human resources.

WYZE: For predictive maintenance you need more information than a photo-shooting of machines and components…

Matthias Schulz: That’s right – this is where “Industrial IoT” comes into play. To put it very simply, the possibility of being able to monitor the interior of a machine with sensors and cameras is ever better. This is precisely the “continuous” networking of cyber-physical systems that Industry 4.0 is targeting.

But even if you can monitor every single gear wheel, have every raw material error reported by the machine and forecast every potential inhibition threshold in the material flow, this does not in itself create added value, but only a gigantic amount of information and information interfaces. This is exactly the “Babylonian data confusion” that companies rightly fear so much today.

WYZE: What, in turn, calls AI to the plan as an ordering and integrative authority?

Matthias Schulz: But there is more at stake than just structure and order. This “train-the-trainer” situation, which we have already described, in which employees and Watson learn from each other “without limits”, is completely different from a “simple” automated, digital design of learning and work processes.

An AI like Watson is, in a way, familiar with the entire production system and not only learns from new error situations, but also independently updates the knowledge pool of guidelines and standard situations on the optimal machine status – or when a new machine generation is used. In conjunction with chatbot communication and image recognition, this is a completely different, much more efficient level of performance than the “If X, then Y” logic of simple, standardized learning programs.

WYZE: So, AI turns out to be some kind of ideal pedagogical director for the operational level in the industry?

Matthias Schulz: I wouldn’t put it like that. It would be wrong to reduce the potential for change in this cooperation between AI and humans for a more efficient value stream and product lifecycle management. This may be the motivation to drive the development of AI in the industry.

WYZE: What changes are you thinking about?

Matthias Schulz: From an AI perspective, the assembly line is a microcosm. Especially in comparison to the cultural, organizational and ethical changes that are emerging on the horizon with this development: Is a “Code of Conduct” or “AI Governance” necessary for human-machine cooperation and, if so, who bears the responsibility – the AI manufacturer or the company? Can I rent a particularly advanced AI as a “consultant” to industrial companies with other products but similar processes in the sense of a new business model – and how do I protect data protection? How do you insure damages that occur because you have followed or not followed the advice of the machine? You should not lose sight of this “penetrating power” of AI.

WYZE: Especially since in this case we are also quickly talking about other AI applications in the company, for example in the finance or HR sector…

Matthias Schulz: Right – but in which other rules apply. When it comes to preparing a purchase decision for a certain component, AI can be used on the basis of quite clear, objective criteria: Material properties, size and weight, and prices.

Now, however, you have a software algorithm assessing how well an applicant for a management position fits into your company – that’s a huge difference. In this case, AI could base its recommendation, for example, on the team constellations, industries and periods in which the manager worked. Even if a great deal of information is made available, this is by no means enough to replace human judgement, let alone to surpass it. AI fails because of “predictive behaviour”.

WYZE: Thank you very much for the interview.

For more information about IBM Watson and John Deere’s Future Factory, visit this link.
This online article at 
“Computerworld” provides an overview of current AI variants.