The 5 biggest pitfalls in the introduction of AI in industrial production

The age of AI has begun 10 years ago...

Oct 2020
Michael Welsch

… and most companies have started to use AI in a productive way. Although some experts still consider AI technology as an underdeveloped technology in their respective domain. The argument that AI does not work reliably compared to human expertise and established procedures brings us straight to the first pitfall.

1. Consider AI as an add-on only

Those who want to use AI technology to improve existing systems hope to achieve this without making changes to the actual system, machine or IT. This is understandable, but the full potential of AI is basically nipped in the bud. The introduction of AI technology should always be associated with the goal of automating human work. Only then can the introduction of AI be calculated; the investment risk can be reasonably assessed and changes to existing processes can be justified. Experiences for this can be taken from the technical automation of processes with robots and the associated technical risks. The introduction is not always successful here either. If you only want to introduce assistance to ease the workload of existing processes, you usually have a problem with the return-of-investment calculation and thus too weak a basis for argumentation to achieve something beyond the POC phase and hold out when the first difficulties arise.

2. An AI-First Strategy

Those who do not maneuver themselves into the add-on pitfall and, conversely, want to approach the topic of AI strategically, e.g. by setting up their own department, create a problem at the other extreme. The development of AI technology is never an end in itself. It is usually used in domains where previous approaches have not been successful. Under certain circumstances, AI can even make it possible to solve problems that one thought could not be automated. So you should take your time to understand what you want to achieve with AI technology. These goals can be very individual, depending on the industry and company. To make matters worse, the technology is developing so fast that it takes a lot of time to stay on top of it. A separate department of two or three people is not suitable for strategic development in the sense of a “doing” decision, but for project management and the acquisition of skills to work with AI vendors at eye level in the sense of a “buying” decision.

3. This (one) AI approach (does not) work for us

AI technology, especially algorithms and learning methods have become an almost unmanageable field of research and application. We often hear that “this AI thing” has been tried before. But there is not this one AI. It is rather a large technology park. You can compare it with a workshop. Sometimes you just have to drill a hole in a sheet of metal to fix something. Well off is the one who does not have to explore or even build the workshop first but has someone who can solve his problem professionally. He needs someone who does not turn the desire for a solid fastening into a rocket engine and who sees this as an opportunity for open-ended research cooperation, as university research institutes often do. On the other hand, you also have a problem if someone like a start-up only knows how to use a hammer, because he will see a nail in every problem.

Someone who is an expert for robots only, will want to use a robot everywhere. The same is true for algorithms and IT. So if you have a problem, and the potential partner only knows Computer Vision and Python, he will always want to set up a camera, label very expensive data, and train neural networks with Python like he always does. On the other hand, if you have a background in statistics and are not familiar with mainframe computers, you will want to avoid image processing and do stunts with machine data in Excel rather than getting involved in image processing.

Check with possible “workshop” partners how universal their workshop is, which makes much more sense than looking for someone who has used the desired niche solution several times. You will also find it difficult to google a hole-drill-fixing specialist. Google’s AI is not that good at understanding what you are looking for.

4. Data first, then algorithm

In order to be able to apply algorithms and determine the best method, data is needed. However, the algorithm determines which data is needed. In the worst case, you can’t even tell what is causing the failure. Simple example: do you need camera images and 3D scan images for product control?

To avoid this problem, you should take a holistic approach to data acquisition and algorithm selection. The working is always the combination of data and algorithm. If you don’t find someone who takes the overall responsibility, there is a high risk that the project will fail, and you won’t be able to figure out exactly what the problem is in the end. The algorithm developers will say that the data is not suitable, the data providers will say that the algorithm is obviously not working. Typical example: You already have data and in the end you realize that the essential data was not even there to achieve something. So first there is not data or algorithm, but a planning of both.

5. Failing to take corporate policy and people with you

The last pitfall is the biggest, but not specific to AI and not even specific to technology. In principle, it affects all innovation projects. There are no simple solutions here either. Based on experience, I would say that a large part of the hurdle has been overcome if the technically oriented employees are enthusiastic about the technology and the business management-oriented employees can calculate the ROI on a coaster. The wording probably plays the biggest role in innovation projects. For example, one should talk about new skills gained with AI technology in the company’s interest. On the one hand, this signals that the introduction of AI is not a means to an end, and on the other hand, capabilities still leave enough room for the design of solutions rather than immediately talking about concrete solutions. In the spirit of the Harvard communication concept, for example, one does not directly negotiate concrete things and positions with the works council, but first adjusts the respective interest. For innovation projects, it therefore makes perfect sense to bring in consultants who fully understand and can handle this fifth hurdle. This is where Captain Picard skills are needed.

Follow me on
We do not only optimize production processes, but also our website! For this, we use tools such as cookies for analysis and marketing purposes. You can change your cookie settings at any time. Information and Settings