To get the most from AI manufacturing, avoid short-term thinking

Digitalization is transforming the workplace. What are its main benefits and challenges?

Tough question. First, there is no "one" type of digitalization. Instead, there are different facets of new technologies – from AI to collaborative robotics, from low- or no-code to Industry 4.0.

Secondly, work activities are heterogenous and complex: the same professions can require very different tasks in different sectors, companies, and even departments. Third and finally, it depends on how much digitalization has been driven forward in a working environment in previous decades. Many companies are only now digitalizing work processes with technologies that have been around for 10 years. Others have always introduced state-of-the-art technology and can therefore really utilize the potential of AI.

But across all differences, there is the challenge of designing digitalization in such a way that it supports human autonomy and learning ability in the long term, and at the same time does not leave too large an ecological footprint (a major problem with AI in particular).

The greatest dangers are to use digitalization primarily for control or to want to exploit the supposed short-term potential of replacing humans prematurely – without having considered the costs of fallback solutions, redundancies and maintenance (e.g. also necessary for learning systems).

You are one of the few researchers who have looked into the implementation of AI solutions in typical blue-collar environments such as manufacturing. What are your key findings and what are the main implications you see for businesses?

A typical application of AI on the store floor is predictive maintenance. If the AI has been well trained for this, and the maintenance personnel have been involved from the outset, AI can help to predict faults and, for example, forecast the failure of an expensive spare part. What is often forgotten: Even the best AI predictions are never 100% correct. There can be false-positive and false-negative results. So, qualified maintenance personnel are still absolutely necessary.

When implementing AI on the shop floor, companies often forget that it's not just about designing a man-machine system. For example, if the AI predicts a spare part failure, it is not maintenance that orders the spare part, but procurement. But procurement is reluctant to order the expensive spare part because of a suspected impending failure. Who decides whether the order is economically justified at a certain point in time? It is therefore very important to always design the organizational processes "around" the AI application. This has hardly been done to date. So, it's not just about “keeping the human in the loop”; it’s also about “keeping the organization in the loop.”

From your own experience of working on the shop floor and providing technical support to customers in manufacturing, what would you say are the biggest fallacies in the current discussion on AI and smart manufacturing?

The greatest danger lies in truly believing in the intelligence of AI. AI uses stochastic models and makes predictions. You cannot always tell from its results whether they are correct and you can never rely on it 100%. AIs are learning systems, so they change over time, and can unfortunately become worse, and less precise.

In production, just as in critical infrastructure or medicine, every decision has real-life consequences. And these are often irreversible. That's why we need to be aware of the limits of AI. Only then can we really exploit its potential and implement it well and sensibly.

How can we overcome these issues?

Above all, those who decide how to use AI must understand its limitations. Unfortunately, managers sometimes naively believe in the omniscience of AI. All too often, they overestimate AI and underestimate the expertise and experience of their employees. It is important to bring both together well, in participatory implementation processes.

We first need to train decision-makers so that they really understand AI.

In light of accelerating digitalization, what can policymakers do to promote a meaningful upgrade in working conditions?

When it comes to the use of technology within companies, politicians can only intervene to a very limited extent. But politics can shape the basic conditions of digitalization and mitigate its negative consequences.

For example, it would be important to develop basic IT and AI knowledge on a broad scale; no degree course, no apprenticeship, no secondary school should be designed without such qualification topics. In view of the coming transformative digital change, it is clearly a political task to organize and broadly ensure the qualification of the population on these topics.

Where jobs are lost, politicians should create good framework conditions for further training or retraining of those affected. An important factor shaping digitalization in the world of work is also lively co-determination in companies. Thus, it is up to politicians to set the right political course for more co-determination.