1 Introduction

Artificial intelligence, short AI, is probably the most promising and discussed technology in the current media discourse regarding the impact it will have on future working life, what tasks humans fulfill as well as the conditions they will face at work. However, the views about future modifications AI will bring in the field of labor differ strongly from dystopian conditions to salvation promises. Many public voices believe in the substitution of most jobs by AI which might result in the loss of many jobs [1,2,3, 7]. AI is expected to change how we perform work tasks and offer new opportunities to think about the design of work processes to be useful in everyday working life. The rise of AI can be roughly attributed to digitization, which has already changed how we work and is still changing work operations. Driven by the competition on the market buzzwords like AI, Machine learning, and algorithms are used frequently in this context, the real meaning of this terminology and what these technologies are capable of is rarely clarified. This leads to uncertainty among employees about their future work, especially in technical fields such as engineering, as they cannot imagine any possible prospects on this topic. Furthermore, every technology is shaped by individuals like e.g. software engineers or mathematicians. Therefore, even if the expectations of the automation possibilities offered by AI in certain areas of work such as engineering are extremely high from a technical perspective [4], the question of what the change in this area will look like and what impact these changes may have cannot be fully answered at this point. Especially not if the perspectives of the people who ultimately have to deal with these technologies in their daily work are not taken into account. The goal should be to create AI technologies in the most useful, meaningful, and ethical way possible to help employees and facilitate their daily work lives. Co-determination of employees therefore needs to be put into the focus of the development and design process [8,9,10].

As previous research shows most of the time the focus surrounding AI lies on the capabilities and possibilities of this technology and how it might be impacting the sphere of labor from a broader perspective [11,12,13,14,15,16,17,18]. Often a tool will be created and integrated into a work surrounding top down. Employees will then have to adapt to the new program or tool they are suddenly confronted with. What is missing in this is the perspective of employees themselves. Their insights, contextual information, and process dependencies due to company or industry structures, especially in a complex field like engineering are important in work processes can be fulfilled. In addition, social, and interpersonal information may seem unimportant at first glance. However, they may be the only missing element for a process to work [19,20,21,22]. As employees are experts in their work, they have specific insights into the actual processes in their working environment. Therefore, they can assess more realistically what opportunities exist to replace processes with new technologies such as AI.

To focus on employees' assessments of AI, this article is not about the technology of AI and its possibilities, but about what employees need to be supported in their day-to-day work as well as how AI should be designed to meet these needs in practice. By bringing the employee perspective into the discourse, the transformation of work finally receives all the necessary perspectives, which thus form the basis for the design of human-centered AI. In this way, the development of AI is not only focused on the technical possibilities, so that a purely technology-oriented perspective, which constructs speculative possible use cases, can be related to the work practice of employees. Considering the employees’ perspective, the following article provides insight into engineers' views on AI technology. To this end, 11 interviews were conducted with engineers working either in German engineering companies (3) or in research institutions (4) in the field of engineering. First, a brief overview of previous research on employee perspectives on AI is given, followed by a detailed consideration of the analyzed data material. The conclusion ends with implications from the analyzed data on how to create a human-centered approach integrating AI in engineering.

2 The perspectives of employees on the technology of AI

Previous research on digitalization has already emphasized the importance of empowering employees in terms of digital transformation. Only in this way, technology can be designed to meet employees needs in their day-to-day work. Therefore, the co-determination of employees and the support of work councils are considered crucial when focusing on supporting sustainable ways of digitalizing work [23, 24]. This is also the case referring to the integration of AI, as this technology brings up new potential for contributions to work processes. At the same time, many challenges also need to be considered, e.g., the intended and not intended effects of AI-supported processes for other working steps connected to those.

The lack of definitions and information about AI as well as the lack of predictability regarding the impacts of AI are also mentioned when focusing on operational co-determination in shaping how AI technology should be used in the working context [25]. As Giering examined, the perspective of employees is underrepresented when thinking about the integration of AI at work. In particular, the concrete effects of the use of AI in work practice, how and where it changes the tasks and work steps of employees, and what effects this has on the career field have not yet been sufficiently investigated in the empirical data available to date [5]. And even though the topic of AI is discussed lively in public, the actual usage of this technology in Germany is not as diffused as one might assume. Accordingly, only 5.8% of all German companies were using AI technologies in 2019. In the industrial sector, electrical and mechanical engineering show the highest rates of 6.8% percent when it comes to companies using AI ([4], pp. 3–4). This indicates that there is still a lot of potential to shape AI as the diffusion is not as comprehensive as one might assume following the public discourse.

Furthermore, Seifert et al. already examined the perspectives of experts regarding the integration of AI in the field of manufacturing. The findings focus on the technological potentials of applications using AI and how to get the full potential out of these in the value chain. Although the study recommends providing more information about AI tools for AI users. However, the concrete perspective of future AI users in companies which are the employees themselves does not get as much attention as the interviewed experts from science and economy [27].

Other research pleads for a complementary approach to the integration of AI into the field of work. Instead of the substitution threat concerning jobs, the focus should shift to a concept of human–machine interaction, which needs to bring human-centered as well as technological approaches together instead of leaving them isolated from each other. If one perspective is ignored, this can result in tools being designed that have no added value for practical work, but rather complicate processes or create additional work for employees. The aim should therefore be to bring these two points of view together. Even though the substitution hypothesis sounds promising regarding the aims of the capitalistic value system of our economy, a complementary concept tackles functional aspects e.g. boundaries of AI automatization, non-intended outcomes of automatization, or human work that is not formalizable [28].

Instead of following the dictum of a rather negative outlook on the future of work and the use of AI technologies, which mostly focuses on the technical aspects and possibilities, a human-centered approach offers the opportunity to design AI-supported tools in such a manner that the best way for integration is found from both perspectives, the technical and the human-centered. This is also supported by research in the field of human-centered AI. Accordingly, the expertise of human experts, human skills, and their experiences must be integrated into the process of designing AI-based tools. As Hermann and Pfeiffer claim this participative approach to developing and shaping AI should also be complemented by keeping the organization as well as the organizational practices in the loop of this process to do justice to the complex process of integrating AI into organizations ([7], p. 1539).

If the concrete integration of AI at the workplace needs to be designed well, it is crucial to understand the work situation of employees. Context factors can vary from case to case as well as from company to company, even when they are active in the same field. Therefore, the expert knowledge of employees must be incorporated into the whole process of considerations for implementing AI. Complementary, 82% of employees find it important to be included in the design process of AI. Therefore, the early participation of employees in this process is recommended [28, 29].

All this previous research shows, that the perspective of employees regarding the integration of AI in their work lives has not yet been sufficiently investigated even though it seems to be a crucial part of designing AI in an employee-oriented way, which supports and relieves their daily work life. With more detailed information about workers' views, needs, and concerns about AI technologies, more insight can be gained into how this technology could reshape future careers and influence workforce dynamics in specific industries. As context factors and requirements vary from field to field the individual consideration about the implementation should be the way to go for companies by integrating their employees as experts of their work and co-creators in the design process of AI [30].

3 Data and methodology

To examine the employees' perspective on how and where AI might find its place in the sector of German engineering, experts from this field were recruited as interviewees to get perspectives on AI from an engineering specialist perspective. The main aim here is to shed light on qualitative contextual information on procedures, processes, specific situations, and dispositions from this field of work regarding the potential use of AI in engineering. To obtain realistic assessments of the potential and limitations of this technology, care was taken when selecting the experts to ensure that they were familiar with the topic of AI and had already had contact with this technology, whether through a concrete practical application in companies or an exploratory discussion outside of day-to-day business. The involvement and role of the interviewees in the operational context ensures that they can provide a realistic assessment of the points in engineering at which the integration of AI technologies is feasible and, above all, makes sense based on the needs of employees. A total of 11 engineering employees from 7 private German companies and four from German research institutions were interviewed (Table 1). The average age of the interviewees was 34, with an age range from 26 to 52. The interviewees had been employed at their current company for between 1 and 21 years at the time of the interview, with an average length of service of 6 years across all interviewees. In terms of education, the sample includes people who have completed vocational training (1), a vocational baccalaureate (1), and a bachelor's degree (1), and the majority of interviewees have a master's degree (7). One interviewee had completed a doctorate. Most of them are working in technical fields: Electrical Engineering (1), Electromechanics/Tool Mechanics (2), Engineering (1), Mechanical Engineering (1), Informatics/Computer Science (2), but Physics (2) and Industrial Engineering (2) are also represented.

Table 1 Demographic characteristics of interviewees
Full size table

The interviews were conducted between 05/02/21 and 22/04/21 with the help of a guideline. The previously created guideline served as an orientation during the interview and enabled comparability between the interviews. Due to the pandemic situation at the time, all interviews were conducted online via video telephony systems whereas only the corresponding audio track was recorded separately. The duration of the interviews ranged from 40 min to 1:46 h.

After the audio data was anonymized and transcribed, the qualitative analysis was carried out using the coding program MaxQDA. For the qualitative content analysis, the approach of Mayring [31] and Hsieh and Shannon [32] was chosen. The data material was initially coded in a first session, followed by several rounds of revision, which led to continuous processing and changes. In further rounds of analysis, this iterative process led to adjustments to the codes based on the data material, differentiation of sub-codes, and fundamental revisions to the code tree. In total, the code tree was revised three times until all findings had been incorporated accordingly. The entire process of coding, changing, and revising codes until the final codes were found and the code tree was finalized took several months and was done recursively. After this, the data material was brought into bigger categories, to sum up the findings referring to the statements of the experts. At first, a more detailed look into the definitions, terms, blurs, and methods regarding AI is given, followed by the specification of the three employees’ perspectives which could be condensed from the material. Afterwards, fields of application of AI as well as the boundaries from the point of the interviewed experts are described.

4 AI in the field of German engineering

4.1 Definitional ambiguities: context dependencies of AI terms

Before going into detail about the evaluations of the interviewed experts regarding the technology of AI and its potentials and risks in the field of engineering an overview of used terms regarding AI is useful. Thus, the context of how interviewees frame their definition of AI gives important information on how the following states need to be interpreted. This results in a broader picture of what the experts initially understand and classify under this technology or term. In the comments on these questions, it also becomes clear that the experts surveyed often find it difficult to give a precise and clear definition. The explanation for this is that this is influenced by the context and depends on how this definition is formed.

But if you're talking to a different bubble, then you first have to clarify with each other, okay, what do you see as that, what do I see as that? And you can also have different opinions. I don't think that's a problem at all, you just have to understand what the other person means by it and then you can have a normal conversation. (IV-No.10, pos. 23)

Reference is made here to various areas in the engineering context that are referred to as bubbles, in which divergent perspectives and correspondingly many different definitions of AI can be found. For this reason, the interviewees point out the need for mutual understanding so that the meaning behind concepts becomes clear for all parties involved in the interaction. In addition, some statements indicate that it is often not possible to find a uniform definition within their own company and that they therefore tend to illustrate the technology using specific examples.

I actually believe that there is not yet a real understanding of artificial intelligence in our company. So, I think many people know the word. Simply because it is publicized a lot and is somehow the new technology. But I think very few people know what's really behind it or what the added value is. (IV-No.2, pos. 58)

Overall, the statements illustrate how important and necessary uniform terminology and language in defining the term AI is for unambiguous interaction. It is therefore important to focus on the individual understanding and discursive exchange of the terms used at first to work out differences and ultimately ensure that everyone involved is talking about the same thing(s).

4.2 Conceptual ambiguities: blurs of AI definitions

The definition of what AI is differs between the interviewed experts, depending on their experience and environment, which also reflects in the actual vagueness of the definitions of this term as seen in the sample.

To examine this more closely, it is useful to look at the statements that make it clear what AI does not mean from the respondents' point of view: for example, the mere evaluation of data material does not constitute a form of AI from the perspective of a respondent. As long as a system receives specifications such as parameters, i.e. rules and instructions on how it should handle the available data material, it is not classified as intelligent or "smart" by several experts. Specifically, they say that anything that is explicitly programmed in terms of behavior is not a form of AI. This demarcation leads directly to a definition of what AI actually is, also given by some experts. Broadly speaking, they define AI as the attempt to replicate human capabilities so that this technology is able to do what humans can do. In other words, the goal is to provide a performance that is equivalent to human performance.

Some of the interviewees particularly emphasize the human ability to act independently that characterizes this technology. AI technology is therefore described as a program that can act independently and also constantly optimizes itself, which in turn is a comparison with humans, who can solve unknown situations without specific instructions. About the concept of intelligence, it is therefore pointed out that everything that is understood as intelligent in humans also applies to what a machine defines as intelligent.

Other experts refer directly to specific processes such as machine learning, data-driven engineering, or algorithms, which involve recognizing patterns in data material and generating suggestions or recommendations for actions based on this. In addition, the aspect of the learning ability of such a system is emphasized, which, however, must be trained by example in order to cultivate the handling of certain situations. However, this ability is differentiated from autonomy, as the aspect of learning and the derivation of rules of action can only take place after a program can act independently.

But from my point of view, what people usually mean when they say AI is actually machine learning. In other words, how it learns something itself. And yes, it's always a bit blurred. When people say they mean AI, they actually mean machine learning (IV-No.8, pos. 106)

The problem is that the term intelligence is used in connection with AI in everyday life without a clear definition, which leads to the capabilities of methods such as machine learning being equated with the intelligence of humans, which in the opinion of the experts cannot be regarded as equivalent.

4.3 AI and its methods

Even though many ambiguous and vague references to what AI is can be found in the sample, at some points the interviewees are more specific and definite about their understanding of this technology. Thus, one interviewee refers specifically to the concept of weak AI and describes it as "quasi-specialist idiots", as individual issues are outsourced and then solved by the system. It is also pointed out elsewhere that AI is not a tool:

It's not really a tool, is it? It's not a screwdriver. It's an algorithm and when you start using it, it takes quite a bit of time and you don't know where it's going to go. (IV-No.3, pos. 93)

While one expert questioned whether a standardized definition is even possible, another described the definition of AI as not fixed, as it is in flux. This broad range of statements continues with the question of how AI works and which methods this technology uses to produce output. Here, statements can be found that identify the processes behind AI solutions. Therefore, machine learning (ML) is referred to as the specific method behind AI. While ML describes the method of how algorithms are trained, other knowledge-based methods that describe different approaches to machine learning are also mentioned: Deep learning, federated learning, supervised learning, and reinforcement learning. In other places, reference is made to statistical learning, which has been a common term for decades and is also categorized as a machine learning method.

Machine learning has been around forever, which means that I train the computer to discover patterns, but with deep neural networks this is possible in a way that was not possible before. (IV-No.5, pos. 28)

It is precisely at this point that one of the many ambivalences with regard to the statements of the experts interviewed becomes apparent, as previously elsewhere they had denied precisely this specification of parameters as a characteristic of an AI. What is particularly interesting is that these contradictory statements come from one and the same interviewee. On closer inspection, it becomes clear that the terms AI, algorithms, and machine learning are often used as synonyms in the current discourse from the interviewee's perspective.

In terms of data, AI ultimately represents the ability of a system to determine the parameters for the instructions itself to deliver the optimal output or optimization, which also includes prioritization.

These robots also have to be fed with a program. […]. And you have to do the same for this […] software. So, it really always has to be a consistent process. And then it's not AI for me, so to come back to the question, it's a program for me. I'm basically teaching the program what it should do. It doesn't teach itself, like that. And in that case, it wouldn't be an AI. (IV-No.7, pos. 24)

Elsewhere, in response to the question of how AI works, it is described as being about extracting knowledge from data in order to ultimately make recommendations regarding decisions based on this knowledge. This statement is much more specific with regard to the output of an AI system, but the process is not discussed in more detail.

4.4 Employees perspectives on AI in the field of German engineering

Based on the evaluation of the sample, three types of employee perspectives on the application of AI can be identified with regard to the assessment of the benefits of such technologies in the German engineering context: the critical, the ambivalent, and the positive employee perspective. These three types make reference to technological aspects dependent on the engineering field, its special setting, related requirements resulting from the surroundings, and infrastructural conditions of this work area.

4.4.1 The critical employee perspective

Employee statements that can be identified as rather critical of the use of AI in German engineering refer to the high effort required to make this technology usable in their working environment and everyday working life. It is mentioned that other approaches, especially procedures that have already been used before, are associated with less effort due to the existing expertise of employees and software engineers in this area and therefore are easier and quicker to implement compared to AI methods. In addition, the use of AI processes is often not necessary from the customer's point of view or customers do not need this technology in their products as it does not add any value regarding the given requirements. So, there is no justification for the application of AI from this perspective. Furthermore, experts having a critical perspective emphasize that complexity is something that AI cannot handle well at the current stage of development. Also, the increase of complexity in data requires much more sophisticated programming algorithms to achieve reliable results. If parameters change, the system needs to be reviewed again to make sure, outputs are still reliable to match given demands. The experts interviewed thus make a simple cost–benefit calculation, which then turns out in favor of methods that are already known and used. This is also because examining the possibilities and designing and programming suitable tools requires time and money as well as qualified personnel, which is rarely or not at all available in the experts' day-to-day business.

When it comes to the applicability of AI tools in engineering, it is also important to consider the requirements that this technology entails and that make an application possible in the first place. One point that is repeatedly mentioned by employees with this perspective is the missing standardization of processes that an AI should map. Whether this might be a language model that is implemented in the company or other processes that are to be executed by an AI tool, experts emphasize the importance of standards:

In any case, yes, and this is actually a great message that is also received by our customers in the industry, even from the AI and ML projects that don't work so well, is that I need a) extremely high standardization, for example how I measure my car should always be the same. If it's not always the same, my data is almost worthless. (IV-No.5, pos. 61)

The work steps must be carried out in the same way in order to establish sustainable process standards. This is not the case referring to the interviewed experts due to the fact that development processes and product life cycles are always different. Especially e.g. in special machine engineering, deriving rules for AI is difficult, which makes the integration of AI from their view a paradoxical matter.

4.4.2 Ambivalent perspectives on the use of AI in German engineering

Statements that can be assigned more to the ambivalent employee perspective address the still too generic orientation of current AI applications and, as already mentioned elsewhere, call for more specific use cases instead of quite generic application scenarios. At the same time, this perspective of AI sees both sides: the possible positive results that they might get from using an AI tool and the possible negative aspects of such an integration. This perspective also raises the question of what is acceptable from an ethical point of view, but also about the question of responsibility for decisions. Specifically, experts with this perspective ask what AI should do, what humans should do, and what is acceptable in terms of ethical standards in work processes.

The interviewees also mention the availability of data for implementation, which they believe is not available in many areas in this current state. This means that the direct applicability of AI is not given due to the lack of relevant data. Before this technology can be used, relevant data must first be generated to process it in further steps. For example, from the experts’ view, this means that a machine should be running quite unobstructed to gather data that can be processed by an AI tool in the next steps.

When we develop a system or a machine, we can only use an AI directly in the rarest of cases, because the machine has to run first, has to generate data that can then be used to use such a system. (IV-No.4, pos. 75)

However, this also means that the data needs to be inspected, cleaned, and checked by experts for further analysis to have a reliable and nonbiased database.

The analysis of the ambivalent employee’s statements shows that the area of engineering offers some points of integration of AI technology so that many process steps could be automated from their point of view. But for sure not the entire development process. There is a consensus of the interviewed experts that certain tasks can be performed better by humans especially when it comes to special mechanical engineering. In addition, the topic of transparency with regard to AI applications is attributed a major role, as background information for analyses is considered an important requirement in order to be able to understand decisions made by an AI system more precisely. This is the only way that the process behind the output can be tracked and reconstructed by employees. Therefore, this view asks concretely about the arrangement of this technology based on the skills of each party.

4.4.3 Positive perspectives on AI in German engineering

The positive employee perspective on AI focuses strongly on only the advantages the implementation of AI could or will bring to their daily work life as well as to the field of German engineering. Employees with this perspective generalize the positive outcomes of AI. Accordingly, very positive assessments can also be found regarding the possible applications and associated benefits of AI in engineering. For example, respondents believe that this technology will be able to make work easier and lead to results more quickly compared to traditional methods. From this perspective, AI could also potentially "provide support almost everywhere". In particular, the computing capacity, which exceeds the capabilities of a human being, is addressed here. For example, errors of the last hundred years could be analyzed by an AI tool, in order to give more information about this topic, whereas employees do not have the capacity as well as no time during their daily work tasks to handle so much data volume.

But when it comes to taking into account, for example, the error cases of the last hundred years, a person will never be able to do that, simply because they can't read that quickly. A computer, on the other hand, could simply analyze all the data from the last hundred years and then draw a conclusion. And in line with that, I think you can make certain activities much more efficient, but you can also create new standards. (IV-No.11, pos. 26)

This can lead to faster insights and results, which in turn speeds up decision-making as it can provide the right information when needed. It also could help to further educate knowledge. In addition, AI approaches are rated as more flexible in terms of changes and adjustments compared to traditional methods. One expert also mentions the positive result from the use of statistical learning methods and a high hit rate, which came as a surprise to him.

[…] I was rather surprised after the first few attempts that it worked relatively well. So rather the other way around. I had rather thought that the algorithms wouldn't work so well with the project I had, with the pictures, which are relatively disgusting. But I had relatively good results with 97, 98 percent hit rates, where the algorithm was just right, although I haven't gone through all the optimizations yet, so in that respect I was relatively positively surprised that it was relatively good. (IV-No.1, pos. 42)

4.5 Fields of AI application in German engineering from the employees ‘ perspective

4.5.1 Current fields of AI application

When it comes to specific areas of the application of such AI tools in engineering, the interviewees mention many areas in which support would be conceivable and feasible. Some of these statements refer either to areas where AI is already being used, albeit in very narrow areas, or to areas that the respondents classify as very time-consuming and could therefore use some relief from their individual work experiences. If these statements are placed in the context of the interviews, it becomes clear that more generic and rather optimistic statements tend to come from people who have no direct connection to technology in their daily work and comment on it from a more distanced position. Therefore, these statements can be classified more as expectations and hopes for this technology. In contrast, experts who have more detailed knowledge of AI give more concrete application scenarios.

As a field that is already making use of AI measurement technology was mentioned several times by interviewees. The sector is viewed as a very promising one for more supporting and efficient implementation, as collecting data is the main focus of this area. Therefore, data from this field could be easily processed, analyzed, and filtered to use it for maintenance as well as for predictive maintenance in the engineering area. It also appears to be promising, as a large amount of data has been collected there, which is needed to obtain reliable results.

[…] And then I might want to know, ah, whenever one sensor reports something, the other one reports it too and I might want to see small correlations and that costs, so if you want to do it manually, it's virtually impossible and with ML he can say exactly, ah, okay, okay and the engineer can then say, ah yes, of course, whenever my shaft breaks, this and that happens, I'll say that now. In other words, it can somehow assign causality to this correlation that this algorithm finds. (IV-No.5, pos. 42)

Another field where AI is already being used by engineering employees is image processing. The preparation of images in order to process images driven by special questions is something the technology of AI made much more feasible. In addition, libraries with images for image processing and training algorithms have already existed for around twenty years, according to one interviewee. Against this background, one interviewee's statement about the development of AI becomes clearer:

But I don't think you can say from the outside that it's a revolution, it's more of an evolution, yes. (IV-No.5, pos. 32)

4.5.2 Desired areas for the application of AI

The desired areas when the experts talk about the application of AI are driven by the experiences they have had in their individual fields of engineering and therefore tackle their needs as well as hopes regarding time-consuming and complex tasks. In general, the wish here is to become more efficient by using AI tools. Less effort is also desired for engineering and feature engineering. More specifically the field of constructing machines offers the possibility of having AI as a tool in order to be able to respond better to customer wishes. Therefore, both requirements management and conceptualization are seen as potential areas of application in which AI could help to support and improve processes.

Also, very simple tools using AI which could offer recommendations about often used documents like charts would help employees in this field and support them towards predictive maintenance. From the experts’ view, this would add value to the daily work life of engineers as it would help to keep all relevant context factors in perspective which is hard for employees due to their cognitive capacity and the complexity they are facing day by day. If such a tool were also able to continuously learn from its application in practice, the process would be simplified from the respondents' point of view and provide greater help in setting priorities.

Things that you use a lot are just in some chart, I say, and there are some quick accesses. And that would be AI for me, so to speak. The machine notices that you use it a lot, so, yes, I might prioritize it a bit more. (IV-No.6, pos. 19)

4.6 The boundaries of AI from an employee’s perspective

When talking about the boundaries the implementation of AI is facing in the field of engineering the interviewed experts address the data that needs to be available in order to process reliability as mentioned before. That the record of data is in most companies poorly makes human intelligence as a sparring partner for AI tools indispensable to make these tools useful. Another basic requirement already mentioned is standards as a basis for actions and decisions on how systems should interact. As these standards are most of the time not given in companies the experts are sceptic about the usage of AI.

[… ] Because I believe that before you have to use an AI properly, you first have to have certain standards that this AI can access. And that's an issue for us. But yes, it's very difficult in special machine construction. So that's also... There are two words that absolutely clash: standard and special. We are in special machine construction and are supposed to work according to standard. That's absolutely paradoxical at this point. (IV-No.7, pos. 28)

Also, definitions, terms, and what employees understand under these can vary from company to company. Therefore, the context and language of any company need to be kept in focus in order to even enable standardization. Accordingly, when developing AI tools, the company-specific vocabulary should be taken into account and incorporated into the system. As stated by the interviewees, a supporting and helpful application of AI needs standards that ensure that working methods are 100% the same, thus employees and their mode of operation needs to be brought into harmony.

We would definitely have to acquire a way of working that is one hundred percent the same so that this AI can work properly. You have to have a certain structure beforehand. And, as I said, this is due to the special machine construction and yes, it is very difficult to harmonize everything. Because sometimes there is no right or wrong. With one project it makes sense to proceed in this way, with another project it makes more sense to proceed in this way. And you first have to reconcile all the people involved. Ultimately, people have to work with it. (IV-No.7, pos. 34)

From the interviewee’s perspective at this point, technology and human intelligence need to work hand in hand in order to tackle such situations. This is also supported by the fact that the experts state that development processes and product lifecycles are always very individual, which is why it is difficult to derive clear instructions for an AI tool and the expertise of employees for unknown situations is considered important in order to provide appropriate instructions for systems. Furthermore, not enough time capacities for application is a huge boundary the interviewed experts state. This also goes back to the insecurity around projects as the development of such is oftentimes not foreseeable and thus jeopardizes structured project plans from the outset, which then leaves time for projects that are not part of day-to-day business. At the same time, the hype around AI is so big, that if companies do not put effort into investigating the possibilities of AI the risk of being overtaken by the competition is high. Even though often there is no certain application of AI in the industry the pressure of competition on the market is forcing companies to apply AI whether it is useful or not.

Apart from this, the reliability of the outcome of AI-processed data is seen as critical by the interviewed experts. This goes back to the fact that often AI is acting like a black box, therefore the process of how a tool came to a decision or generated the output cannot be traced.

It's basically a black box and you have ... it's also very difficult to validate these things, i.e. the AI, in an industrial context because ... yes. In principle, you don't know exactly what ... according to which rules the decision is ultimately made. And yes, you have to put a lot of effort into securing and validating it. That's definitely a big problem. (IV-No.4, pos. 64)

In order to be fully aware of the reliability of the output, the data must be trained well after it was made sure, that the data is valid. Otherwise, biases as well as mistakes from the collected data will be reproduced by the AI. Verification of data material is a crucial part of getting useful outputs by AI, but this procedure takes time and expertise from human intelligence. Also, complexity as well as context factors like connections or dispositions are difficult to handle for AI tools, as the programming of such algorithms is very time-consuming.

[…] but it's actually the case that you have to say that complexity is actually something that involves so many connections, context, context, which ML doesn't master so well, where people can still look at it. (IV-No.5, pos. 65)

5 Conclusion

The aim of this study was to investigate the views of employees from the German engineering sector on the possibilities of using AI in their workplace. As already stated at the beginning of this paper, the perspective of employees has not been investigated thoroughly in previous research even though it is seen as a crucial part of designing AI tools that actually facilitate the work of employees instead of causing more issues. Furthermore, context factors vary from field to field which therefore makes it impossible to have one approach fit all areas. Therefore, an individual and specific inspection of the field of interest is crucial in order to apply AI in a useful and human-centered way. For this paper the field of engineering was chosen as the work areas of the interviewed experts are quite similar and therefore make it possible to compare their views. However, the results show that despite the great similarity of the work areas, the definition of AI varies and depends on the context, e.g. the company, the environment, or the vocabulary used there. It also becomes clear, that even though AI is a topic especially at the management levels, as well as driven by competition in the market, the actual application of AI in the field of engineering is happening in a much narrower way than the public discourse is stating [26]. Accordingly, AI can only be used properly if relevant data is already digitized to be utilized by AI technologies. While the limitations regarding the use of AI, such as a lack of standardization or data that is too complex, are mentioned by the interviewees, the existing advantages of using this technology are also addressed. The extended possibilities to shift capacity limits in the processing of data as well as the increase in accuracy are considered valuable attributes of this technology for the work of engineers.

In order to design AI tools in such a way that employees in the field of engineering are supported and relieved by the introduction of AI, their perspective should be integrated into the process from the outset. Employees must therefore be trained in such a way that they are able to participate in this process. Specifically, they need to know what AI is and what the technology can and cannot do. If they do not know what AI is able to handle well, the individual requirements for tools will be too complex and therefore not possible to transfer into an algorithm. As previous research has already found out, this has also an impact on the acceptance and trust employees have in such tools [33]. Instead of pinning too much hope on this technology, realistic use cases can be created and evaluated by all stakeholders affected by these use cases. What seems crucial after the results are represented is that circumstances and processes need to be considered individually. Sufficiently detailed contextual information from the employees, who serve as experts for their own work, must also be taken into account, as there can be no one-size-fits-all solution based on the analyzed data material. Therefore, more research on the actual needs of employees in everyday life should be considered in the future. While the qualitative approach could provide very detailed information about individuals and specific circumstances due to the sector of work, quantitative analyses or surveys could provide results that are easier to generalize. A mixed-methods approach combining qualitative and quantitative data [34, 35] could thus provide more insights to develop a general process for implementing AI tools in the workplace.

Nevertheless, the possibilities of AI in the area of engineering are constantly growing due to the rapid development in this area. This must also be taken into account, as the analyzed data material was collected before the introduction of the generative AI ChatGPT [36]. However, since the field of engineering is very complex in terms of requirements, such as in the field of special machine construction, and the complexity cannot be managed well at the moment, AI should be seen as an opportunity to facilitate or even take over time-consuming and unpopular tasks in the engineering process instead of compulsively trying to integrate this technology into processes due to competitive pressure, even if no added value is created there as a result. Further research focusing on all perspectives of the parties involved in specific areas is therefore needed. This is the only way to draw the right conclusions for individual areas of work as to how AI technology could be used in a sustainable and supportive way in the future.