Against the background of the launch of ChatGPT and the associated growing importance of artificial intelligence, the bidt is investigating the use and acceptance of generative AI with a survey among approximately 3,000 people in Germany. Generative AI is a collective term for systems with artificial intelligence that generate new content such as texts, images, program code, videos or music. In this study, the bidt presents results for all internet users in Germany and for internet-using employees, focusing on the use, regulation and expected impact of generative AI on the working environment.
According to a study by the McKinsey Global Institute, generative AI could generate up to USD 4.4 trillion per year in global corporate profits. Generative AI could increase productivity, help create creative content, particularly for marketing and sales, and generate software code without programming skills based solely on voice-based input (McKinsey 2023). At the same time, however, employees fear losing their jobs, because generative AI could take over their tasks (Spiegel 2023). Therefore, generative AI is hotly debated in the working world and is becoming increasingly important in society and politics. The bidt takes these current discussions as an opportunity to conduct a survey investigating the use and acceptance of generative AI in the population and the workforce.
To this end, the bidt hired the market research institute DCORE to conduct a representative quantitative survey using a standardised online questionnaire. From 20 July to 4 August 2023, DCORE surveyed 3.020 internet users aged 16 and above in Germany, including 1.008 full-time-employed persons. The following results are representatively weighted for the population of German internet users, with weights based on age, gender, education and federal state. Weights were generated separately for the subset of full-time-employed people, again using age, gender, education and federal state (b4p 2022).
About one-third of internet users have already used generative AI
Most internet users in Germany (81%) have already heard of using generative AI for creating and/or checking texts or program code or creating images or videos. However, only a good third of internet users (36%) have used such systems at least once. Among those in employment, the proportion of people who have already used generative AI for one of the purposes mentioned is 40%; for unemployed people, this proportion is substantially lower at 27%.
A total of 32% of internet users have already used generative AI to create and/or check texts or program code, 17% once and 15% several times. 45% have heard of generative AI for these purposes but have not yet used it.
Younger people and those with a higher level of formal education with at least a university entrance qualification use generative AI more frequently for creating and/or checking texts and programme codes than older people or those with a lower level of formal education. Those who indicate knowing much about AI are also more likely to use text-based generative AI.
Most users of generative AI for creating and/or checking texts or programme code have used it to research information (56%), summarise texts (48%) and create new texts, and hold conversations (42% each). In contrast, just over a quarter have had programme code translated into another programming language (27%) or checked (26%).
Focusing on the risks of generative AI rather than the opportunities
While almost half of the internet users in Germany (46%) believe that the opportunities and risks of generative AI are balanced, 26% mainly see risks and only 18% focus on opportunities. 11% of internet users are unable or unwilling to assess in this regard.
People with a lower level of formal education are more likely to see risks than those with at least a university entrance qualification. In addition, people likely to emphasise the opportunities of generative AI also rate their knowledge of AI as high.
A similar correlation can be seen with previous use of generative AI. Respondents who use this technology more frequently are more likely to see opportunities rather than risks. For 42% of internet users who have already used generative AI several times to create and/or check texts or program code, the opportunities outweigh the risks, while the opposite is true for only 9%. Among respondents who have heard of generative AI but have not yet used it for the abovementioned purposes, only 13% primarily see opportunities, while 30% mainly perceive risks.
Compared to an earlier survey on the opportunities and risks of artificial intelligence, conducted by bidt prior to the release of ChatGPT, the assessment of opportunities and risks has shifted somewhat, leaning towards an increased perception of risk. In the bidt-SZ-Digitalbarometer 2022, 21% of internet users in Germany still predominantly saw opportunities in artificial intelligence. 50% thought the opportunities and risks were balanced, and only 20% saw the risks as predominant. The difference for the group of 16-29-year-old internet users is particularly striking: while 36% of them still saw the opportunities of AI in the bidt-SZ-Digitalbarometer 2022, only 20% of them said this about generative AI in the current survey.
Almost half of internet users in Germany believe in a loss of control over AI
Just under half of all internet users in Germany agree or strongly agree that the increasing use of generative AI could lead to humanity losing control over the technology in the future. Only 18% believe that this is not a realistic scenario. This pessimistic view is particularly interesting in light of the population’s relatively balanced opportunity/risk assessment. Even among users of generative AI, the proportion of those who are rather or completely afraid of losing control over the technology is 44%. This proportion is also 44% among 16-29-year-olds and 45% for those with a higher formal education, meaning at least a German university entrance qualification.
Around a quarter of employees have already used generative AI in a professional context
The majority of internet-using employees have not yet used generative AI in a professional context to create and/or check texts or program codes (74%). Only around 4% of those in employment use generative AI for these purposes exclusively in a professional context, while 23% use the technology professionally and privately.
In most cases, the use of generative AI at work is communicated openly. For example, 79% of users of text-based generative AI state that their employer or supervisor is aware of this, with only 18% reporting the opposite.
In the professional context, the use of generative AI differs from its general use in the population. Employees who have already used generative AI to create and/or check texts or program code use the technology most frequently to summarise texts (45%) and to create new texts (42%). Researching information, the most popular use overall, is the third most common use of text-based generative AI in the subgroup of employees (38%). This use is followed by checking (32%) and classifying information (29%), as well as creating (25%), checking (25%) or translating programme code (23%). Conducting conversations only plays a subordinate role in the professional context, while this purpose is significantly more important for all users of text-based generative AI.
Most employees are aware of the risks associated with using generative AI
39% of internet-using employees who have heard of generative AI (strongly) agree that they can recognise, in which cases the use of generative AI helps complete tasks and in which it does not. Among employees who have already used text-based generative AI in a professional context, this proportion is 57%. In contrast, 11% think they are (completely) unable to assess this.
Regarding data protection, 62% of employees who have already used text-based generative AI in a professional context agree or fully agree that they carefully consider which data they provide when using generative AI. Many employees, therefore, seem to be aware of potential data protection issues.
Furthermore, most internet-using employees in Germany know that results created by generative AI can be factually incorrect (57% agreement) or at least incomplete, unbalanced, contradictory and inappropriate (56% agreement). These insights are even more common among users of text-based generative AI in a professional environment (66% and 65% agreement, respectively).
The spread of generative AI also leads to a growing interest in this technology among employees. Half of all employed people agree or strongly agree with the statement that they would like to understand better how generative AI works; only 18% do not. There is also a clear positive correlation between the interest in learning more about generative AI and previous knowledge or experience with generative AI. Only 35% of employees who have never heard of generative AI tend to agree or fully agree with this statement, compared to 63% of employees who have already used text-based generative AI in a professional context.
The majority of professional users perceive generative AI as beneficial
The majority of employees who have already used text-based generative AI in a professional context are satisfied with the application. For example, 60% agree or strongly agree with the statement that they could use the generated results in a meaningful way at work. Additionally, 64% agree or strongly agree that generative AI has saved them working time. Further, 56% state that generative AI is helpful for tasks that are difficult for them. Finally, 63% of employed users agree or strongly agree that generative AI has helped them gain a positive perception of their work without requiring additional effort. In general, we find a positive correlation between the likelihood of agreement with these statements and the frequency of use of generative AI.
Despite the high level of satisfaction with generative AI, the majority check the results: 67% of employees who have already used text-based generative AI in a professional context agree or strongly agree that they have checked the results provided by generative AI for accuracy. This assessment gains significance in light of the understanding that results of generative AI are not always trustworthy, but may also be factually incorrect or incomplete, unbalanced or contradictory. 77% of employed professional users who say that they are aware or fully aware of generative AI being able to produce incorrect results also agree or fully agree that they have checked the results in a professional context. This means that the more employees state that they are aware of this potential problem, the more they review the results of generative AI.
Generative AI is receiving attention in many companies
The awareness of the use of generative AI is heterogeneously distributed in companies. For example, 39% of all internet-using employees in Germany who have heard of generative AI state that little to no attention is paid to this topic in their company or institution. In contrast, 31% report that generative AI receives a high or very high level of attention in their organisation. There is also a clear association with the professional use of text-based generative AI: of those employees who have already used text-based generative AI professionally, 61% state that the use of generative AI is given a high or very high level of attention in their company or institution. However, only 17% agree among employees who have already heard of generative AI but have never used text-based generative AI in a professional context.
Most Companies still lack guidelines for generative AI
Despite the medium to high level of attention paid to generative AI in many companies, only 21% of internet-using employees who have already heard of generative AI state that they have specifications and guidelines for the use of generative AI at work. At the same time, a clear majority of two-thirds report that they do not have any specifications or guidelines in their professional life. This proportion is 46% among employees who have already used text-based generative AI professionally and 43% who state that the topic receives a high or very high level of attention in their company or institution.
Almost half of those in employment who have heard of generative AI but have no professional specifications and guidelines would like to have them. At the same time, a quarter of those still do not want any specifications and guidelines.
Again, there is an association with professional use: focusing on employees without guidelines who have already used text-based generative AI professionally, we find 30% who do not want clear specifications and guidelines, slightly more than among those without professional use (23%).
The assessment of opportunities and risks is balanced in a professional context
When asked about the opportunities and risks of using generative AI in companies, employees answered somewhat more balanced than all respondents. Almost a quarter of employees believe that the opportunities outweigh the risks, while another quarter take a more pessimistic view, mostly emphasising the risks. 42% of employees state that they think the opportunities and risks of generative AI in companies are balanced.
The majority expects generative AI to substitute their work activities
Focusing on their professional activities in connection with generative AI, employees expect changes: almost half of internet-using employees predict that all or at least some of the activities in their job will become superfluous in the next ten years due to the increasing use of generative AI (47%). However, almost as many do not see any risk of substitution here (43%). We see clear differences comparing employees who have already used text-based generative AI in their jobs with those who have not yet done so. For example, 77% of employees who have already used text-based generative AI at work expect that at least parts of their activities could be replaced, in contrast to only 36% among employees who have not used AI at work.
Due to digitalisation, the bidt-SZ-Digitalbarometer 2022 asked whether parts of activities will be eliminated or become redundant in the next ten years. Compared to this more general question, our current study shows an increased expectation about eliminated work activities. In the bidt-SZ-Digitalbarometer 2022, only 24% of employed people expected parts of their job to be eliminated due to digitalisation in the next ten years. Only 1% feared that all activities would become redundant.
There are clear differences when looking at different age groups. Younger employees up to the age of 29 have a much more pessimistic view than 50–64-year-olds. For example, 64% of the younger employees assume that at least some of the work activities will be replaced by generative AI, while only 34% of the older ones are of this opinion.
We also find a heterogeneous picture in terms of education levels. Among formally low-educated employees without a German school-leaving qualification or with a lower secondary or primary school leaving certificate, 23% believe that at least some of the activities in their profession will become superfluous in the next ten years. 39% of the formally medium-educated employees and even 56% of the formally higher-educated employees with at least a German university entrance qualification 56% are of this opinion.
The proportion (54%) of internet-using employees who somewhat or fully agree with the statement that jobs will be lost in the future as a result of generative AI is correspondingly high. Accordingly, a high proportion (54%) of internet-using employees somewhat or fully agree with the statement that jobs will be lost in the future due to generative AI; only 10% expect that there will be no job losses.
Majority of the population wants stronger regulation
Over half (52%) of all internet users in Germany would like to see greater regulation of generative AI; only 11% see little or no reason for this. Around a quarter even believe that the further development of generative AI should be stopped. At the same time, 37% also agree or fully agree with the statement that generative AI will make life easier in the future. Furthermore, 33% agree or fully agree that using generative AI will strengthen Germany as a business location, while 20% are of the opposite opinion. Overall, the assessments of generative AI among the population thus paint a heterogeneous picture. On the one hand, people emphasise fears and anxieties along with a strong desire for regulation. However, on the other hand, some voices stress the potentially positive implications of generative AI.
The results of this bidt survey show that with the launch of ChatGPT a year ago, the use of generative AI spread relatively quickly among the population and in German workplaces. This expansion went hand in hand with growing fears of job loss and specific concerns of loss of professional activities among those in employment. It is striking that primarily those with a higher level of formal education fear a loss of work in their profession due to the use of generative AI.
This assessment aligns with findings according to which cognitive skills and, thus, knowledge workers are particularly affected by advancing automation in the current transformation processes triggered primarily by generative AI (Gmyrek et al. 2023). In particular, the activities of specialists in business and administration, as well as science and technology, will be increasingly taken over by generative AI (Oschinski 2023). Additionally, creative fields are increasingly affected, as recently illustrated by the labour dispute among US screenwriters and actors (The Guardian 2023). Whether the fears of large-scale job losses will materialise is uncertain. In some professions, tasks will change, and people will increasingly carry out activities that still cannot, or not well enough, be performed by AI. The support for employees in their work activities provided by generative AI identified in the survey shows this development direction. Jobs will indeed be lost in other professions while new professions and jobs will emerge. This transformation process is also described in studies about labour market forecasts (Schneemann et al. 2023). Here, the authors predict that jobs will be lost and new jobs will be created at the same time, leading to only a moderate change in the total number of jobs overall. However, this transformation process should not be underestimated, as it means a considerable upheaval in the demand for labour and certain skills.
In most cases, new jobs will be accompanied by new and higher skill requirements for employees. Therefore, adult education and lifelong learning become more important than ever. Further training opportunities and support options must be expanded to maintain people’s future employability. On the one hand, this applies to promoting important general skills for the jobs of the future, such as analytical and creative thinking, as well as resilience, flexibility and agility, motivation, and self-confidence (WEF 2023). On the other hand, this also applies to skills in dealing with generative AI and an understanding of this technology to promote safe and productive use. Therefore, it should also be in companies’ interest to take into account the desire to learn more about how generative AI works, expressed by many employees in this survey. Acknowledging this request could be achieved by offering suitable training courses. Companies should also consider providing clear guidelines on how to deal with generative AI – not only to meet the wishes of many employees, but also to counteract any unwanted leakage of trade secrets or other data worth protecting.
In the long term, the impact of generative AI on society and the economy will depend crucially on where and how generative AI is used and regulated. The goal that AI should be designed, developed, deployed and used in a way that is safe, human-centred, trustworthy, and responsible (Bletchley Declaration 2023), recently expressed by 28 countries in the Bletchley Declaration, must be followed by appropriate policy measures. These measures must maintain a balance between different interests and ensure that the predicted high additional gains from generative AI do not only benefit a few individuals. The efforts of the European Union to pursue a risk-based approach to the regulation of AI with the AI Act, in which risky applications, in particular, are regulated more strictly (European Parliament 2023), are, in principle, a step in the right direction in light of the study results. Not only do many people desire stronger regulation of AI, but they also fear losing control over the technology. This concern becomes more serious with the increasing risk associated with the application of generative AI.