Generative AI passes exams and is used to create homework assignments. The bidt took this as an opportunity to survey 252 adult internet-using pupils and 981 adult internet-using students on their use of generative AI. Generative AI is a collective term for systems with artificial intelligence that generate new content such as texts, images, programme code, videos or music. The results for adult learners at schools and universities in Germany are presented below. Questions regarding the use, regulation and expected impact of generative AI on the education sector are examined.
Generative artificial intelligence (AI) passes the Bavarian Abitur (Tagesschau 2023), the American bar exam (ZDF.heute 2023) and the text questions of the first and second medical state exams in Germany (Ärzteblatt 2023). These are just a few examples of how generative AI is causing a stir in the school and university sector and how great the transformative potential of this technology is in the education sector. Against this background and the associated increasing importance of generative AI, the bidt investigates questions regarding the use, application and regulations of generative AI in schools and universities.
To this end, the bidt hired the market research institute DCORE to conduct a quantitative survey using a standardized online questionnaire. From 20 July to 4 August 2023, DCORE surveyed 3,020 internet users in Germany, including 252 pupils aged 18 and over and 981 students aged 18 and over. The student group also includes working people who are only part-time students. The following results are representatively weighted for the population of internet using pupils and university students of legal age living in Germany, with weights based on age, gender, and federal state. For separate analyses, we developed group-specific weights for pupils aged 18 and over who use the internet, as well as for the group of internet using adult students; also based on age, gender and federal state (BMBF, Destatis). Therefore, the results presented in this study generally reflect the population of school pupils and students aged 18 and above living in Germany who use the internet. For sub-group analyses, there is further specification provided.
Widespread use of generative AI among pupils and students
The vast majority of pupils (92%) and students (98%) aged 18 and over have already heard of generative AI for creating texts and/or program code as well as videos and images. Also, the use of generative AI is widespread among pupils and students. For example, 73% of the pupils surveyed and 78% of the adult students have already used generative AI to generate text and/or images.
A total of 68% of the pupils of legal age surveyed have already used generative AI at least once to create and/or check texts or program code, 39% have even used it several times. Of the students surveyed, a total of 71% have used text-based generative AI at least once, 38% several times. At the same time, 12% of the pupils and 5% of the students surveyed state that they have not yet heard of text-based generative AI.
When looking at the repeated use of text-based generative AI among all learners aged 18 and over, there are clear differences by gender: repeated use is more common among male learners (43%) than among female learners (33%). One-time use was equally common among the male and female learners surveyed, at one third each.
However, for using generative AI to create images or videos, there is a gender difference, even for one-time use. Male pupils and students used image-generating generative AI both once (38%) and several times (25%) more frequently than female learners. Among the females, 26% state that they have already used image-generating generative AI once and 16% say that they have used it several times.
Purpose of use depends on the context of use
Of the adult pupils surveyed who have already used text-based generative AI, 89% use it for both school and other purposes, like in a professional or private context. For the corresponding students aged 18 and over, this proportion is 78%. Only 3% of the pupils examined here do not use text-based generative AI in a school context. At 16%, the proportion of students who do not use these AI systems as part of their studies is considerably higher.
The data shows clear differences in the purposes for which text-based generative AI is used between schools and universities. While adult pupils most frequently use text-based generative AI at school to summarise texts (68%), students most frequently use text-based generative AI at university to research information (59%). The least common use of text-based generative AI among the pupils examined was to conduct conversations (16%). This purpose of use is clearly more common among students with a share of 32%.
In terms of private use, the group differences are generally smaller. Adult pupils who use text-based generative AI in a private context utilise these applications most frequently for conducting conversations (53%) and researching information (50%). Among the students considered here, researching information (54%) and summarising texts (45%) are the most common purposes of use in the private setting. Consequently, different requirements at school or university appear to lead to more heterogeneous usage behavior in these areas than in the private context.
Majority of users of generative AI in an educational context benefit from its use
Satisfaction with the use of text-based generative AI at schools or universities is high, only a minority of the users surveyed reported the opposite. The majority of the adult pupils (52%) and students (55%) surveyed, who have already used text-based generative AI at school or university, agree with the statement that they were able to make good use of the results obtained from generative AI. This statement is rather or not at all true for only 19% of the corresponding pupils and 15% of the corresponding students.
Furthermore, 54% of the adult students and 55% of the adult pupils who use text-based generative AI in their studies or at school agree that the use of generative AI has made learning easier for them. This statement is somewhat or not at all true for 17% of the corresponding students and 13% of the relevant pupils. 62% of the considered students and 59% of the considered pupils report that the use of generative AI has saved them time.
The statement that the use of generative AI at schools or universities leads to an increase in performance applies to around half of the pupils and students aged 18 and over, who have already used text-based generative AI at school or university. However, just under one in five of them disagree with the statement. Further, 42% of the pupils and 45% of the students in question somewhat or fully agree that generative AI has helped them to achieve better grades without having to perform adequately. This is not the case for 18% of the corresponding pupils and 27% of the corresponding students.
There is a clear association between the frequency of use of text-based generative AI and the perceived benefit. Across all the satisfaction statements surveyed, the following applies on average: learners who have already used text-based generative AI several times at school or university agree more often to those statements than those who have only used such systems once. The average difference is 17 percentage points. This difference is most pronounced for the statement that the use has saved time. Here, 72% of the multi-time users, but only 50% of the one-time users surveyed, agree. The data shows the smallest difference (12 percentage points) for whether the use of generative AI at school or university has helped them to get better grades without having to perform adequately. Here, among the learners surveyed, 48% of the multi-time users and only 36% of the one-time users agree. One possible explanation of why satisfaction with text-based generative AI increases the more it is used may be because developing skills for the application of generative AI is built up over multiple usages. Specifically, over time, better and more refined prompts can contribute to improved and more targeted results. Furthermore, people who see a clear benefit for themselves at the very first application will likely use generative AI more frequently.
Interest in deeper understanding increases with level of knowledge
According to self-reporting, 49% of the adult pupils surveyed who have already heard of generative AI for text or image creation state that they know and understand the basics of generative AI systems. Among the corresponding university students, this figure is slightly higher at 54%. In contrast, for 16% of these pupils and students, this statement is rather or not at all true. Even among all the learners surveyed who have already used generative AI, 13% disagree with the statement that they know and understand the basics of generative AI. Thus, those learners use a technology whose basics they say they do not understand. However, the proportion of learners who state they know and understand the basics of generative AI increases with the frequency of use of such systems.
Almost half (47%) of the surveyed pupils aged 18 and over, who have already heard of text- or image-based generative AI, agree with the statement that they can recognize when generative AI is helpful in a task. However, just under a fifth (19%) say the opposite. Among the corresponding university students, a slightly larger proportion are confident that they can decide when generative AI is helpful and when it is not: 53% of them somewhat or strongly agree with the statement, while the opposite is true for 14%. Users of generative AI state more frequently that they can assess the applicability of generative AI in terms of usefulness, compared to non-users. Again, agreement with the statement increases with the frequency of use.
44% of pupils and 50% of students aged 18 and over, who have already used generative AI, state that they know how generative AI can be used depending on the task to obtain suitable results. Simultaneously, however, 26% of the corresponding pupils and 17% of the respective students somewhat or completely disagree.
The statement "I would like to better understand how generative AI systems work" applies either somewhat or completely to 45% of pupils aged 18 and over and 57% of the students surveyed. Of the corresponding pupils, 20% and 14% of the students state the opposite. This finding shows that although the pupils in question have less knowledge about generative AI and how it is used, their desire to build up knowledge is also less pronounced than among the university students surveyed.
Looking at learners aged 18 and over who have at least heard of image- or text-based generative AI, there is generally the pattern that those who do not know and do not understand the basics of generative AI are also less interested in gaining a better understanding, compared to those who already have a basic understanding of the technology.
University students more likely to see opportunities in use of generative AI than adult pupils
In the general opportunity/risk assessment of the application of generative AI, 22% of all pupils aged 18 and over believe that the risks outweigh the opportunities. Among the university students surveyed, this figure is only 16%. The adult pupils surveyed are also more cautious when it comes to the opportunities of using generative AI. Of them, 20% state that the opportunities of using generative AI outweigh the risks. Among the students surveyed, 29% do so.
For the use of generative AI in education specifically, a total of 32% of the university students surveyed see opportunities. Among the pupils surveyed, 30% are of this opinion. This means that the application of generative AI in the education sector in particular is more frequently associated with opportunities, compared to general application. At the same time, however, the perception of risk is also more pronounced. 25% of the pupils and 20% of the university students surveyed see the use of generative AI in education as predominantly associated with risks.
Overall, the pattern seen in previous studies for other population groups (Stürz et al. 2022, Schlude et al. 2023) is also visible among learners: Those who state that they know a lot about AI or those who have already used generative AI themselves are also more likely to see the opportunities of this technology. For example, of all learners aged 18 and over with experience of using generative AI, 36% predominantly see opportunities for its application in education. In contrast, this only applies to 17% of the learners surveyed who have never used generative AI.
Critical engagement with results of generative AI more pronounced than concerns about data privacy
26% of the adult pupils and 38% of the students surveyed, who have heard of generative AI, somewhat or strongly agree with the statement that they review the data protection guidelines of generative AI that is being used or will be used in the future. This statement is slightly more common (35%) among learners who have already used generative AI themselves, compared to learners without usage experience (33%). 38% of the surveyed pupils and 45% of the students surveyed report that they carefully consider which data they provide when using generative AI. At the same time, this is not the case for 31% of the adult pupils and 20% of the university students.
Half of the pupils aged 18 and over as well as 56% of the students surveyed state that they know that generative AI can sometimes produce factually wrong results. Of all the learning users of text- and image-generating generative AI, 55% are somewhat or fully aware that the results of generative AI can be factually incorrect. For 17%, this is rather or not at all the case.
40% of the pupils and 58% of the students surveyed stated that they are aware that generative AI can produce incomplete, unbalanced, contradictory and inappropriate results. At the same time, a quarter of the corresponding adult pupils and 15% of the university students in question disagree or strongly disagree with this statement.
Among adult users of text-based generative AI, 59% of pupils and 58% of university students state that they have verified the correctness of the results created by generative AI. This statement applies somewhat or not at all to 11% of the corresponding pupils and 14% of the corresponding students. Differentiated by frequency of use, among adult learners at schools and universities, 10% of the one-time users and 18% of the multi-time users state that they have somewhat or not at all checked the generated results for correctness.
Rarely clear and helpful standards or guidelines for the use of generative AI in education
Of the surveyed students who have already heard of or used generative AI, 45% report that the topic of generative AI receives a high to very high level of attention at their university. With 40%, this percentage is slightly lower among the surveyed pupils.
Despite the sometimes high level of attention paid to the topic and the widespread use of generative AI among learners, some teachers and lecturers lack awareness of its application. For instance, 31% of the adult pupils surveyed and 35% of the university students surveyed, who have already used text-based generative AI for school or university, state that the teaching staff is unaware of its use. In contrast, slightly more than half of the corresponding learners at schools or universities report that their teachers and lecturers are informed about the usage of generative AI.
In addition, over half of the adult pupils surveyed, who have at least heard of generative AI, report that there are no standards or guidelines for its use at their school. Among the surveyed students, this figure is slightly lower at 41%. In general, learners who perceive a high level of attention to the topic of generative AI at their educational institution are more likely to report the existence of standards or guidelines for the use of generative AI in their school or university.
However, such standards and guidelines for the use of generative AI are not always clear or helpful, according to the respondents. For example, only 61% of the learners, who have already heard of generative AI and have standards or guidelines for the use of generative AI at their educational institution, agree or strongly agree that these guidelines clearly regulate how generative AI should be used. However, 17% of those learners disagree or strongly disagree with this statement. For around half of these learners, it is rather or completely true that the standards and guidelines help with the use of generative AI. For 20% of them, this is rather or not at all true.
Overall, 61% of the adult learners who report the existence of standards or guidelines and who use generative AI at schools or universities, indicate that they tend to or fully comply with these standards and guidelines. 16% of the corresponding learners follow them rather or not at all.
44% of pupils age 18 and above and 57% of university students who have already heard of generative AI and state that they have no standards or guidelines for the use of generative AI at their school or university would like to have such regulations. Further, 40% of these pupils and 51% of these students assume that such regulations would help them decide how generative AI should be used in the context of schools or universities.
The desire for more rules among the university students surveyed is also reflected in the desire for more controlled examination formats. For example, 47% of them somewhat or strongly agree with the statement that more controlled examination formats, such as oral examinations or supervised closed-book in-person exams, are needed. This value is only 38% for the corresponding adult pupils. In general, users of generative AI surveyed are more likely to agree with the statement than non-users.
42% of the adult pupils and 37% of the university students are somewhat or completely opposed to a general ban on the use of generative AI in school or university contexts. In contrast, 28% of these pupils and 31% of these students surveyed are somewhat or fully in favor of such a ban. Here, there is only a slight difference in whether generative AI has already been used or not.
Majority sees use of generative AI as a challenge for the education system
The majority (52%) of the university students surveyed believe that the use of generative AI poses significant challenges for the education system, while only 45% of the pupils aged 18 and above share these concerns. At the same time, 20% of this specific group believe that the use of generative AI will not pose any major challenges for the education system.
Conclusion
Text-based generative AI can not only pass final exams but, as this survey shows, is already used by around 6 in 10 adult learners in their everyday school and university life. According to the respondents, the results created by generative AI are predominantly useful, save time, and lead to higher performance among adult pupils and university students. Thus, the question is no longer whether generative AI will or should enter classrooms and universities, but rather how this technology should be handled there. The majority of learners at schools and universities surveyed in the study are aware that the use of generative AI will pose major challenges for the education system.
One challenge is to establish clear and helpful guidelines for the application of generative AI in schools and universities. So far, there has often been a lack of such rules. At the same time, many adult pupils and university students would like to see regulations and guidelines on the use of generative AI to better understand when and how to utilise this technology.
A second challenge is the need to develop skills in dealing with generative AI, among both learners and educators. The survey data shows that at least some of the adult learners at schools and universities have so far used the technology without basic knowledge of its functionalities and its possible limitations. However, the study also shows that respondents with a better understanding of the technology predominantly see opportunities in the use of generative AI, and that learners who have already used generative AI several times also benefit more from its use. Thus, if not all learners are taught the necessary knowledge and skills in dealing with generative AI at an early stage, there is a risk that the digital divide among groups of the population continue to increase. Specifically, if only those benefit who already have access to digital devices and generative AI and have already been able to use them, other groups of the population will initially lag behind in school and university and later in the workplace. Since in the professional setting, skills in dealing with generative AI will also become increasingly important (Schlude et al. 2023). In this context, educational institutions have a dual responsibility: first, teaching the appropriate skills in dealing with the technology and, second, using generative AI in such a way that learning and teaching are further developed and improved. Therefore, teaching staff play an important role: on the one hand, they should impart the necessary skills to pupils and students. On the other hand, educators also require knowledge of appropriate areas of application for generative AI in order to benefit from the technology in the class room and, for example, to better meet the needs of individual learners. Simultaneously, teachers can also benefit from generative AI when preparing lessons or courses. For example, some federal states are already relying on a start-up that offers further training for teachers as well as AI assistants for teaching (News4teachers 2024). The expansion of further training for teachers and lecturers in this area should receive special attention in all federal states, as should the expansion of a corresponding interdisciplinary range of courses at schools and universities.
A third challenge consists of the adequate assessment of learners' performance in times of generative AI. This is because determining whether performance was provided by learners themselves or by generative AI is often difficult, and in many cases, even impossible. Therefore, examination formats at schools and universities should be adapted, as requested by many respondents. For example, the increased use of oral examinations can make performance assessment fairer. Thereby, there is a need to take into account the increased resource investment compared to non-supervised forms of examination, particularly at universities.
A fourth challenge will be to keep pace with the rapid development of technology and to react quickly with suitable measures. Overall, educational institutions need to become more flexible and agile. Furthermore, appropriate monitoring is necessary to recognise developments at an early stage and enable corresponding reactions.
