Generative Artificial Intelligence (henceforth: GenAI) is a form of Artificial Intelligence that allows for creation of new content (text, image, video), based on large amounts of training data used to train AI models (Crawford, 2021; Whitaker, 2024). GenAI has increasingly gained popularity especially since November 2022, when Open AI released ChatGPT, a GenAI model with a chat interface that allows users to interact with the tool to create new content.

GenAI development is a controversial issue from the perspective of individual rights (use of data for training purposes see Mejias & Couldry, 2024); because of its impact on the environment and sustainability; labor market and inequality, among other issues (Humble & Mozelius, 2024; Whitaker, 2024). Hence, GenAI development is highly relevant to young people’s future and it is important to understand to what extent they have a critical understanding of the possible impact of this technology on their lives.

Young people tend to be early adopters of new technologies, often playing the role of canaries in a coalmine, (Staksrud, 2016), which makes it even more relevant to study how they are using various GenAI tools and the associated risks and opportunities. Yet, little is known about how young people engage with GenAI tools, especially in developing countries such as Serbia (Klarin et al., 2024; Yu et al., 2024).

The primary objective of this research is to understand and map how adolescents are engaging with various GenAI tools; how they might be using them for educational purposes and what consequences their use of these tools might have on their formal and informal education. We are also interested in their digital literacy with respect to these technologies: how much they know about limitations and caveats of using these technologies related to model-training; as well as where they learn about these technologies (via parental mediation, traditional media and digital media channels, formal educational channels etc.).

To that end, the EU Kids Online network^[1], a network of researchers on children and young people’s digital media use, is undertaking an international comparative qualitative research with adolescents in 16 European countries. This abstract is part of this broader effort, and it only details the results of this research undertaken in Serbia.

The theoretical framework for this study is the socio-ecological model on young people’s digital media use (Livingstone et al., 2018). This framework maps factors at the individual, family, peer and broader societal and country levels that determine whether a young person’s engagement with digital technology and any experience of risk will result in harm or benefits. The underlying premise of the model is that the experience of risk is not inherently harmful, but rather that various factors in the socio-ecological model influence the outcome of experiencing risk online.

Previous survey research on children and adolescents’ digital media use in Serbia, on a nationally representative sample, found high prevalence of risky experiences online, such as underage use of social media as well as exposure to sexual content online (the highest prevalence rates among the 19 European countries studied, see Smahel et al., 2020). At the same time, children’s levels of confidence in their digital literacy skills are among the highest in Europe (Kuzmanovic et al., 2019). Nonetheless, it is important to emphasise that these are self-assessed confidence levels and not objective digital literacy measurements.

In order to understand patterns of risks, harm and opportunities in Serbian adolescents’ GenAI use, we ask the following research questions:

RQ1: How are adolescents aged 13-17 in Serbia using Generative AI tools?

RQ2: What are the risks and opportunities that adolescents in Serbia experience when using GenAI tools?

[1] https://www.lse.ac.uk/media-and-communications/research/research-projects/eu-kids-online

In order to understand these issues, fifteen semi-structured in-depth interviews with adolescents aged 13-17 have been conducted in each of the sixteen European countries, including Serbia. This means standalone tools such as ChatGPT, Midjourney, DALLE, Claude etc. or use of GenAI tools that are integrated into other software such as Microsoft Copilot or GenAI features in social media such as Snapchat, Instagram etc. A research protocol was developed by the EU Kids Online research team which included questions about adolescents’ activities with these technologies; usage patterns and reasons behind their use (social, educational, creative, political/citizenship-related etc.); parental and other mediation of such technology use; perceptions of GenAI and emotional relationships with such technologies; hopes and fears related to their future with these technologies. The interviews lasted between 45 minutes and one and a half-hours each. The interviews were conducted both in-person and on zoom. Sampling was purposive: The recruitment criterion was at least occasional use of conversational or visual Gen AI tools. We aimed for age and gender balance (we were not able to recruit any non-binary participants). This meant that we recruited one male and female participant per each age cohort (male and female age 13, 14 etc.). Adolescents were recruited with the help of school staff in primary schools, secondary (vocational schools) and high schools. One participant was recruited as a family acquaintance of one of the researchers. We sought to have a diversity of socio-economic backgrounds in our sample. Most of the interviewees were from the capital city (Belgrade), while four participants were from other towns in Serbia. Privacy and confidentiality were protected, with child assent and parental/caregiver consent collected. Ethical review and clearance was obtained via the Ethics Committee at the London School of Economics. This research is still in progress and we have thus far conducted 14 interviews. The data is being analysed from a post-positivist methodological framework by applying a thematic analysis guided by the RQs (Braun & Clarke, 2024).

The results will contribute to the body of knowledge on adolescent GenAI use in Serbia by providing empirical evidence on specific risks and protective factors, as well as opportunities related to technology use in the context of the socio-ecological theoretical framework (Livingstone et al., 2018). Thus far, the findings indicate that adolescents, across various age groups and sexes, have a limited understanding of GenAI, how it works and the rights-based and social implications of such technological development. There is a lack of systematic school-based education about this technology and adolescents primarily learn about it via parents and peers. Some of the adolescents in the sample have used GenAI to produce school assignments, and despite threats from their teachers, their assignments were not detected as plagiarism. There are examples of uses for creative and playful purposes such as hobbies and image generation. Adolescents largely lack awareness about privacy risks of sharing personal information with GenAI when asking for advice about their physical or mental health; they are inclined to trust the information provided, even when they might double check it in a browser. Awareness of problems such as discrimination and bias is largely missing; as well as any knowledge of environmental implications of GenAI development.

Braun, V., & Clarke, V. (2024). Thematic analysis. In Encyclopedia of quality of life and well-being research (pp. 7187-7193). Cham: Springer International Publishing. Crawford, K. (2021). The Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press. Humble, N., & Mozelius, P. (2024). Generative Artificial Intelligence and the Impact on Sustainability. In The 4th International Conference on AI Research, ICAIR 2024 (Vol. 4). Klarin, J., Hoff, E. V., Larsson, A., & Daukantaitė, D. (2024). Adolescents' use and perceived usefulness of generative AI for schoolwork: exploring their relationships with executive functioning and academic achievement. Frontiers in Artificial Intelligence, 7, 1415782. Kuzmanović, D., Pavlović, Z., Popadić, D., & Milosevic, T. (2019). Internet and digital technology use among children and youth in Serbia: EU Kids Online Survey Results, 2018. Retrieved from: https://www.lse.ac.uk/media-and-communications/assets/documents/research/eu-kids-online/participant-countries/serbia/EU-Kids-Online-ENG-2019.pdf Livingstone, S., Mascheroni, G., & Staksrud, E. (2018). European research on children’s internet use: Assessing the past and anticipating the future. New media & society, 20(3), 1103-1122. Mejias, U. A., & Couldry, N. (2024). Data grab: The new colonialism of big tech and how to fight back. In Data Grab. University of Chicago Press. Smahel, D., Machackova, H., Mascheroni, G., Dedkova, L., Staksrud, E., Ólafsson, K., Livingstone, S. and Hasebrink, U. (2020). EU Kids Online 2020: Survey results from 19 countries. EU Kids Online. https://doi:10.21953/lse.47fdeqj01ofo Staksrud, E. (2016). Children in the online world: Risk, regulation, rights. Routledge. Tlili, A., Shehata, B., Adarkwah, M. A., Bozkurt, A., Hickey, D. T., Huang, R., & Agyemang, B. (2023). What if the devil is my guardian angel: ChatGPT as a case study of using chatbots in education. Smart Learning Environments, 10(1), 15. https://doi.org/10.1186/s40561-023-00237-x United Nations Children's Fund (UNICEF) (2021). Policy guidance on AI for children 2.0. Retrieved from: https://www.unicef.org/innocenti/media/1341/file/UNICEF-Global-Insight-policy-guidance-AI-children-2.0-2021.pdf Whitaker, M. (2024, May 15). Inaugural Speech, Helmut Schmidt Prize. Retrieved from: https://www.helmut-schmidt.de/en/news-1/detail/the-prizewinners-speech Yu, Y., Sharma, T., Hu, M., Wang, J., & Wang, Y. (2024). Exploring Parent-Child Perceptions on Safety in Generative AI: Concerns, Mitigation Strategies, and Design Implications. arXiv preprint arXiv:2406.10461.