Swedish teenagers aged 13-16 understand AI through nine distinct conceptual frameworks ranging from helpful tools to scary robots, revealing significant gaps in AI literacy that vary dramatically across cultural contexts.
Objective: The main goal of this study was to investigate how Swedish students aged 13-16 conceptualize and understand artificial intelligence, and to compare these perceptions with previously collected data from Azerbaijani students to examine cultural and contextual differences in AI understanding among young learners.
Methods: The researchers employed a comprehensive qualitative survey approach using Google Forms to collect data from 75 Swedish secondary school students in the Kalmar and Växjö regions over a two-month period. The survey included demographic questions and five open-ended prompts adapted from previous research, asking students to describe what they think AI means, what it is used for, how it works, why it is used, and to name related words or objects. Each student was required to provide at least two sentences per prompt, generating 375 individual responses. The researchers applied BERT-based topic modeling using the BERTopic technique, which leverages Bidirectional Encoder Representations from Transformers (BERT) embeddings combined with class-based TF-IDF and Unified Manifold Approximation and Projection (UMAP) for dimension reduction. This approach used Hierarchical Density-Based Spatial Clustering of Noise Applications (HDBSCAN) algorithm to identify latent themes without requiring a predetermined number of topics.
Key Findings: The analysis identified nine distinct topics reflecting how students conceptualize AI: (1) "AI" - general associations with helpful, intelligent, human-like entities; (2) "Use" - utilitarian framing focusing on daily life applications; (3) "Help" - emphasis on AI's assistance-oriented functions; (4) "Robot" - anthropomorphic associations equating AI with robots and intelligent machines; (5) "Just like" - comparisons suggesting AI is similar to or indistinguishable from humans; (6) "Don't know" - open expressions of uncertainty and ignorance; (7) "Computer" - associations with generic computing technologies; (8) "LLMs" - specific awareness of conversational AI systems like ChatGPT; and (9) "Cool but scary" - emotional ambivalence expressing both appreciation and concern about AI's potential consequences. The hierarchical clustering revealed that topics like "Use" and "Help" were closely related, while "Cool but scary" was nearest to "Don't know," suggesting fear often stems from lack of understanding. Students who received smartphones as early as age five showed more intuitive, tool-oriented understanding of AI, with the mean age of first phone acquisition being 8.3 years.
Implications: This research contributes significantly to AI literacy education by providing evidence-based insights into student conceptions and demonstrating how sociocultural factors shape AI understanding. The findings emphasize the critical need for contextually grounded AI literacy curricula that address diverse student perceptions, from anthropomorphic misconceptions to uncertainty gaps. The study highlights the importance of incorporating explainable AI and hands-on tools like Teachable Machine or MachineLearning4Kids to help students develop accurate understandings of AI's inner workings. The research also underscores the necessity of addressing ethical considerations in AI education, as students expressed both excitement and concerns about AI's potential misuse and impact on employment.
Limitations: The study acknowledges several important limitations that may affect the generalizability of findings. The sample may not be fully representative of the Swedish student population, as 93.3% of participants attended private schools, which could influence both technology access and AI engagement patterns. The brevity of many open-ended survey responses may have limited the depth of interpretation possible through topic modeling. Additionally, the study's scope was confined to a specific age group (13-16 years) and geographic regions within Sweden, potentially limiting broader applicability across different educational contexts and cultural settings.
Future Directions: The researchers suggest several promising avenues for future investigation, including combining topic modeling approaches with interviews or classroom-based longitudinal studies to better understand students' reasoning and conceptual development over time. Future research should explore how early technology exposure and digital literacy influence AI understanding across various demographics and educational backgrounds. The study recommends investigating the long-term impact of integrating explainable AI tools into K-12 curricula and examining how different cultural and educational contexts shape AI perceptions. Additionally, researchers should explore the effectiveness of hands-on, transparency-focused AI education tools in addressing misconceptions and building more accurate conceptual frameworks.
Title and Authors: "How Do 13–16-Year-Olds Understand AI? A Topic Modeling Study of Swedish Students' Perceptions" by Mirka Saarela (University of Jyväskylä), Ahmed Taiye Mohammed (Linnaeus University), Johanna Velander (Linnaeus University), and Rafael Zerega (Linnaeus University).
Published On: 2025
Published By: 2025 3rd Cognitive Models and Artificial Intelligence Conference (AICCONF), IEEE