Exploring AI Literacy: Voice Recognition Project in Vocational Education

Abstract

This study examines how a voice-recognition project may support vocational secondary students’ AI literacy. In this applied scenario, students used Arduino hardware and an AI tools platform to collect data, train models, and deploy a basic voice-recognition device, linking introductory AI concepts with practical engineering applications. A mixed-methods design combined pre–post self-report assessment using the AI Literacy Questionnaire (AILQ) with post semi-structured interviews. Emerging gains were associated with the maker-learning pathway, particularly in the affective, behavioral, and cognitive AI literacy domains, whereas ethical outcomes were limited within this intervention window. Qualitative insights provided complementary interpretive context, suggesting that learning through making was experienced as more engaging and personally relevant, while hands-on linked with emerging understanding of AI model behavior and limitations. Overall, the study extends AI-literacy research to a vocational classroom setting, where evidence remains limited. It also highlights a domain-level AI literacy analysis for identifying which components strengthen through making and which may require more explicit instructional scaffolding in this specific vocational context. The exploratory nature of the study offers evidence that maker activities can provide a feasible approach for engaging vocational learners with multidimensional AI literacy.

1. Introduction

The rapid diffusion of artificial intelligence (AI) across everyday technologies has intensified global efforts to cultivate AI literacy among young learners [1,2,3]. Beyond tertiary education, vocational secondary schools are increasingly expected to help students understand how intelligent systems operate and influence society [4]. AI literacy is therefore emerging as a fundamental competence for navigating modern technologies.

Following recent research, this study adopts a contextual view of AI literacy, conceptualized as a discipline and role-sensitive capacity to understand, interact with, and make informed, ethically responsible decisions about AI within a specific educational or professional setting [4,5,6]. Contextual AI literacy involves four interrelated domains—cognitive, behavioral, affective, and ethical—that enable learners to interpret, use, and evaluate AI systems responsibly [7]. Comparable frameworks reinforce this multidimensional perspective, linking contextual AI literacy to sustainable, ethical, and socially informed technology engagement [8].

Despite increasing global interest, AI literacy initiatives remain unevenly implemented. Studies mainly focus on university-level settings [9,10], while classroom-based studies in secondary education, and particularly in vocational settings, remain limited [11,12,13,14]. In Greece, vocational secondary students specializing in electronics, automation, or computing may have limited opportunities to engage with authentic AI workflows during school practice, despite the applied orientation of their curricula. This underscores the need for feasible classroom designs that embed AI workflows in vocational tasks.

In vocational education and training (VET; often referred to internationally as Career and Technical Education, CTE), engagement in technology-rich STEM interventions depends on learners’ motivation, while perceived readiness for further study and careers is partly shaped by self-belief of success in challenging tasks [15]. Prior studies also suggest that vocational pathways may differ in self-esteem and academic self-concept [16,17]. Such differences may shape how learners approach challenging tasks and how confident they feel while working with emerging technologies. Since engagement is linked to completion and learning trajectories [18], VET interventions involving emerging technologies (e.g., sensors and AI) should be evaluated not only for what students learn, but also for how they support motivational engagement and perceived competence.

Recent evidence from technology-oriented secondary/VET-adjacent settings suggests that AI literacy can be meaningfully profiled through “big ideas” lenses and linked to learner agency in technology education, reinforcing the need for contextualized AI literacy development in applied school settings [19,20]. In parallel, evidence from AI-supported teaching indicates that AI literacy is associated with learners’ engagement, acting as a key mechanism connecting AI use to skill-oriented outcomes, including creativity and applied design performance [21,22].

Maker Education offers a promising response to this educational challenge. Rooted in constructionist learning theory, maker pedagogies promote knowledge construction through design, fabrication, iteration, and experimentation with real artifacts [23,24]. Recent studies demonstrate that maker-based experiences foster collaboration and sustained engagement in STEM subjects. These benefits may be especially salient when AI tools are incorporated into maker activities, enabling learners to investigate how intelligent systems learn and operate through hands-on experimentation [25,26]. Building upon these insights, the present study introduces an educational scenario in which vocational secondary students implement a machine-learning voice-recognition device using Arduino hardware (Nano 33 BLE Sense) and the Edge Impulse web-based platform for developing and deploying machine-learning models on embedded devices. From a digital-technology perspective, the intervention operationalized an end-to-end embedded AI workflow: students generated a small labeled voice dataset, trained and iteratively refined a model in the web-based platform environment, and tested real-time inference on-device, making model behavior and limitations observable. Through this process, abstract AI concepts were linked with functional technological practice.

Unlike many maker-based AI activities that do not include embedded deployment and real-time inference on-device, this study examines a classroom-feasible, machine-learning workflow in a vocational setting and reports a domain-level AI literacy profile (cognitive, behavioral, affective, ethical), interpreted as context-bound patterns in students’ self-perceptions rather than generalizable effects. Its aim is not generalization or direct transferability; rather, it seeks situated understanding within this specific vocational context and offers exploratory evidence relevant to feasible approaches for introducing emerging technologies in education.

2. Literature Review

2.1. AI Literacy as a Multidimensional Construct

Artificial intelligence literacy (AI literacy) refers to the set of knowledge, skills, attitudes, and values that enable learners to understand how AI systems work, interact with them, and make informed, ethically responsible decisions about how AI should be used. Several frameworks have converged on the idea that AI literacy is not a single component, but a multidimensional construct. Across recent studies, there is strong agreement that AI literacy is a multidimensional construct comprising at least four core components, cognitive, behavioral, affective, and socio-ethical, each essential for developing a well-rounded understanding of AI [2,8,11,27,28,29]. The cognitive dimension involves understanding the basic ideas behind AI and machine learning (e.g., training with data; model strengths and limitations). The behavioral dimension addresses what learners can do with AI tools (e.g., collecting data, applying models, using AI in tasks). The affective dimension captures learners’ interest, curiosity, confidence, and motivation. The socio-ethical dimension concerns risks and benefits (e.g., bias, privacy, fairness) and responsible decisions about AI use. Recent work has refined these ideas into validated frameworks and instruments for school and university students [10]. For example, a project-based AI literacy course for senior secondary students strengthened cognitive, metacognitive, affective, and social dimensions of AI literacy [11], while AI maker projects have been shown to support both lower and higher cognitive domains, including ethics and collaboration [12].

In this study, we adopt a contextual view of AI literacy, emphasizing the knowledge and skills students need to use AI tools meaningfully and responsibly within the educational and professional context of vocational STEM education. The importance of contextual AI literacy is particularly pronounced in vocational settings. Vocational students’ AI literacy has been linked to digital learning sustainability and practical achievement [30], while reviews note uneven implementation across age groups and levels and call for more school-level evidence, particularly in applied, hands-on environments such as vocational education [14,27,31]. Aligned with this literature, the present study uses a four-domain definition of AI literacy (cognitive, behavioral, affective, and ethical) and examines it through the AI Literacy Questionnaire (AILQ) [32]. In the present study, AI literacy is examined as students’ self-reported, context-sensitive capability, and findings are interpreted as indicative patterns rather than generalizable effects. By adopting this multidimensional and contextualized view, we position AI literacy not only as knowing about AI but as being able to engage with AI technologies in a competent, motivated, and ethically informed manner within vocational STEM education.

2.2. From Constructionism to Maker Learning

Maker Education, grounded in constructionism [23], is a learning approach in which students develop understanding through the design and creation of tangible artifacts that are personally meaningful. Rapid technological advances and affordable fabrication tools gave rise to what became known as the Maker Movement, which expanded into formal schooling, supporting creativity, agency, and computational thinking [24]. Situated within the global Maker Movement, this approach emphasizes creativity, design, iteration, and the use of physical and digital tools to explore ideas and solve problems in personally meaningful ways [33]. Maker Education positions learners as active creators who learn through making, designing objects, testing prototypes, manipulating materials, and refining their solutions as their understanding evolves. This process encourages students to see themselves as capable creators who can shape ideas into concrete outcomes and take ownership of their learning trajectories [34]. In this study, Maker Education is treated as a constructionist pedagogical approach characterized by iterative design–build–test cycles, collaboration, and learning through the creation of meaningful artifacts within tool-rich environments.

Educational research describes making as a spectrum ranging from open-ended tinkering to purposeful design, where learners follow engineering-like cycles of planning, constructing, testing, and revising. These modes together support learners as they navigate through uncertainty to try new approaches and develop confidence in their creative and technical abilities [34,35]. Recent reviews show a shift from unstructured tinkering toward more standards-aligned, curricular implementations in secondary education, where Maker activities are embedded into formal STEM courses and project-based learning experiences [26]. When making is supported by computing environments that incorporate programmable hardware and sensing components, learners are able to externalize their thinking and engage in iterative cycles of construction and evaluation that mirror authentic engineering practices [36,37]. Such technology-rich environments promote engagement with interdisciplinary STEM concepts as they design systems that integrate coding, electronics, and scientific reasoning [25,38].

Within STEM education, making brings together physical construction, programming, and scientific reasoning, enabling students to test ideas in concrete ways and receive immediate feedback from the artifacts they build. These experiences cultivate essential competencies for STEM practice, including technological creativity, adaptability, teamwork, and the ability to reason across disciplinary boundaries [39]. In this sense, Maker Education helps bridge the gap between theoretical knowledge and real-world application by enabling learners to explore scientific ideas through hands-on creation [40]. Recent reviews characterize Maker Education as a constructionist synthesis of project, problem-, and inquiry-based learning situated within technology-rich, collaborative contexts [33,41].

Beyond cognitive gains, maker learning is frequently associated with affective outcomes central to this study, cultivating learners’ confidence, persistence, and a strong sense of agency as they experience themselves as capable problem solvers [37,42]. Makerspaces, whether formal classrooms, fab labs, or informal community environments, provide supportive, low-risk settings in which students can experiment freely, take intellectual risks, and learn through cycles of trial and error. These school-based makerspace environments foster motivation and a willingness to approach complex tasks rather than avoid them—attitudes closely aligned with problem-solving confidence and approach tendencies [43]. Maker learning also supports engagement with emerging technologies when tools such as microcontrollers, sensors, and AI platforms amplify opportunities to test ideas and refine designs [7,44]. The utilization of Arduino boards is a prime example of inexpensive and easily accessible technology that connects theoretical coding with concrete results [36,45]. At the same time, the literature notes that access, resources, and inclusion can shape participation and benefits, which is especially relevant in vocational contexts [46]. For vocational learners who value applied and hands-on learning, such activities may provide a practical link between engineering-oriented content and emerging digital competences by making AI workflows observable through tangible artifacts.

2.3. Maker Learning and the Development of AI Literacy

As artificial intelligence increasingly shapes scientific work, industry applications, and everyday technologies, schools are under growing pressure to help students understand how AI systems are built, how they behave, and how they can be critically evaluated [47]. A growing body of studies suggests that AI literacy can be supported through active, experiential learning, where students engage directly with AI tools rather than only studying them theoretically [11,44,48,49]. However, while conceptual frameworks for AI literacy have expanded rapidly, practical models for school-based implementation remain uneven across educational systems [44]. Questions persist regarding how to structure age-appropriate AI literacy activities, how to balance conceptual depth with accessibility, and how to support teachers with limited AI expertise [50].

Maker learning is increasingly discussed as a promising approach for addressing these challenges. Maker activities incorporating AI components, such as embedded sensors, data-collection routines, and simple machine-learning models, can enable learners to design intelligent artifacts that respond to real-world inputs. When students train and deploy machine learning models within artifacts, they encounter iterative features of AI development (e.g., data quality, model performance, and refinement). Early studies show that such activities may help students connect abstract principles to observable system responses, supporting their conceptual understanding and realistic expectations about AI capabilities and limitations [7,11,51]. Maker environments can provide opportunities for learners to engage across the four AI literacy domains highlighted in contemporary frameworks. In this context, making can support (a) conceptual reasoning about data and models, (b) practical AI-related actions such as collecting datasets and testing outputs, (c) affective engagement such as interest and confidence, and (d) ethical reflection when system behavior raises questions about fairness or responsible use [8,27,31,48].

Cloud-based machine-learning tools such as Edge Impulse, Teachable Machine, and lightweight AutoML systems can support these learning experiences by offering interfaces through which students can collect data, train models, and deploy them on embedded devices without advanced mathematical or programming prerequisites. These tools can enable students to visualize training data, adjust parameters, and observe how models respond in physical contexts. Such tangible interactions make visible key AI principles (e.g., noise, confidence levels, threshold), helping students understand both the logic and limitations of intelligent systems. However, simplified interfaces may obscure important modeling decisions, so guided reflection remains essential for understanding limits and variability in performance [44,50].

Although interest in maker learning is increasing, vocational secondary settings remain underexplored despite being particularly well suited to applied, hands-on engagement. Empirical evidence from vocational classrooms remains limited, particularly for how students engage across affective, behavioral, cognitive, and ethical domains with such activities. In response to this gap, the present study adopts an exploratory classroom-based approach to examine how students’ AI literacy evolves within a maker learning context. Based on the above, the study addresses the following research questions:

• RQ1: How do pre–post changes in vocational secondary-school students’ AI literacy scores differ between the maker learning approach and traditional instruction?
• RQ2: What are vocational secondary-school students’ perceptions of learning through the maker activity?

To ensure methodological coherence, RQ1 is addressed through pre–post administration of the AI Literacy Questionnaire (AILQ), while RQ2 is addressed through post-intervention semi-structured interviews analyzed thematically. Through this dual focus, the study provides exploratory evidence that can inform future research and pedagogical design related to maker learning in vocational STEM education.

3. Methodology

3.1. Participants

The participants of the present study were students aged 15 to 16, of a public vocational upper secondary school (EPAL) in Greece (wider Athens area), attending the Electrical, Electronics and Automation specializations. EPAL schools in Greece offer three-year programs combining core STEM education with technical and laboratory training, preparing students both for further studies and direct entry into the labor market. The participants (N = 38; 94% male, 6% female; Maker Learning: n = 19 [18 male, 1 female], Traditional Learning: n = 19 [18 male, 1 female]) were enrolled in the same grade level and followed the same school curriculum, which ensured a relatively comparable educational background and exposure to STEM subjects.

3.2. Research Methodology

The research utilized a mixed-methods design, combining quantitative and qualitative data to obtain a more comprehensive insight into how the educational intervention influenced students’ AI literacy [52]. The quantitative component consisted of the AI Literacy Questionnaire (AILQ), administered at two key stages of the intervention, while the qualitative component consisted of exploratory semi-structured interviews conducted after its completion. The two components served complementary purposes: the questionnaires captured students’ self-reported AI literacy, and the interviews provided insight into students’ experiences and perceptions during the project. A side-by-side comparison during interpretation was used to contextualize the findings [53].

The intervention was structured in two instructional phases, each serving a distinct pedagogical and methodological role. Before the start of Phase I, all students completed the pre-test questionnaire (pre-test) to establish their initial AI literacy. Phase I introduced foundational concepts to ensure that all students began the project with comparable baseline knowledge (Arduino programming, sensor integration, and introductory machine-learning procedures). In Phase II, students engaged with the same educational scenario through two distinct learning modalities: a Maker Learning (ML) instructional group in hands-on prototyping and a Traditional Learning (C) instructional group in teacher-guided instruction. Placement into the ML and C pathways was not treated as fully randomized at the individual level but followed the existing classroom structure. Classroom-related influences cannot be fully excluded and include potential differences in peer dynamics (e.g., uneven participation), teacher guidance and scaffolding (e.g., the amount of feedback), and contextual constraints. Therefore, any between-pathway comparisons are interpreted cautiously as descriptive associations rather than causal effects. This structure enabled an exploratory comparison of pre–post changes in AI literacy across instructional modalities. After completing Phase II, all students completed the post-test questionnaire (post-test), enabling the examination of changes across the entire intervention. Figure 1 presents a schematic overview of the research design and instructional phases.

3.3. Learning Environment and Instructional Design

The educational intervention was implemented in a blended learning environment resembling a school-based makerspace (in-class hands-on work supported by digital simulation and ML platforms). This environment supported hands-on experimentation and provided access to sensors, electronic components, multimeters, circuit boards, soldering equipment, signal function generators, oscilloscopes, and general-purpose laboratory tools. In addition, simulation platforms, such as Tinkercad and Edge Impulse, were used to help students explore machine learning concepts, set up simple models, and understand how data can be collected, processed, and classified. Such school-based makerspace environments can support individual tinkering, collaborative problem-solving, and iterative design, and have been shown to support STEM learning in maker-based settings [43].

The project was organized around a maker project design cycle, in which students addressed the challenge of designing a machine-learning voice-recognition device. Across a sequence of twelve 40 min class lessons delivered over a three-month period, students learned how to collect and train voice-data samples and to create a simple machine-learning model capable of recognizing spoken keywords, thereby linking abstract AI concepts with real technological functionality.

• Phase I (Foundations—4 lessons).

All students received a common introduction to core concepts required for the project: Arduino programming, digital sensors, block-based coding, and introductory machine-learning tasks. Students engaged in guided simulations of data collection and experimented with simple model-training procedures in Edge Impulse. These activities were designed to establish baseline conceptual readiness and ensure that learners entered the project with comparable prior knowledge. Instruction combined demonstration, hands-on exploration, and structured reflection to help students articulate their understanding of how input signals such as voice are processed, interpreted, and used by computational systems.

• Phase II (Applied Implementation—8 lessons).

In collaborative teams, students then advanced to solving the authentic engineering challenge: designing and implementing a machine-learning voice-recognition device. At this stage, the class was organized into two learning groups following the study design:

• Maker Learning (ML): hands-on prototyping with Arduino.
• Traditional Learning (C): teacher-guided instruction supported by static visualizations and explanations.

To make the instructional contrasts explicit, the two pathways differed in how construction, feedback, collaboration, and guidance were operationalized, while the project theme, objectives, materials, and target AI concepts remained constant. In the Maker Learning (ML) pathway, construction involved physical prototyping (wiring, assembling, testing, and iteratively refining the device) and on-device deployment of the trained model. Feedback was primarily derived from artifact behavior (e.g., sensor readings, circuit functioning, and model outputs on-device), supplemented by facilitator-style teacher support. Collaboration centered on shared building, troubleshooting, and iterative refinement within groups, and guidance was provided through milestones and just-in-time scaffolding (e.g., brief prompts or targeted support when groups encountered difficulties).

In the Traditional Learning (C) pathway, construction was more limited, and the workflow was taught primarily through teacher-led explanation, guided completion of key steps, worked examples, and tightly scaffolded checkpoints. Feedback was predominantly teacher-provided (e.g., explanations, worked examples, and corrective feedback). Collaboration was more discussion-oriented, focusing on interpreting procedures and outputs rather than sustained co-building, and guidance was high-structured, with a tightly sequenced progression through the project steps.

• Alignment of instructional activities with AILQ domains.

To support domain-level interpretation, the instructional design targeted the AILQ domains through specific activity elements:

• Cognitive: guided explanation of ML basics; interpreting training outputs; discussing model limits and error sources.
• Behavioral: collecting voice samples; training models; testing outputs; deploying inference on-device (ML) or completing structured model steps (C).
• Affective: hands-on prototyping and iterative testing (ML); structured progress checkpoints (C); reflective prompts to support confidence and sustained engagement.
• Ethical: structured discussion prompts on reliability, safety, privacy, responsibility, inclusion, and accountability, linked to observed misclassifications and performance variability.

All groups engaged with the same design challenge: to understand how a machine-learning voice-recognition device operates, and design a working model. Students examined how voice signals change with the different sounds of the words, how data can be processed, and how a simple classification model can support automated recognition. Using Edge Impulse platform, students trained machine-learning models capable of distinguishing voice patterns of simple keywords “red” or “green.” Through this process, students were introduced to concepts related to digital signal processing, feature identification, and basic classification tasks. As they engaged in this workflow, they examined sampling choices, model accuracy, and sensor reliability and discussed how AI can support decision-making in applied contexts. Figure 2 illustrates a representative Edge Impulse training interface used during the activity.

Collaboration was embedded throughout the process. Students worked in teams to plan their prototype, divide responsibilities, test hypotheses, troubleshoot errors, and articulate design decisions. Each lesson concluded with short reflective prompts aligned to the maker design cycle (e.g., clarifying the problem, testing ideas, evaluating constraints, and proposing improvements). Figure 3 presents examples of student prototypes and voice signal analysis.

In the concluding lesson, students demonstrated and discussed their solutions, responding to reflective questions such as “What is AI?” and “How does a machine-learning voice-recognition device operate?” Ethical considerations related to AI, such as reliability, safety, privacy, responsibility, inclusion, and accountability, were also addressed and discussed. Across the 12-lesson sequence, this progression from foundational learning (Phase I) to applied implementation (Phase II) was intentionally designed to provide opportunities for students to engage with AI Literacy while enabling students to experience in the classroom how AI workflows can be integrated into engineering-oriented tasks.

3.4. Data Collection and Analysis

Data collection followed the mixed-methods design described earlier, consisting of quantitative self-report measures and complementary qualitative interviews. Learning was assessed using (a) AILQ domain scores collected at pre-test and post-test (self-perceived AI literacy) and (b) post-intervention semi-structured interviews (students’ perceptions).

The quantitative data were obtained through two administrations of the AI Literacy Questionnaire (AILQ) (pre-test and post-test), and were analyzed using a mixed ANOVA (Time × Group), with follow-up t-tests as appropriate. These analyses were used to explore pathway-associated differences in the observed pre–post score change, which are interpreted cautiously given the classroom implementation.

Qualitative data were collected after the intervention through exploratory semi-structured interviews and were analyzed thematically to contextualize the quantitative findings, rather than as confirmatory evidence for pathway-associated differences. Gender was not included as an analytic factor because there was one female student in each group, precluding meaningful gender-based comparisons.

3.4.1. AI Literacy Questionnaire (AILQ)

The study used the AI Literacy Questionnaire (AILQ), a validated instrument designed to measure students’ self-perceived competencies and attitudes toward AI across four interrelated domains: Affective (interest and motivation), Behavioral (engagement and intended use), Cognitive (knowledge and application), and Ethical (values and responsibility) [32]. The Greek-adapted version consisted of 32 items rated on a five-point Likert scale (1 = Strongly Disagree to 5 = Strongly Agree). In line with the original validation study, which reported excellent internal consistency (α = 0.93), the present sample also demonstrated strong reliability (α = 0.88 at post-test). The instrument was administered at two key points (pre-test, post-test), allowing the study to examine pre–post changes in AI literacy development across the two instructional pathways.

3.4.2. Semi-Structured Interviews

After completing the intervention, exploratory semi-structured interviews were conducted to deepen understanding of how students experienced the learning activities and to provide explanatory insight into the attitudinal patterns observed in the AILQ data. This qualitative component is consistent with recommendations in recent AI-literacy studies, which emphasize the importance of capturing affective and experiential dimensions that may not be fully visible through survey measures [7,27].

Each interview lasted approximately ten minutes and followed open-ended questions designed to elicit students’ reflections on their learning processes, engagement, and challenges. Sample prompts included “What did you enjoy most about the AI activity?”, “How do you think you learned through this process?”, and “What challenges did you face?” During each interview, the researcher documented key points, representative statements, and emerging ideas using structured field notes. These notes captured meaningful expressions related to students’ attitudes, learning processes, and perceptions of the educational environment. The qualitative data were analyzed using a six-step framework thematic analysis [54]. First, the field notes were read repeatedly to establish familiarity with the data. Second, initial codes were generated to identify meaningful points related to students’ experiences. Third, the codes were grouped into preliminary categories. Fourth, these categories were reviewed for coherence and refined into themes. Fifth, the themes were defined and named to capture patterns across the dataset. Finally, illustrative excerpts from the field notes were selected to support each theme.

The qualitative findings were used to contextualize and enrich the quantitative results, rather than to confirm them. The analysis aimed to identify recurring themes aligned with students’ engagement, perceived learning processes, and ethical reflections during the activity, which are reported as specific themes with illustrative excerpts in the results section.

4. Results

4.1. Quantitative Findings on Self-Reported AI Literacy

This section presents the quantitative findings derived from the AI Literacy Questionnaire (AILQ). Because the AILQ captures students’ self-reported AI literacy, it does not assess objective performance and results are interpreted as perceived competence and attitudes. Shapiro–Wilk tests indicated that AILQ scores did not significantly deviate from normality; therefore, parametric analyses were conducted as an appropriate exploratory approach for summarizing pre–post changes in this sample [55]. To examine both within-group pre–post change and to explore whether pre–post scores differed by instructional pathway, a two-way mixed ANOVA (Time × Group) was performed for each AILQ domain. Hereafter, the Time × Group term is referred to as the interaction. In this model, the Time main effect summarizes overall pre–post change across both groups, whereas the interaction examines whether the magnitude of pre–post score change differs between groups. When a statistically significant interaction effect was observed, Independent-Samples t-tests were used as follow-up comparisons to describe post-test differences between groups. Likewise, paired-samples t-tests were conducted to examine within-group pre–post changes for each group [56]. In line with the exploratory framing and to support interpretive clarity, within-group change is reported primarily using pre/post mean differences (Δ), with paired-samples t-tests indicating whether these within-group changes were statistically significant. AILQ scores were computed as the mean of the items within each domain, and all analyses were performed using SPSS v26. Effect sizes were reported to describe the magnitude of observed differences in this sample, following recommended guidelines for parametric analysis: partial eta squared (ηp²) for ANOVA effects and Cohen’s d for between-group t-tests [57,58]. Given the classroom implementation, these analyses are used to summarize and explore observed associations, not to support causal inference.

• Between-group differences in pre–post score change (interaction) across AILQ domains

Table 1. Between-group (interaction) comparisons for ML and C groups across AILQ domains.

Questionnaire Domain	Results
Overall	F(1,36) = 11.300, p = 0.002 **, ηp2 = 0.239
Affective	F(1,36) = 14.077, p = 0.001 **, ηp2 = 0.281
Behavioral	F(1,36) = 6.452, p = 0. 016 *, ηp2 = 0.152
Cognitive	F(1,36) = 12.470, p = 0.001 **, ηp2 = 0.257
Ethical	F(1,36) = 1.411, p = 0.243, ηp2 = 0.038

Statistically significant difference * p < 0.05; ** p < 0.01. Exact p-values reported.

summarizes the results of the two-way mixed ANOVA for each AILQ domain, examining how students’ self-reported AI literacy changed from pre-test to post-test in the Maker Learning (ML) and Traditional Learning (C) groups. This analysis supports an exploratory comparison of whether the observed pre–post score change differed by instructional group.

The two-way mixed ANOVA revealed a significant interaction for Overall AI literacy (F(1,36) = 11.300, p = 0.002, ηp² = 0.239), suggesting that the magnitude of pre–post score change differed between groups in this classroom sample, with the ML group showing a larger observed pre–post increase than the C group. This interaction is reported as a context-bound group difference within this sample, rather than as proof that one pathway caused greater improvement. A follow-up independent-samples t-test at post-test was used to further describe the post-test difference between groups (see

Table 2. Post-test comparisons of AI Literacy domains for ML and C groups at post-test.

Questionnaire Domain	Results
Overall	t(36) = −4.830, p < 0.001 ***, d = 1.57
Affective	t(29.725) = −4.766, p < 0.001 ***, d = 1.54 (Welch)
Behavioral	t(36) = −4.580, p < 0.001 ***, d = 1.49
Cognitive	t(36) = −5.010, p < 0.001 ***, d = 1.63
Ethical	t(36) = −0.489, p = 0.628, d = 0.16

Statistically significant difference *** p < 0.001. Exact p-values reported.

), while interpretations remain context-bound to this classroom sample.

For individual domains, significant interactions were observed in the Affective domain (F(1,36) = 14.077, p = 0.001, ηp² = 0.281), the Behavioral domain (F(1,36) = 6.452, p = 0.016, ηp² = 0.152) and the Cognitive domain (F(1,36) = 12.470, p = 0.001, ηp² = 0.257), indicating statistically significant between-group differences in pre–post score change in these domains. In practical terms, the largest interaction effects were observed for affective, behavioral and cognitive AI literacy domains.

By contrast, the Ethical domain (F(1,36) = 1.411, p = 0.243, ηp² = 0.038) showed no statistically significant interaction, suggesting that evidence for between-group differentiation in pre–post score change was not clear for this domain, within this intervention window. Accordingly, this domain is interpreted descriptively in terms of within-group pre–post change and post-test differences, without implying between-group differences in pre–post score change.

• Post-test differences (between groups)

Table 2 presents the independent-samples t-tests conducted at post-test to describe post-test differences between groups. For AI literacy domains with significant interactions (Overall, Affective, Behavioral, Cognitive), these post-test comparisons are treated as follow-up probes of the interaction. For AI literacy domains without a significant interaction (Ethical), the post-test differences are reported descriptively. Levene’s tests indicated homogeneity of variances for Overall, Behavioral, Cognitive, and Ethical domain post-test comparisons. The Affective domain post-test comparison violated homogeneity; therefore, Welch’s t-test was used.

Furthermore, post-test mean differences (C − ML) and effect sizes are reported with 95% confidence intervals to support interpretation across domains.

Overall, AI literacy differed between groups at post-test (t(36) = −4.830, p < 0.001, d = 1.57, 95% CI [0.83, 2.29]). The post-test mean difference (C − ML) was −0.383, 95% CI [−0.544, −0.222], indicating higher post-test scores in the ML group. This represents a large observed post-test separation in this classroom sample and is interpreted as a descriptive magnitude consistent with the interaction result.

Consistent with the interaction findings, the most pronounced post-test difference emerged in the AI literacy Affective domain (t(29.725) = −4.766, p < 0.001, d = 1.54, 95% CI [0.81, 2.27]) (Welch correction applied due to unequal variances). The post-test mean difference (C − ML) was −0.432, 95% CI [−0.617, −0.247], indicating that students in the ML group achieved higher scores than those in the C group. This large post-test separation is reported as an observed magnitude in this sample and should not be interpreted as a population-level “effect.”

For the AI literacy Behavioral domain, a statistically significant post-test difference was also observed (t(36) = −4.580, p < 0.001, d = 1.49, 95% CI [0.76, 2.20]). The post-test mean difference (C − ML) was −0.461, 95% CI [−0.664, −0.257], corresponding to the interaction result for this domain within the present sample.

The AI literacy Cognitive domain also showed a strong post-test difference (t(36) = −5.010, p < 0.001, d = 1.63, 95% CI [0.88, 2.36]). The post-test mean difference (C − ML) was −0.605, 95% CI [−0.850, −0.360], in line with the interaction and indicating a sizeable between-group difference in pre–post score change for self-reported conceptual-related AI literacy by the end of the intervention, within this classroom sample.

For the AI literacy Ethical domain, no post-test difference was observed (t(36) = −0.489, p = 0.628, d = 0.16, 95% CI [−0.48, 0.79]). The post-test mean difference (C − ML) was −0.079, 95% CI [−0.406, 0.248], consistent with the non-significant interaction and suggesting limited group separation in the ethical domain of AI literacy within this intervention period.

Taken together, the interaction effects indicate between-group differences in pre–post score change for AI literacy (Overall, Affective, Behavioral and Cognitive), with post-test differences aligning with these findings.

• Within-group pre–post changes

Within-group analyses address a different question, describing whether each group showed pre–post change within itself, without implying that one group improved more than the other. To examine within-group change, paired-samples t-tests were conducted for the Maker Learning (ML) and Traditional Learning (C) groups across all AILQ domains. Paired-samples t-tests indicated statistically significant pre–post gains across all AI literacy domains. Across domains, these within-group results indicate a consistent pre–post upward trend in self-reported AI literacy for both Maker Learning (ML) and Traditional Learning (C) groups. Notably, the Maker Learning (ML) group showed larger observed mean increases (Δ) across domains: Overall (Δ = +0.985 for ML vs. +0.587 for C), Affective: (Δ = +1.063 for ML vs. +0.632 for C), Behavioral (Δ = +0.763 for ML vs. +0.342 for C), Cognitive (Δ = +1.228 for ML vs. +0.728 for C), Ethical (Δ = +0.928 for ML vs. +0.671 for C). These Δ values are reported as mean-score changes (post minus pre) and are presented as descriptive estimates of within-group change. Accordingly, inferences about between-group differentiation in pre–post score change are anchored in the interaction. Interpretations about between-group differentiation are limited to domains where the interaction was statistically significant (Overall, Affective, Behavioral, Cognitive). For Ethical, which did not show a significant interaction, the within-group increases are interpreted as overall pre–post change within this intervention window rather than as differential change between groups.

Taken together, these within-group findings indicate that both groups were associated with pre–post growth in self-perceived AI literacy in this vocational classroom context, while the between-group analyses suggest differences in pre–post score change for Overall, Affective, Behavioral and Cognitive AI literacy within this classroom sample. In line with the study’s context and design, these results are interpreted as context-bound exploratory tendencies that can inform future, larger-scale research in vocational STEM education.

4.2. Qualitative Findings from Semi-Structured Interviews

To complement the AILQ findings, exploratory semi-structured interviews were conducted with students from both learning groups. The interviews provided insight into how students experienced the learning activities and how they described their interest, motivation, engagement, understanding, and emerging ethical awareness regarding AI. The thematic analysis revealed three main themes that relate to the multidimensional structure of AI literacy—affective, behavioral, cognitive, and ethical—and offer interpretive context for the quantitative results across the Maker Learning (ML) and Traditional Learning (C) groups. Rather than serving as confirmatory evidence, these themes are used to contextualize the quantitative results observed in this classroom sample, including domains with significant interactions (Overall, Affective, Behavioral, Cognitive) and the Ethical domain, where the interaction was not significant within this intervention window.

• Theme 1: Affective Engagement and Perceived Relevance with AI

Across both groups, students emphasized that learning about AI felt important and timely. However, ML students frequently described the activity as “learning by doing” and referred to feelings of excitement and interest, emphasizing that the lesson felt “more exciting” and “less like ordinary theory” when the voice-recognition model responded to their own voice. Several ML students reported statements such as “it was fun to see that the model recognized my voice” or “I wanted to try more words to see what it could learn.” These descriptions point to strong affective engagement in students’ accounts, which is consistent with the quantitative pattern in the Affective domain.

Students in the Traditional Learning (C) group also acknowledged the importance of learning about AI, often stating that “AI is everywhere now; it is good to learn about it” but tended to frame their learning experience as “useful but hard to imagine.” Although they appreciated the relevance of the topic, the absence of hands-on interaction was frequently described as limiting the depth of their emotional engagement, as described in their interviews. These accounts provide interpretive context for the AILQ results by illustrating how the maker-based environment may have translated AI from an abstract topic into a personally meaningful and emotionally engaging experience in this classroom context.

• Theme 2: Behavioral Engagement and Perceived Practical Competence with AI

This theme provided insight into how students described their willingness and perceived competence when interacting with AI tools. Many ML students stated, “at first it seemed too difficult, but now I understand how a machine actually learns,” capturing the shift from initial uncertainty to practical confidence. One ML student noted that “AI is not magic”, while several other ML students described that “I can teach the machine, and train it” emphasizing that the step-by-step process of collecting data, training, and testing the model made AI workflows more transparent. Students also reported that debugging misclassifications, adjusting samples, and collaborating with peers strengthened their sense of practical competence. These experiences provide interpretive context for students’ reported engagement with AI tools and perceived competence and illustrate how students linked hands-on work to their perceived readiness to use AI tools in future learning or work contexts. Consistent with quantitative findings of the Behavioral domain (significant interaction), these interview accounts are interpreted as providing interpretive context for the stronger behavioral engagement described by the ML group within this sample.

Students in the Traditional Learning (C) group also reported improved understanding, especially regarding basic AI concepts and terminology. Some expressed that “the teacher’s examples helped me understand the logic of AI more clearly,” but also added that they were “not sure” how to implement such systems themselves. This contrast in students’ narratives suggests a difference in how learning was experienced (hands-on agency vs. explanation-based understanding) in this sample, and it is consistent with the between-group differentiation indicated by quantitative results in behavioral engagement by the end of the intervention.

• Theme 3: Emerging Ethical Awareness through Direct Interaction with AI Systems

Students’ interviews revealed reflections on fairness, reliability, and responsible use of AI. Ethical issues were not the primary focus of any learning group. Yet ML students occasionally questioned why the model “sometimes recognizes one voice and not another,” or why “it needs many examples to be fair.” Some ML students questioned why the system “got some voices wrong” or “treated similar sounds differently,” which led them to speculate about data quality, bias, and the reliability of AI. These comments illustrate instances of early awareness that AI performance depends on data quality and that misclassifications may have consequences in real-world applications. These reflections suggest that conceptual understanding and ethical awareness were intertwined, emerging as students observed their model’s limitations and errors.

By contrast, students in the Traditional Learning (C) group rarely raised ethical concerns spontaneously. Their reflections focused more on understanding what AI is and how it works at a general level, with limited reference to bias, privacy, or fairness. The interviews, therefore, suggest that direct interaction with AI systems—observing both accurate outputs and errors—may provide a more concrete point for ethical reflection than abstract explanation alone in the classroom context. At the same time, this theme is presented as descriptive of the interview content, while the quantitative results did not indicate clear between-group differences in pre–post score change in the Ethical domain within this intervention window.

• Synthesis of Qualitative Findings

Taken together, the qualitative findings provide interpretive depth and context to the quantitative results. Students in the maker learning group reported higher engagement and a clearer sense of relevance, corresponding to the domains where significant interactions were observed (Overall, Affective, Behavioral, Cognitive). Their descriptions of increased confidence and “being able to teach the machine” are consistent with the identified affective, behavioral and cognitive domains where the interaction was statistically significant in the quantitative analysis. At the same time, their spontaneous reflections on misclassifications, fairness, and reliability offer examples of how ethical considerations can emerge during interaction with model errors and limitations, while the quantitative analyses did not show clear between-group separation in the Ethical domain.

Students in the Traditional Learning (C) group also recognized the importance of AI and reported self-perceived competence gains; however, their experiences remained more observational and less embodied, corresponding to smaller affective and behavioral shifts and limited ethical development. Overall, the qualitative data support an interpretive reading that maker learning may be experienced as a more engaging and multidimensional learning experience for vocational STEM students in this specific classroom context, while avoiding causal interpretations of pathway differences and treating ethical-domain contrasts as tentative given the absence of clear between-group differentiation in Ethical pre–post change within this intervention window.

5. Discussion

The present findings provide exploratory, classroom-based evidence that a maker-based learning intervention may support vocational students’ self-reported contextual AI literacy. Given the classroom-based non-randomized implementation and reliance on AILQ, interpretations are limited to context-bound tendencies in this sample rather than population-level effects. Although the mixed ANOVA framework is often used in experimental research, here the Time × Group interaction is treated as an exploratory summary of observed group-by-time score patterns, not as evidence of causal impact.

A key contribution is the domain-specific profile of change. Between-group differentiation in pre–post score change was observed for Overall AI literacy and for the Affective, Behavioral and Cognitive domains, whereas the Ethical domain did not show clear evidence of between-group differentiation in pre–post within this intervention window. This domain-sensitive interpretation aligns with validated multidimensional perspectives of AI literacy—affective, behavioral, cognitive, and ethical—and with intervention studies in secondary education where motivation, confidence, collaboration, and engagement often show the strongest gains [7,8]. At the same time, the present results underscore that “overall improvement” can mask uneven development across domains, suggesting that domain-level reporting can function as a practical diagnostic lens for instructional design. To interpret the observed domain-level changes, we draw on Self-Determination Theory, Kolb’s experiential learning cycle, and Cognitive Load Theory as complementary interpretive lenses.

From an intervention-evaluation perspective, domain-level results can function as a diagnostic lens by indicating which capacities are being cultivated (e.g., affective engagement, behavioral readiness and conceptual understanding) and which may require redesign (e.g., ethical reasoning). In this sense, a positive overall AI literacy result does not guarantee balanced development across domains, reinforcing the need for holistic designs that intentionally target all components of AI literacy. These findings are consistent with the view that AI literacy development is shaped by learners’ disciplinary context and their experiential interaction with AI, as reflected in recent frameworks that define AI literacy and competence through authentic educational and professional practices [4,5,6]. From a motivational standpoint, this domain profile is compatible with Self-Determination Theory (SDT), which proposes that engagement and learning are strengthened when learning environments support autonomy, competence, and relatedness [59]. In vocational learning settings, motivation and engagement are especially consequential for learning trajectories and transition planning [18], and SDT-based evidence further highlights the role of basic psychological needs in shaping engagement [60,61]. Within this framing, the clearer affective and behavioral differentiation in the Maker Learning group is plausibly aligned with learners experiencing greater agency in the workflow, competence through iterative feedback, and relatedness through collaborative troubleshooting [59].

By linking engagement with practical experimentation, the present study is consistent with broader evidence that experiential and socio-constructivist pedagogies foster sustained interest in STEM learning contexts [25,62]. Qualitative findings add interpretive context by showing that ML students repeatedly framed learning as “learning by doing,” describing the activity as personally meaningful and motivating—an account that coheres with the stronger quantitative differentiation observed in the Behavioral and Affective domains. From a theoretical perspective, the improvements observed across groups can be interpreted through constructionist learning theory. Constructionism proposes that knowledge is adopted more effectively when learners actively build, test, and refine personally meaningful objects [23]. In this study, by building and training a functional voice-recognition model, students engaged with abstract algorithmic concepts through tangible design decisions and observable system behavior.

In parallel, the cognitive-related gains are also compatible with Kolb’s experiential learning cycle, where learning is strengthened through concrete experience, reflective observation, conceptualization, and active experimentation [63]. Recent studies further support the use of Kolb’s cycle as a mechanism for deeper learning when activities provide observable feedback and structured opportunities for iteration [64]. In the present intervention, model outputs and misclassifications provided immediate feedback that students could compare with expectations, supporting repeated cycles of adjustment and sense-making around data, training, inference, and error [63,64]. The observed cognitive-related findings are also consistent with prior research showing that maker environments foster social learning and co-construction of knowledge [24]. Students’ reports of debugging, testing, and collaborating suggest that peer interaction and iterative refinement supported meaning-making around data, models, and system outputs. Consistent with the domains where between-group differentiation was clearer, the improvement in affective, behavioral and cognitive domains reflects the well-documented progression by which motivation and interest can act as gateways to higher-order reasoning, as noted in Arduino-based and maker-oriented STEM courses [45]. Taken together, these points support the perspective that Maker Education serves as a learning pathway promoting motivation and agency and may enable a more nuanced understanding [26].

At the same time, the domain profile analysis cautions against overgeneralizing the intervention’s impact to all domains of AI literacy. Students’ spontaneous reflections on misclassification, fairness, and reliability suggest that ethical awareness may begin to emerge through direct interaction with imperfect AI systems. However, these comments are best interpreted as early, experience-triggered noticing rather than evidence of differential ethical development, because the ethical domain did not show clear quantitative between-group differentiation within this intervention window, so ethical claims remain descriptive and context-bound. One plausible explanation is offered by Cognitive Load Theory (CLT): engaging with new tools, procedures, and concepts may impose substantial intrinsic load, and ethical reasoning can add additional abstraction that may not become salient without explicit scaffolding [65,66]. Thus, even when ethical-domain scores do not shift measurably, the learning setting can still surface ethically meaningful moments that warrant explicit instructional attention. This is particularly relevant for voice-recognition activities, where ethical considerations extend beyond “fairness” in an abstract sense to include data protection, privacy, and the potential identifiability of voice samples. Accordingly, strengthening ethical AI literacy may require learning designs where socio-technical dilemmas are more explicit and consequential, such as authentic engineering problems, community-based challenges, or problem scenarios where bias, privacy, safety, and accountability are directly encountered and discussed. In addition, when projects use a web-based ML environment, it is useful to explicitly address ethics at the “digital” workflow level. For voice-based datasets, a practical classroom step is to treat voice samples as potentially personal data and embed clear norms for responsible handling. These concrete safeguards may make ethical reasoning more observable and assessable. Measurable differentiation in ethical AI literacy may also require longer exposure and structured reflection activities (e.g., targeted ethics prompts, case-based reflection, or scenarios where consequences of error/bias are salient and assessable). From an SDT and CLT perspective, these additions may also support students’ perceived competence and reduce extraneous cognitive load by making ethical dimensions explicit, discussable, and assessable within the workflow [59,66].

Accordingly, the present results support a holistic approach to AI literacy design. Maker Education can function as an enabling platform, but domain-complete AI literacy may require intentional pedagogical layering that brings real-world constraints and socio-technical dilemmas to the foreground. One practical direction is to embed maker activities within Project-Based, Problem-Based, or Challenge-Based Learning frames, where students work toward authentic goals (e.g., safety, reliability, fairness, privacy) and are prompted to justify design decisions, evaluate trade-offs, and reflect on consequences. In this way, making remains the core activity, while the surrounding pedagogical design makes ethical reasoning more visible, discussable, and assessable within the workflow.

In summary, contextual AI literacy appears to develop through the interplay of affective engagement, hands-on experimentation, collaborative meaning-making, and opportunities for reflection. In this study, the Maker Learning environment supported clearer development in the affective, behavioral and cognitive AI literacy domains by connecting tangible engineering practice with observable AI model behavior and iterative feedback. At the same time, the domain profile indicates where additional design emphasis may be required, particularly for ethical AI literacy, which may not shift measurably without explicit socio-technical dilemmas, structured reflection, and assessment-aligned prompts.

Overall, these outcomes are consistent with core constructionist principles, in which learning emerges through the creation, testing, and refinement of meaningful artifacts. Maker projects that incorporate AI sensing and pattern recognition extend constructionism into a socio-technical learning context, allowing students to experience algorithmic processes as real, observable phenomena rather than abstract descriptions. Importantly, the present study underscores that holistic contextual AI literacy is an instructional design target rather than an automatic outcome, since interventions like the one implemented here may strengthen some domains (affective, behavioral and cognitive) while leaving others (ethical) comparatively less developed unless they are deliberately targeted through authentic constraints, structured socio-technical challenges, and appropriate assessment practices. This framing supports treating contextual AI literacy as an integrated cognitive, behavioral, affective, and ethical competence emerging within interdisciplinary maker environments, while maintaining interpretive restraint consistent with the interaction findings in this sample.

6. Conclusions

This study provides exploratory, classroom-based evidence that a maker learning intervention using Arduino hardware and the Edge Impulse platform was associated with vocational secondary school students’ self-reported contextual AI literacy. By developing and training their own voice-recognition models, students engaged with introductory AI concepts through concrete engineering practices, making AI workflows more observable, testable, and personally relevant. Within this non-randomized classroom sample and using a self-report instrument, observed group differences should be interpreted as context-bound associations rather than causal effects. Within this classroom sample, the Maker Learning (ML) group was associated with larger observed pre–post gains in self-reported interest and motivation (affective), willingness to engage with AI tools (behavioral), and conceptual/knowledge-related understanding (cognitive). These findings support the feasibility of implementing low-cost maker activities within vocational STEM curricula in similar settings and align with multidimensional perspectives that conceptualize AI literacy as comprising cognitive, behavioral, affective, and ethical domains [7,8,28].

At the same time, this study illustrates why a positive “overall AI literacy” result should not be interpreted as uniform development across domains. In the present intervention window, the ethical domain did not show clear differentiation, so conclusions about ethical AI literacy remain tentative and descriptive. This pattern indicates a design implication for future implementations and research, suggesting that ethical AI literacy may be less likely to shift without explicit socio-technical dilemmas and structured reflection embedded in the learning process. Accordingly, Maker Learning can function as a strong enabling platform for engaging students with AI workflows, but this study does not justify general recommendations about what “works best” across contexts; rather, it indicates that more balanced development across affective, behavioral, cognitive, and ethical domains may require additional pedagogical scaffolding (e.g., Project-Based, Problem-Based, or Challenge-Based Learning frames) in similar classroom settings.

Within the limits of this classroom context, the present study offers classroom-grounded design insights for vocational STEM education by highlighting a domain profile in which maker-based learning appears to support growth most clearly in affective, behavioral, and cognitive AI literacy, while ethical AI literacy may require additional instructional design emphasis. Because the AILQ captures perceived competence and attitudes rather than objective performance, these conclusions should be read as describing changes in students’ self-perceptions and orientations toward AI. These conclusions align with broader educational priorities for equitable and ethical AI integration in schools and with calls for sustained collaboration between teachers and academia to navigate an evolving technological landscape [31,67].

7. Limitations and Future Research

There are limitations that should be acknowledged. The qualitative component relied on students’ self-reported data collected through exploratory semi-structured interviews, which may be subject to response bias. The intervention duration was relatively short (three months, corresponding to twelve class lessons) compared with longer implementations reported in the literature. As a result, the findings primarily reflect short-term changes, and no claims can be made regarding the long-term stability, transfer, or retention of these changes. This is particularly relevant for domains such as ethical AI literacy, where measurable differentiation between instructional approaches may require longer exposure and more structured socio-technical dilemmas than a relatively short intervention can provide. Accordingly, qualitative findings are interpreted as complementary interpretive context rather than as confirmatory evidence for between-group differences or causal impacts.

In addition, the study was conducted within an authentic vocational school setting, where the two instructional groups were implemented as part of normal teaching practice. The organization of students into these groups was shaped by instructional modality and learning environment, rather than by full random assignment at the individual level. Although levels were comparable on the study measures, non-random grouping means that unmeasured differences and classroom-related influences (e.g., peer dynamics, teacher guidance, and contextual constraints) cannot be fully excluded. Furthermore, the statistical analyses were interpreted in an exploratory manner. Given the relatively small sample size, the quantitative results are reported as tendencies within these classroom conditions rather than as definitive population-level effects or stable estimates of effect magnitude. The sample was also highly gender-imbalanced, and gender was not included as an analytic factor because there was one female student in each instructional group, precluding meaningful gender-based comparisons. In addition, because the AILQ is a self-report instrument, changes reflect students’ self-perceived competence and attitudes rather than directly observed competence or performance.

Future research involving larger and more diverse samples across vocational and general education contexts would allow for stronger inferences and broader applicability. This work could be extended through longer-term interventions to examine whether the shifts identified in the present study are sustained over time and whether growth in cognitive and ethical domains becomes more pronounced with extended engagement. Future work should aim for more gender-diverse vocational STEM samples to examine whether maker-based AI learning pathways operate similarly across genders and whether specific design features support more inclusive participation. The ethical dimension could also be expanded through structured activities and community-based challenges that foreground socio-technical dilemmas (e.g., bias, privacy, safety, accountability) to further support the comprehensive development of contextual AI literacy among vocational learners. Finally, future work should examine how teacher professional development can sustain maker-based AI pedagogies in vocational contexts, including support for lesson design, classroom implementation, and assessment of student artifacts and AI workflows and, where feasible, incorporate outcomes beyond self-report to triangulate change (e.g., evaluation of student artifacts, AI workflows, or performance-based tasks) and to better align perceived AI literacy with demonstrated competencies.