Conference Program

The rapid diffusion of artificial intelligence in educational contexts is reshaping the epistemic conditions of teaching and learning, particularly in mathematics education, where generative systems provide immediate solutions and procedural explanations. While such tools offer new forms of cognitive support, they also raise critical concerns regarding student agency, pedagogical judgment, and the preservation of democratic learning spaces within increasingly datafied environments (Williamson, 2017). The challenge, therefore, extends beyond technological adoption to the cultivation of critical AI literacy, understood as the capacity to question, interpret, and responsibly engage with algorithmically generated knowledge (Long & Magerko, 2020).

This paper presents a classroom-based study conducted in a classical high school in Bagheria (Sicily, Italy), where AI-powered mathematical applications were integrated through structured pedagogical mediation. Drawing on Vygotsky’s sociocultural framework (1978), AI is conceptualized as a mediating artifact operating within the learner’s Zone of Proximal Development rather than as an autonomous authority. Metacognition (Flavell, 1979) and dialogism (Bakhtin, 1981) provide the theoretical lens through which student–AI interaction is analyzed.

The study adopts a mixed-method case-study design. A structured questionnaire administered to the entire course provided descriptive data on students’ patterns of AI use, while qualitative thematic analysis of student narratives examined processes of cognitive support, cognitive delegation, metacognitive engagement, and teacher mediation. Questionnaire findings indicate that most students initially approached AI as a procedural support tool, whereas a smaller group demonstrated explicit metacognitive monitoring and systematic comparison between AI outputs and classroom reasoning. Qualitative analysis further revealed differentiated trajectories of instrumental use and reflective appropriation.

Rather than interpreting instrumental engagement as failure, the study frames such variability as part of a process of instrumental genesis (Rabardel, 1995), through which learners progressively appropriate new cognitive tools. Sustained teacher mediation functioned as an epistemic anchor, ensuring that AI outputs remained embedded within disciplinary mathematical reasoning rather than substituting it.

The findings suggest that AI integration, when embedded within dialogical pedagogy and supported by metacognitive scaffolding, can strengthen student agency and sustain democratic learning environments. In this perspective, critical dialogue with AI emerges as foundational to preserving mathematics education as a space of reflective judgment, disciplinary rigor, and responsible technological engagement.

The rapid integration of generative AI into educational contexts raises a fundamental pedagogical question: how can schools prevent students from delegating thinking to machines and instead cultivate the critical agency that democratic education requires? This contribution presents an empirical case study developed in a secondary vocational school (Servizi Socio-Sanitari, IISS Don Milani Pertini Morante, Grottaglie, Italy), exploring how a deliberately designed pedagogical framework — the AI-Resistant Debate — can transform AI interaction into a site of metacognitive practice.

The framework rests on a core assumption aligned with Vygotsky’s Zone of Proximal Development: AI does not replace human mediation but can function as a challenging dialogical partner (Bakhtin) whose outputs demand critical scrutiny. In the AI-Resistant Debate, student teams are invited to use generative AI tools (ChatGPT, Gemini) during the preparation phase, but with a crucial constraint: they must keep an AI-Log — a reflective diary documenting every AI suggestion accepted, every one rejected, and the reasoning behind each decision. This protocol becomes a powerful metacognitive mirror (Flavell, 1979): by externalising their decision-making process, students learn to distinguish their own reasoning from automated output, to identify hallucinations, and to exercise what the framework terms “critical prompting” — the capacity to interrogate AI rather than accept it as an oracular authority.

Two complementary activities enrich this core design. In the augmented storytelling modules, students used NotebookLM and Gemini to reinterpret canonical literary texts (Dante, D’Annunzio, Verga) and historical sources (Magellan’s circumnavigation) in multimodal formats: podcasts, infographics, video scripts, and vision boards. Here, the AI-Log evolved into a production diary tracking creative agency — which ideas originated with the student, which were AI-generated, and how human and machine authorship were negotiated. A complementary Prompt-Book for Authentic Assessment provides teachers with structured prompts for metacognitive evaluation, including a rubric specifically designed to assess the quality of students’ AI dialogue — distinguishing passive consumption from critical co-creation.

The debriefing session concluding each activity foregrounds the democratic dimension of the framework. Guided by questions such as “Who was controlling whom?” and “Where did your thinking end and the machine’s begin?”, students reflect collectively on epistemic autonomy, the risks of automation bias, and the civic responsibilities entailed in using AI-generated content in public discourse.

Qualitative data drawn from AI-Logs, self-assessment checklists, and post-activity reflections suggest that the framework effectively nurtures higher-order thinking skills — analysis, evaluation, and creation in the revised Bloom’s taxonomy (Krathwohl, 2002) — while developing a practised scepticism toward AI outputs. Students who engaged most rigorously with the AI-Log demonstrated stronger argumentative coherence and greater ownership of their final products.

This case study contributes to the panel’s inquiry by demonstrating that the critical potential of AI in schools is not inherent in the technology itself, but in the pedagogical scaffolding that surrounds its use. The AI-Resistant Debate and AI-Log offer a replicable framework for ensuring that AI amplifies, rather than erodes, the conditions for democratic learning.

The widespread availability of generative AI poses a structural challenge to educational assessment and curriculum design: when a language model can produce summaries, definitions, lists of causes, and descriptive analyses in seconds, the lower rungs of Bloom’s taxonomy risk becoming a “cognitive commodity”. If schools continue to certify the reproduction of knowledge that AI generates effortlessly, they risk certifying the machine rather than the student. This paper argues that this crisis demands not a rejection of Bloom’s framework, but its reframing for an era of human–AI co-cognition.

Drawing on two years of classroom-based curriculum development at IISS Don Milani Pertini Morante (Grottaglie, Italy) — a secondary vocational school in the Servizi Socio-Sanitari track — this contribution presents a practical reframing of Bloom’s revised taxonomy adapted to the conditions of generative AI. The proposed framework reorganises the taxonomy’s six levels into three pedagogical zones. The AI-Commodity Zone (Curate & Question; Prompt & Explain; Adapt & Apply) acknowledges that AI handles baseline cognitive tasks with statistical precision, and redefines these levels around source verification, critical prompting, and situated judgement rather than mere recall and comprehension. The Human Superiority Zone (Compare & Validate; Challenge & Reflect; Co-Create) foregrounds the specifically human capacities of identifying algorithmic bias, exercising ethical judgement, and driving innovation through strategic human–AI collaboration. At the apex stands the Transform level, inspired by Mezirow’s transformative learning theory: the student becomes an agent of change who deploys knowledge to generate real social impact, a capacity that no algorithm can replicate.

The paper presents a comparative table of “fragile” vs. “AI-resistant” learning objectives, showing how a simple shift in the action verb transforms a delegable task into one that demands authentic human cognition. For example, “describe the figure of Augustus” becomes “refute an AI-generated analysis of Augustus as ‘Saviour’, using primary sources” — a task requiring Compare & Validate. Similarly, “summarise the chapter” becomes “argue the validity of thesis X against AI output Y”, activating Challenge & Reflect. The framework also introduces Enhancement as a metacognitive bridge level: the strategic capacity to integrate AI outputs with traditional academic sources to produce depth that neither human nor machine achieves alone.

The framework has been operationalised through a Prompt-Book for Authentic Assessment — a structured set of rubrics and teacher prompts designed to evaluate not the product of AI interaction, but its quality: did the student interrogate, improve, and contextualise AI output, or merely reproduce it? This assessment instrument embeds the democratic dimension of the framework: by requiring students to take ownership of their cognitive process — to decide what is reliable, relevant, and ethical — it cultivates the epistemic autonomy and civic responsibility that democratic education demands.

This contribution offers researchers and practitioners a replicable theoretical framework and a set of practical design tools for navigating the tension between AI efficiency and democratic agency. The central argument is that Bloom’s taxonomy, properly reframed, does not become obsolete in the AI age: it becomes more necessary than ever, as the compass that distinguishes authentic human learning from the automation of thought.

Integrating Generative Artificial Intelligence (GenAI) into education requires thoughtful decisions that prioritize human agency. One major concern is cognitive delegation: when students (and sometimes teachers) start relying too much on the system for thinking, their learning can actually weaken instead of getting stronger. That’s why we need effective teaching strategies—ones that view AI as a conversational partner that encourages reflection, rather than just a quick fix that takes the place of human judgment. The PAIDEIA project, backed by the Regional Programme of the Marche Region (https://paideia.study/), is built on a theoretical framework that merges learning sciences with computer engineering. Its goal is to promote “human training” that doesn’t just adapt to technology but actively influences how we use it. Addressing the concerns raised in the panel “Critical Dialogue with AI in Schools” (M.03), the project investigates how teachers can intentionally design lessons to make AI a facilitator of critical thinking, rather than just a tool that automates decision-making. The key research question is: how can the collaborative design of Learning Units (LUs) using the “Teacher in the Loop” (TiL-AI) model enhance metacognitive engagement and empower students, pushing back against algorithmic passivity? In line with sociocultural perspectives, the project places AI within the Zone of Proximal Development (Vygotsky, 1978), viewing it as adaptive support that can help manage cognitive load (Sweller, 1988) and promote educational success through a transdisciplinary approach. The heart of this project is all about empowering teachers by training 460 of them to collaboratively guide the learning experiences facilitated by the PaolaGen.AI platform (https://paolagen.ai/). A crucial part of this approach is the Teacher in the Loop AI (TiL) strategy (Balaji et al., 2025), which puts educators front and center in the algorithmic process. This ensures that AI enhances teaching methods instead of replacing professional judgment. With PAIDEIA's Learning Units (LUs), we create tools that not only engage students but also boost their intrinsic motivation and sense of competence—key elements of Self-Determination Theory (Ryan & Deci, 2000). In this way, AI becomes a supportive tutor that helps solidify skills-based teaching. For the 5,800 students involved, the project fosters the development of strong metacognitive skills, which are vital for resisting algorithmic conditioning and avoiding passive content consumption. Following Nelson and Narens' (1990) model of monitoring and regulation, the activities we’ve designed encourage students to critically assess the quality of GenAI outputs, paving the way for genuine Critical AI Literacy (Flavell, 1979). In this self-regulated learning environment (Zimmerman, 2002), AI serves as a “cognitive mirror” (Efklides, 2011), prompting students to plan and adjust their study strategies, thereby enhancing their executive functions (Brown, 1987) and awareness of their cognitive journey. Ultimately, the PAIDEIA project shows that the democratic integration of AI can only thrive through strong teacher collaboration inspired by the TiL model, which emphasizes critical thinking and creativity, ensuring a fair and inclusive learning experience.

This paper documents an intensive Formazione Scuola Lavoro (FSL) module carried out in January 2026 with a fourth-year class (16 students) at a liceo linguistico in northern Italy. The module was jointly designed and delivered by a secondary school teacher of English language and literature and a pedagogy graduate researcher working as FSL tutor for an educational technology company. Both contributors brought distinct but genuinely pedagogical frameworks to the design: the former rooted in narrative and literary-disciplinary practice, the latter in competency-based education and FSL assessment: producing a collaborative authorship that shaped both the project's architecture and the interpretation of its outcomes. The project work took podcast production as its central device: a format requiring students to move through a complete cycle of competencies (critical analysis of AI systems, collaborative negotiation of topic and format, editorial co-design with an AI tutor, and performative audio production). This cycle was deliberately structured as a progression, with each phase requiring a greater degree of initiative and autonomous decision-making than the previous one. An encounter with professional podcasters mid-process introduced an external standard of communicative quality, shifting students' self-perception from task-completers toward communicative authors accountable to a real audience.

Three Black Mirror episodes were used as the analytical entry point, mediated through DOK-structured scaffolding across four cognitive levels. The AI tutor askLea was introduced only after a phase of peer negotiation, designed to function as a critical interlocutor. Conversation logs document patterns of interaction ranging from passive, delegating acceptance of AI output to increasingly directive and evaluative prompting. The conditions that appeared to shift students from one posture to the other constitute the central analytical focus of this contribution.

The paper's primary research question is not whether the module "worked", but under what conditions critical agency with AI tools emerged or failed to emerge. To address this question, the paper will draw on triangulation across three data sources (askLea conversation logs, direct observation records from the sessions, and a final Likert-scale questionnaire on perceived competency development, with particular attention to the tensions between self-reported autonomy and observed interaction behaviour. Where these sources diverge, the divergence itself will be treated as pedagogically significant data, pointing to the gap between perceived and enacted competency.

The experience is presented with deliberate transparency about its unevenness: fragmented group dynamics, intermittent engagement, and products of variable quality. In an FSL framework, this unevenness maps precisely where transversal competencies require further pedagogical attention, and it is from that mapping that the contribution's practical implications are drawn.

When an algorithmic system classifies a student as "low risk" without that assessment being open to verification or discussion, it does not merely make a technical error: it forecloses the very possibility of educational dialogue. The teacher cannot open a conversation about what they cannot see, the student cannot regulate their learning on the basis of criteria they are unaware of, and the school cannot exercise its function if its decisions are delegated to an opaque procedure. This contribution argues that the problem is not the use of AI in education, but the conditions under which it is used: conditions which, when absent, make AI incompatible with any serious conception of critical learning.

Authentic learning requires that students have access to the criteria by which they are judged, that they can monitor and regulate their own progress, and that dialogue with tools and teachers remains open and contestable (Vygotsky, 1978; Flavell, 1979; Bakhtin, 1981). An AI system that produces predictions without explaining them violates all three conditions at once. The AI Act (EU 2024/1689), Arts. 9, 10 and 14, translates this into normative obligations: human oversight, data quality, and explainability of high-impact decisions, treated here not as bureaucratic constraints but as minimum pedagogical conditions.

The contribution adopts a two-level mixed design. At the macro level, an empirical analysis of territorial bias across 190 European NUTS2 regions uses Eurostat data on early school leaving rates and regional GDP per capita (2024) to show which students an algorithm calibrated on the European average renders structurally invisible. At the micro level, four case studies from the Educational Data Mining Conference 2023 are analysed for the centrality they assign to teachers and students.

At the macro level, a statistically significant correlation emerges between regional GDP per capita and the algorithm's prediction error (r = −0.194, p = 0.007): disadvantaged regions are systematically underestimated as "low risk", with a structural gap of 0.68 percentage points (t-test p = 0.044). The bias follows economic, not geographic, lines (Kruskal-Wallis p = 0.32): 82 false negatives appear because students live in low-income contexts the algorithm does not account for.

At the micro level, four recurring conditions emerge. Gabbay and Cohen (2023) show that metacognitive feedback in programming MOOCs produces more autonomous learners than systems providing solutions directly. An Accountable Talk case mediated by GPT-4 (EDM 2023) demonstrates that language models can sustain critical discourse when designed to elicit argumentation, not replace it. Wagner et al. (2023) show that a transparent recommendation system reduces dropout without a self-fulfilling prophecy effect. Karimov et al. (2023) document how an adaptive platform in a disadvantaged context in Azerbaijan improved grades for 68.6% of students, showing that algorithmic equity is a matter of design.

Both levels converge: AI is compatible with education only if verifiable — if teachers and students can interrogate it, challenge it, and correct it. Without explainability and human accountability, the school does not produce critical learners: it produces consumers of predictions who do not know they can refuse them.