Conference Program

When students use generative AI to translate Latin and Ancient Greek, are they thinking more—or thinking less? This paper presents findings from an exploratory qualitative case study conducted in an Italian classical high school, examining how AI reshapes interpretative practices in subjects traditionally centred on judgement, ambiguity, and linguistic precision. Data were collected through structured classroom observations, analysis of student translations, and guided reflective discussions during AI-supported activities.

Two contrasting patterns emerge. Some students fully delegate the task to AI systems, reproducing translations without engaging in grammatical analysis or interpretative reasoning. In these cases, AI functions as a form of cognitive substitution, weakening agency and displacing responsibility for judgement. Other students, however, use AI as a cognitive scaffold: they compare outputs with their own attempts, question lexical and syntactic choices, and refine their understanding in preparation for oral examinations. Here, AI becomes a mediating tool that can potentially extend the learner’s zone of development.

These findings suggest that the educational impact of generative AI does not depend on the technology itself but on the pedagogical structure framing its use. As algorithmic systems trained on vast datasets, generative models embed implicit linguistic and cultural assumptions that shape the interpretative options presented to learners. Without explicit critical mediation, such systems risk becoming oracular authorities.

In response, the paper proposes a three-step instructional protocol designed to foster metacognitive awareness and dialogical engagement: (1) independent translation attempt; (2) systematic comparison with AI output; and (3) guided reflection on divergences, interpretative criteria, and underlying assumptions. By making algorithmic mediation itself an object of inquiry, students learn to interrogate digital power rather than passively absorb it. Such practices cultivate habits of justification, responsibility, and reflective judgement—capacities essential to democratic learning in an age of datafication.

The rise of Generative Artificial Intelligence (GenAI) in schools introduces a fundamental pedagogical tension: how can GenAI be integrated into classroom practice without encouraging cognitive delegation or passive reliance on AI-generated outputs? This issue connects to broader debates on the purposes of education and the risks of delegating educational processes to technological systems (Biesta, 2020; Selwyn, 2019). This study presents an intervention conducted in an Italian upper secondary school using a Specialised Conversational Agent (SCA) grounded in a Metacognitive Tutoring Model (MTM), designed to foster critical GenAI literacy, student agency, and reflective judgment.

We explore the conditions under which AI can enter classroom practice without replacing educational responsibility or teacher judgment. This perspective aligns with critical analyses warning against technologically deterministic narratives (Williamson & Eynon, 2020) and emphasizing the need to preserve teachers’ professional agency in AI-enhanced environments (Lan & Chen, 2024). The MTM underpinning the SCA aims to (1) reject the agent’s role as an oracle-like problem solver, (2) enact dialogic coaching and active scaffolding, and (3) create a “cognitive gym” supporting the transition from passive consumer to active inquirer. The model builds on foundational work on metacognition (Flavell, 1979) and on the role of self-efficacy in sustaining learner agency (Bandura, 1997). Through structured feedback, metacognitive prompts, and guided error analysis, the agent ensures that students remain the primary architects of the problem-solving process, learning to reformulate prompts and critically evaluate responses, including incomplete or misleading outputs. The agent is thus conceived not as a substitute for human intelligence, but as a system designed to augment and extend it (Luckin, 2018).

The intervention adopted a waitlist-controlled classroom design. All students received preliminary training in prompting, framed as a cognitive and deliberative practice emphasizing clarity, intentionality, and iterative refinement. In the first phase, one group participated in structured mathematics sessions with the SCA, while a second group followed traditional instruction. In the second phase, the latter received prompting training and used the SCA, ensuring full participation. This design enabled observation of learning trajectories and the development of competence in using GenAI systems. Rather than focusing on disciplinary performance, the study examines empowerment and metacognitive indicators. Data include anonymized student–agent interaction excerpts, the evolution of prompt patterns, and a self-perception questionnaire measuring perceived agency, awareness of GenAI limitations, prompt reformulation ability, and recognition of incomplete or ambiguous responses. The central question concerns students’ positioning toward GenAI: do they adopt uncritical cognitive delegation (GenAI as an “epistemic oracle”), or maintain cognitive agency, treating GenAI as a co-cognitive dialogic partner? This inquiry reflects concerns about the future of teaching in AI-mediated contexts (Selwyn, 2019) and the need to design systems that support rather than displace pedagogical judgment (Lan & Chen, 2024). We argue that critical AI literacy cannot be reduced to ethical awareness or technical proficiency. It requires pedagogical devices that preserve deliberation (Biesta, 2020), safeguard teacher judgment, and foster students’ responsibility in learning. The proposed model is presented as a replicable instructional framework currently undergoing empirical validation.

Ancient Greek teaching is an intensive practice of evidence-based interpretation: learners must justify morphological parses, evaluate syntactic ambiguity, and defend translation choices that remain contestable. Generative AI can support this work, but its conversational authority risks automating judgment and encouraging cognitive dependency. Responding to Panel M.03’s focus on critical dialogue with AI, this paper connects Freire’s pedagogy of autonomy (ethical self-formation through responsible decision-making) with Illich’s threshold logic (tools crossing into radical monopoly) and critical AI approaches that treat AI as sociotechnical agency that must be pedagogically mediated, not merely adopted.

We propose a Dialogic Metacognitive Protocol for Ancient Greek (DMP‑GR), a five-step lesson sequence that treats AI as a dialogical partner rather than an oracle and that preserves teacher agency:

(1) Problematize: the teacher frames a “philological problem” in a short Greek passage (e.g., participial scope; discourse particles; aspect and modality) and makes explicit democratic criteria (reasons, evidence, openness to alternative readings).

(2) Externalize: students generate multiple AI outputs through prompt-variation (constraints, counterfactuals, “show two parses”, “argue against yourself”), documenting outputs and uncertainties.

(3) Cross-examine: groups test claims against grammars, lexica, and corpora; they identify where AI is plausible but unsupported, and where it is wrong but persuasive.

(4) Metacognitive reflection: students write a short “philological decision log” explaining how they updated their confidence, what evidence mattered, and which biases (AI or human) may have shaped their choice.

(5) Dialogic deliberation: students defend a translation and respond to objections; the class co-constructs a justified version plus an “alternative readings” appendix.

DMP‑GR shifts evaluation from product to process: the teacher assesses traceable reasoning, responsible tool-use, and dialogical participation. Expected democratic outcomes include autonomy in deciding when and how to use AI, epistemic humility, plural reasoning, and practical wisdom (phronesis) in AI-mediated inquiry.

Generative artificial intelligences (GAI), through their rapid development, are profoundly reshaping the educational landscape. Current literature highlights how the use of A.I in education can positively affect several dimensions, including instructional personalization, learning outcomes, collaborative and communicative dynamics, reduction of cognitive load, and overall enhancement of cognitive skills. At the same time significant risks have been identified, such as overreliance and cognitive deskilling. As reported by Tian & Zheng (2025), the effects of A.I on the development of skills such as the 4Cs are also influenced by the role it assumes within the educational process (e.g., tutor, peer, or collaborator).

This makes it strategically important to define conditions and interaction models capable of guiding the use of A.I, framing it as a tool for cognitive augmentation rather than as a purely performative and opaque technology.

Seymour Papert’s constructionist pedagogy offers a potential interpretative key, both due to the centrality of creative activity in the process of knowledge construction and through what he define as “objects to think with”: tools enabling students to engage in cognitively meaningful activities. Levin et al. extend this concept to A.I, shifting from objects to “agents to think with.” These can function as maieutic tutors and cognitive exoskeletons capable of stimulating reasoning while keeping learners pivotal in their cognitive processes, thus preserving agency and cognitive abilities.

This perspective suggests a reversal of the interaction axis between user and A.I where, rather than having a learner passivly searching for an output, we figure an agent guiding him maieutically fostering cognitive processes, divergent thinking, creative and critical skills.

Along these lines, this paper aims to present the Tool Engagement × Human Engagement (TExHE) interaction paradigm. Designed to stimulate user agency as defined by Bandura (2001)—intentionality, forethought, self-reactiveness, and self-reflectiveness—within constructionist learning contexts, TExHE supports the configuration of co-constructionist “agents to think with,” outlining their characteristics and pedagogical aims. The agent’s behaviors (Tool Engagement) should be modeled both according to Levin’s definition of “agents to think with” and according to Resnick’s (2018) principles of effective educational tools: low floors (accessibility), high ceilings (high potential), and wide walls (flexibility). These features are intended to foster Human Engagement, understood as the degree of cognitive and self-regulatory involvement through which the learner guides, structures, and evaluates their own mental processes.

The relationship between Tool Engagement and Human Engagement is therefore asymmetrical and dynamic: the agent’s intervention remains subordinated to, and functionally oriented toward, the preservation and enhancement of the learner’s agency.

A first application of this paradigm, currently under field experimentation, is the Gemini Gem Clair. As a constructionist tutor, Clair supports users in creative processes by integrating the TExHE paradigm, which determines its behavioral logic, with the Creative Problem Solving Framework 6.1, which structures its operational workflow, acting both as a facilitator and as an enhancer of creative performance. Moreover, its implementation through Gemini Gems ensures replicability and accessibility of similar tools within school contexts, offering design possibilities potentially within teachers’ reach.

The widespread adoption of generative AI in educational and professional contexts has outpaced the development of frameworks supporting critical, metacognitive engagement. Most users approach conversational AI as a digital oracle — seeking speed and immediate output — rather than as a dialogical partner for meaning-making. This pattern mirrors a structural problem: the very affordances that make generative AI powerful (fluency, responsiveness, apparent authority) also make it prone to fostering cognitive delegation rather than reflective thinking. This paper presents rAIse (Reflect-Ask-Interpret-Source-Evolve), a five-phase framework for conversational AI literacy developed within the AI4E Laboratory at UnitelmaSapienza, and reports empirical evidence of its effectiveness across multiple learning contexts.
rAIse is grounded in five foundational theoretical traditions: Flavell's metacognition, Vygotsky's Zone of Proximal Development, Bakhtin's dialogism, Schön's reflection-in-action, and Freire's generative pedagogy (Evolve). Each phase is operationalized through stimulus questions that slow down and structure human-AI interaction, shifting engagement from mechanical to strategic and from passive to reflective. The framework explicitly repositions AI not as an authoritative source but as a co-constructor of knowledge requiring active critical participation from the user — a dialogical partner rather than an oracular authority.
A pre-post single-group design was conducted across multiple professional and educational contexts — university students, in-service teachers, private sector professionals, and public administration officers — through the AI4E Laboratory at UnitelmaSapienza. A total of 657 participants completed the rAIse process module and an accompanying validated questionnaire (24 items; Cronbach's α = 0.826 pre, 0.827 post). Two conditions were compared: facilitated workshops (2–4 hours, with theoretical introduction and guided application) versus fully autonomous self-application without prior training or facilitation. Four AI literacy competencies were measured via five-point Likert scales: use of AI for professional support, use of AI for information research, critical analysis of AI outputs, and ethical awareness.
Wilcoxon signed-rank tests revealed large effect sizes across all four competencies (r = 0.619–0.838, p < .001). A counterintuitive and theoretically significant finding emerged: participants who applied rAIse autonomously — without any prior facilitated training — showed systematically greater improvement than workshop participants, with the most pronounced difference in ethical awareness (+0.123 effect size). This finding suggests that the framework possesses intrinsic self-scaffolding properties: learning occurs through structured doing rather than through listening to explanations, positioning the AI conversation itself as the formative environment. Moreover, participants who completed the EVOLVE phase reflective item demonstrated significantly higher gains in critical analysis (U = 47566, p = .044), confirming that metacognitive writing is constitutive to the learning process. Reduced post-intervention standard deviations across all competencies further indicate that rAIse reduces initial variability, functioning as an equalizing scaffold within heterogeneous groups.
Three conclusions emerge. First, rAIse is effective transversally across diverse professional and educational profiles, from primary teachers to public sector officers. Second, traditional didactic mediation is not a prerequisite for its effectiveness: the framework's structure is sufficient to generate meaningful learning independently. Third, the challenge of AI literacy is ultimately cultural rather than technical — it requires reclaiming metacognition, slowness, and democratic agency as non-negotiable dimensions of human-AI interaction in educational settings.

Abstract

The growing presence of generative artificial intelligence (GenAI) in contemporary societies, and particularly in education, is reshaping how knowledge is produced, accessed and interpreted (Brandao, 2025; Rossi et al., 2025). Within broader processes of digitalization and datafication, algorithmic systems increasingly influence the circulation of knowledge and the ways individuals engage with information. However, whether GenAI represents primarily risks or opportunities for younger generations remains relatively underexplored. While its use may become problematic when it is purely instrumental and oriented toward rapid output production, it may also foster meaningful learning when it promotes reflective and critical engagement with knowledge and supports the development of AI literacy and critical thinking skills (Walter, 2024). These transformations raise important questions regarding whether students possess the tools to critically engage with artificial intelligence and whether these technologies can contribute to the democratic mission of education, particularly in relation to citizenship formation.

Within the field of sociology of education, these developments call for renewed reflection on algorithmic authority, digital mediation and the ways students interact with algorithmically generated knowledge. In response to these challenges, this paper presents a classroom laboratory designed as a pedagogical and sociological experiment in which university students engage critically with generative AI systems while exploring sociological concepts such as identity, social roles and inequality. The objective is not merely to employ AI as a technological tool, but to examine how algorithmic responses may reflect, reproduce or reshape existing social narratives.

Methodologically, the study develops a classroom laboratory adopting a qualitative and exploratory approach grounded in participatory peer-based research methods. The activity takes place within the university course Sociology of Education in the Digital Age at the University of Turin (academic year 2025–2026) and involves a cohort of 35 students. The study dynamics are implemented through the use of the generative artificial intelligence platform ChatGPT (OpenAI) and structured instruments designed around key sociological concepts such as identity, social roles and inequality. The data generated—including students’ prompts, AI responses and collective classroom reflections—are analyzed through a qualitative interpretative approach aimed at identifying patterns of reasoning, emerging narratives and processes of metacognitive reflection on algorithmically mediated knowledge.

By encouraging students to interrogate the assumptions embedded in AI-generated discourse, the laboratory promotes metacognitive awareness of how knowledge is constructed within digitally mediated environments and supports the development of student agency in relation to algorithmic systems. The study argues that such pedagogical practices can strengthen critical engagement with emerging technologies while addressing broader democratic challenges associated with artificial intelligence, including the capacity to recognize algorithmic bias, counter disinformation and foster forms of digital citizenship in contemporary educational contexts.