AI Impact Assessment

PROODOS EduAI is a doctoral research platform developed at the International Hellenic University (IHU), Thessaloniki, for the professional development of K-12 teachers in artificial intelligence literacy. The platform is structured around 15 UNESCO-aligned modules followed by a synthesis Epilogue, and uses AI-generated feedback as a core pedagogical mechanism.

PROODOS is the empirical instrument of John Dourvas's doctoral dissertation, under the supervision of Asst. Prof. Georgios Kokkonis, Department of Information and Electronic Engineering.

Five AI-driven features are active during a participant's journey:

• RAG-based reflection feedback: after each module's reflection submission, a retrieval-augmented generation pipeline (Google Gemini 2.5 Flash) produces a personalised feedback narrative grounded in pedagogy literature.
• Reflective Tension Mapper (RTM): from the participant's reflection text, the system extracts up to three pedagogical tensions and asks the participant to self-position on a 5-point scale between two contrasting poles.
• Developmental Trajectory Predictor (DTP): a narrative summary of the participant's reflection trajectory across modules. Available from Module 2 onwards.
• Peer synthesis: drawing on pseudonymised peer reflections, this produces a comparative view of how other participants engaged with the same content.
• AI Literacy Sandbox (Aletheia): an optional practice environment unlocked after Module 5. Participants construct prompts across 48 curriculum-anchored scenarios (8 subject clusters × 6 responsible-prompting strategies) and receive AI-generated responses from Google Gemini 2.5 Flash. Sandbox sessions are recorded as research data per the research participation consent.

After Module 15, the PROODOS Epilogue presents a Personal Evolution Dashboard synthesising the participant's developmental trajectory across the 15 modules. The Epilogue is a non-AI synthesis surface: it aggregates existing AI outputs (DTP themes, RTM tension trajectories) and existing platform telemetry into a single read-only view; no new AI inference runs during the Epilogue itself.

Under our assessment of Regulation 2024/1689 (the EU AI Act), PROODOS is classified as a Limited Risk AI system falling under Article 50 transparency obligations.

The platform:

• Does not produce automated decisions about participants' professional, academic, or employment standing.
• Does not perform biometric identification, social scoring, or any practice listed as Prohibited (Article 5) or as High-Risk (Annex III).
• Generates content that the user reviews and may dispute (see Section 4).

The AI's role is assistive and pedagogical: it surfaces patterns, asks reflective questions, and offers commentary. Authority over interpretation remains with the educator.

Per Article 50 and the spirit of Article 14 (human oversight) of the EU AI Act:

• Explicit consent: use of AI assistance is disclosed before account creation and is revocable at any time from the Privacy dashboard.
• Output dispute: every AI-generated artefact (RAG, RTM, DTP) carries a dispute submission button. Flagged outputs feed back into the research dataset for AI-alignment analysis.
• No automated decision-making: AILST self-assessment scores are computed deterministically from the participant's own responses; no AI inference is involved in score calculation, gating, or certification.
• Pedagogical framing: AI feedback is presented as one voice among many, not as authoritative pronouncement. Module copy reinforces this stance throughout the programme.
• Continuous evaluation: AI dispute submissions, RAG telemetry, and AILST trajectory data are part of an ongoing research programme to refine the AI's pedagogical alignment.

Lawful basis under the EU General Data Protection Regulation (GDPR, Regulation 2016/679) and Greek Law 4624/2019:

• Primary basis: Article 6(1)(a) — explicit informed consent, captured at onboarding Step 3.
• Special category basis: Article 9(2)(j) — processing for scientific research with appropriate safeguards.

Retention and identifiers:

• Research data is retained for three years after data collection for this research ends, then transformed as described in the Privacy Policy. Participants can request this transformation earlier via the Privacy dashboard's account anonymisation action.
• All analysed data is pseudonymised before analytics processing. Free-text fields that may contain identifiers are cleared on participant-initiated erasure.
• IP addresses are captured at consent time only, automatically redacted after 30 days, and immediately cleared on erasure.

Optional data sharing for secondary research: a separate consent allows pseudonymised data to be used in further analyses approved separately by the IHU Research Ethics Committee. Optional and revocable independently of primary research consent.

Guaranteed by GDPR Articles 7, 15, 17, and 21 and operationalised in the Privacy dashboard at /profile/privacy/:

• Withdraw any consent at any time (Article 7(3)): three independent revoke endpoints, one per consent type.
• Download a full copy of your personal data in JSON format (Article 15).
• Anonymise your account permanently (Article 17): identifying information is removed; pseudonymised research data remains attached to an anonymous account and is excluded from future analyses through the opt-out flag set at the same time.
• Object to processing or contact the research team for questions about your data (Article 21).

For questions about this platform's AI use or your data:

• Principal Investigator: John Dourvas, doctoral researcher - idourvas@ihu.gr.
• Academic Supervisor: Asst. Prof. Georgios Kokkonis - IHU, Department of Information and Electronic Engineering.
• Institution: International Hellenic University (IHU, Διεθνές Πανεπιστήμιο της Ελλάδος), Thessaloniki.

This document is version v4_pre_irb (August 2026). Section 5 was corrected to state the retention period consistently with the Privacy Policy (three years after data collection ends, not seven), and the department name in this section and Section 1 was corrected to its official English form. It will be revised again after IHU IRB review.