Submission 36
Generative AI in Computational Thinking: Assessing Student Reasoning
Presented by: NOGA REZNIK
Computational Thinking (CT) is a core competence in higher education and the labor market (Wing, 2006). The rise of generative AI (GenAI) is reshaping the epistemic and pedagogical foundations of CT and programming education (Kohen-Vacs et al., 2025). Because many novices already struggle to articulate their logic, AI further complicates the interpretation of learning processes. Higher education is therefore reexamining teaching and assessment practices to ensure responsible AI integration and sustain meaningful CT engagement (Kurtz et al., 2024).
This paper presents the initial phase of a design-based research study focused on the use of an open-source environment using Blockly library providing the functionalities available in Scratch environment which are extended as part of our development effort with GenAI integrations. Connected to OpenAI and Gemini APIs, the system enables students to insert AI-query blocks during tasks, positioning AI as an explicit, student-controlled tool rather than an automated guidance mechanism.
The environment enhances the visibility of learners’ reasoning, preserves agency in AI-rich contexts, and shifts assessment toward process-oriented evidence. It captures artefacts and interaction traces while embedding metacognitive prompts at two key moments: students articulate their initial plan at task onset and reflect upon submission. By combining reflections with programming traces, the system operationalizes a process-aware model of CT. During Fall 2025, the environment was implemented with 60 first-year undergraduate students in an “Introduction to Computational Thinking” course. Students completed structured programming tasks as part of regular assignments. Initial findings suggest promising outcomes: students demonstrated high engagement, and the course lecturer reported improved visibility into reasoning processes, effective monitoring of AI use, and more targeted formative feedback. These early indications align with the environment’s design goals preserving learner agency in AI-rich contexts, making thinking processes pedagogically visible, and re-anchoring assessment in process-oriented evidence rather than solely in final artefacts.