R&D Unplugged

What happens when students use AI to study and where is the line between support and doing the thinking for them?

In this episode, we explore, together with our guest Liudmila (Mika) Piatnitcskaia, how students use ChatGPT during learning, what kinds of tasks they delegate to AI, and why the most “effortful” parts of thinking may be the ones we need to protect most. We also discuss an experiment with two AI tutors: one that gives answers more directly, and one designed to guide students toward the answer themselves.

Together, we ask how educational AI can be designed to support learning without replacing human thought.

The proliferation of complex and often conflicting digital information demands advanced cognitive strategies to discern source trustworthiness effectively. Yet, many readers remain on “autopilot” or favour personal justification over investigating who's behind the information. Let's explore how different educational strategies can break these automatic habits and foster a more analytical mindset. Drawing on a study of 179 university students, Gaston Saux and his team compared traditional teacher-led instruction with symmetric peer-tutoring. Their findings reveal that while both methods were effective, peer-tutoring's effect was clearer when participants justified their evaluation decisions in a specific task and reported their criteria for justifying their trust in online information in general. By bridging the gap between cognitive theory and classroom practice, this research aims to reflect on what interventions are more effective to promote critical source evaluation.

More about the speaker
Gaston Ignacio Saux is a researcher at the Argentinean National Scientific and Research Council (CONICET) and head of the “Text Comprehension and New Reading Contexts” program at CIPP-UCA (Buenos Aires). After earning a Ph.D. from the University of Buenos Aires and completing a CNRS-funded postdoctoral position at the University of Poitiers, he shifted his focus from inference generation to strategic source evaluation while reading in unreliable digital environments. Currently, his research team seeks to bridge the gap between basic cognitive research and effective, reflective reading practices.

In this episode, we explore a personal and professional journey beyond the code. Our guest, Colin de la Higuera, shares the realization that guiding research toward human values is not something that can be improvised; rather, it has become a vital field of research in its own right.

We move beyond the personal narrative to examine the critical interactions between Artificial Intelligence and the sharing of knowledge. With a primary focus on education, we discuss how we can intentionally steer AI to empower learners and democratize information, rather than letting it be defined by market forces alone.

In this episode, we explore creativity not as a mysterious gift, but as a process that can be studied, modeled, and nurtured. At the crossroads of computer science, cognitive science, and the learning sciences, our guest, Chloé Mercier, examines creativity as a situated and iterative learning process shaped by reflection and feedback.

Using computational approaches, we discuss how creative work can be analyzed without reducing its educational richness. These models help us understand how learners generate ideas from prior knowledge and environmental stimuli, revise their work, and gradually develop creative strategies.

Looking ahead, the conversation turns to writing education and the goal of fostering creativity alongside metacognitive awareness, not automating production. We also reflect on digital humanities and STEAM education, asking whether bridging the sciences and humanities could support more inclusive pathways and help address persistent social and gender inequalities.

In this episode, Jean Vassoyan explores a fundamental question in machine learning: how can we design “teacher” algorithms that effectively guide “student” agents?

Using a broad and abstract framework, he connects fields such as Curriculum Learning, Active Learning, and Reinforcement Learning, showing how they share a common mathematical foundation aimed at optimizing learning.

Applied to human education, this approach becomes Adaptive Learning, intelligent tutoring systems that personalize instruction for each learner. By viewing machine and human teaching through a unified lens, Jean invites us to rethink what it truly means to be a teacher and how this role can be defined and formalized in quantitative terms.