Dr. Maciej Pankiewicz’s Keynote Talk

From ChatGPT to Classroom Impact: Causal Evidence for LLMs in Education

Generative AI has entered education at an unprecedented pace, expanding from simple chatbots into feedback tools, tutors and content generators. Early studies reveal a surprisingly strong body of positive evidence, reporting improved task performance, faster feedback, and higher learner satisfaction across many contexts.

As researchers, we should read these encouraging results with a measure of skepticism, asking what is actually being measured and whether the method truly provides evidence for the claims being made. Are we observing genuine learning, or only short-term performance and satisfaction? A related question then follows: are these tools truly helping all learners, or mainly those who are already confident and engaged? If AI assistance primarily benefits students who need it least, it may quietly widen existing gaps even as average outcomes improve.

This keynote explores how to understand AI's real classroom impact. Rather than asking whether a tool "works," I focus on how to make defensible claims about learning with LLMs: clarifying what the AI is doing instructionally, measuring outcomes beyond short-term performance, and separating tool effects from surrounding pedagogy.

Grounded in learning theory and real classroom evaluation, the talk emphasizes identifying mechanisms such as scaffolding, feedback timing, and self-regulation, measuring actual engagement with support, and treating heterogeneity as central: who benefits, who does not, and why.

Drawing on real deployments of LLMs in authentic learning workflows, I will discuss how to develop well-supported conclusions about what works, for whom, and when. The goal is to strengthen the empirical foundation for generative AI as a tool that broadens access to effective support rather than reinforcing existing advantages.

About Dr. Maciej Pankiewicz

Dr. Maciej Pankiewicz leads innovative research at the intersection of education and technology. With a background in computer science, he utilizes a contemporary toolbox of learning analytics, artificial intelligence, and large language models to enhance educational outcomes.

He has extensive experience as a software developer, data analyst, and researcher in educational data mining. His technical expertise encompasses a wide range of IT skills such as software development, data management, and predictive modeling using various technologies and programming languages. This expertise has been instrumental in his latest research work, aimed at designing intelligent educational systems using large language models.

Back to 2026 Keynote Speakers