My initial thesis presentation will show how serious games are becoming an important direction for education and how the rise of generative AI enables more personalized tutoring within serious games. This benefits both students’ learning progress and the efficient use of teacher resources. However, single GenAI solutions face limitations when used as tutoring systems in serious games. Agentic AI may address these limitations using specialized agents.
I will also present my research questions that aim to compare a single GenAI system with an agentic solution regarding the following metrics: correctness, didactical response quality, tool use, and computational resource utilization. The evaluation will be conducted with the VirtuChemLab, which currently employs a single LLM as its virtual tutor. Additionally, I will briefly introduce agentic AI, present the planned evaluation methodology and give an outline on the workplan.