Google Research is testing a system that lets teachers generate interactive STEM simulations tailored to a lesson rather than relying on a fixed library. The experiment uses generative user interfaces, meaning AI constructs the controls, visuals and feedback for an activity instead of drawing from a screen coded in advance.
Teachers first define and approve learning objectives. The system then creates progressively harder challenges, hints, formulas, feedback and worked solutions. Automated agents test whether each simulation can be solved, whether its controls work and whether extreme inputs expose errors. The generation loop repeats until the activity meets the project’s pedagogical, mechanical and visual criteria.
Google has released a public sample library of more than 30 English-language activities covering middle- and high-school physics, chemistry, biology, mathematics and computer science. Every published example was reviewed by teachers. In an initial US study, 12 teachers each requested three custom activities and gave them an average quality rating of eight out of ten. That is encouraging but still a small early evaluation, not evidence of improved student outcomes. Schools using Google Workspace for Education can apply to a pilot program and provide feedback. The prototype keeps educators in charge of the objectives and where an AI-generated simulation fits into the curriculum.