Google Research says its AI experiment lets teachers create custom learning simulations

Google Research describes curriculum-focused creation

Google Research says its new experiment allows educators to create interactive educational simulations tailored to their teaching objectives and curriculum. The organization describes a system that generates interfaces dynamically and announces an English-language sample library containing more than 30 activities across physics, chemistry, biology and mathematics, focused on middle and high school.

According to Google Research, teachers start by suggesting a topic. The system then generates learning objectives that teachers can modify and must approve before an interactive is produced. Those objectives guide a sequence of game-like levels, with challenges designed to become progressively harder as students explore the topic.

Google Research outlines guidance and review

Google Research says the activities include introductions, relevant formulas and theories, layered hints, feedback and worked solutions. It describes these features as support for learners working through challenges and says they were tested and refined with teachers and students. The supplied account presents this guidance as part of the instructional design.

The organization also describes automated checks that examine educational design, interface behavior and whether challenges can be solved. Some checks interact with a simulation and test extreme control settings. A positive automated result does not constitute factual certification or human approval. Separately, Google Research says teachers approved the released library activities and that new additions require teacher validation and approval.

What the technical report actually evaluated

The linked technical report, arXiv version 1 dated September 17, 2026, describes 12 U.S. teachers making 36 requests. Researchers generated five candidates per request and selected a result internally; three requests produced no candidate considered good enough to show the teacher. Feedback on the selected results therefore does not measure the reliability of an unfiltered first attempt.

A separate evaluation used 40 STEM requests and subject teachers to rate generated simulations. Both exercises assess the simulations and teacher experience. They do not establish improved student learning in classrooms. The report identifies that evaluation as future work. This is an author-reported preprint, not an independent replication by The Illustral.

A practical review before classroom use

Our suggested review starts with one learning objective: what should a student be able to explain after the activity? Work through one ordinary case and one boundary case, compare the result with the course material, and inspect the hints and worked solution. Record any mismatch before considering the activity for students.

For a hypothetical arithmetic activity, a teacher could check zero and a negative input as well as a positive one. A working button alone would not answer whether the mathematical explanation is correct or whether the exercise meets the lesson objective. These are proposed checks, not a report that we ran Google’s simulations.

Update history and evidence limits

Updated October 13, 2026 (UTC), following source review on the same date. The technical report has now been consulted alongside the original announcement. The update adds study-design context and a suggested review exercise. We have not reproduced the research, measured learning outcomes or tested the product. The original publication date is retained.