STEPS, developed by Shima Salehi and chemistry lecturer Jennifer Schwartz Poehlmann, asks students questions as they work through STEM problems instead of giving the answer. In classical Japanese, Ariel Stilerman’s BungoBot likewise responds to a submitted sentence with questions that guide the student’s interpretation rather than immediately supplying a translation. Stanford describes both approaches as ways to support the process of working toward an answer.
These examples put a different requirement on the model: its response must move the learner’s reasoning forward without taking over the task. A useful review would look at the questions it asks, when it reveals information and whether the student can explain the next step. The university report offers instructor and student accounts; it does not establish comparative learning gains through a controlled evaluation.
Across the examples, the practical design choice is where responsibility stays. A lecturer answers grouped questions, subject experts curate prerequisite materials, and learners work through a tutor’s prompts. Choosing that boundary first gives a teaching team something more concrete to assess than whether an AI tool sounds knowledgeable.
AI-generated conceptual illustration of an instructor organizing questions while adult learners work through ideas. It does not depict the Stanford instructors, students or tools described in this report. Original artwork by OpenTools Team, based on Stanford’s teaching report.