How We Use AI
Stemble offers many features for STEM courses; AI-assisted feedback and assessment is one option that is available to support instructors and students. Not all courses use AI-assisted features; this is up to the discretion of the institution and the instructor (on a course by course basis). In other words, Stemble provides the tool, and the institution and its instructors decide whether to use it. Instructors/TAs can choose to review AI feedback before it is released to students. Tasks that use AI-assisted grading are clearly marked for instructors when setting up assignments.
What AI does Stemble use?
Stemble uses commercially licensed OpenAI models through OpenAI’s business API, under a signed Data Processing Addendum. This is a separate service from consumer ChatGPT and operates under different data terms. Stemble does not build or train its own models for these features.
Is student or instructor work used to train AI?
No. OpenAI’s business terms state that it will not use the data we send to develop or improve its services, and Stemble does not fine-tune models or build training sets from institutional content. Under our data processing agreement, OpenAI deletes data sent through its API within 30 days, unless the law requires it to keep it.
What does the AI actually see about a student?
The question, the rubric, Stemble’s grading instructions, any example answers written to guide grading, and the student’s submitted work, including any images they upload. When a student resubmits, the model also sees their earlier answer and the feedback on it. We do not send identifying metadata such as names, email addresses or student IDs. Submitted work is sent as the student wrote it, so anything a student types into an answer or captures in an image is included.
Who sets the grading standard: the AI or the instructor?
The instructor. Each AI-graded task is assessed against a rubric developed with the instructor’s input, and that same rubric applies to every submission for the task. Instructors can review and override any individual assessment.
Can students see why they received a grade?
AI-assisted assessments use a rubric and come with written feedback. Instructors can choose whether to make this feedback visible to students. Our goal is that students and instructors can see the reasoning behind a result instead of an unexplained score.
Could grading change without warning when AI models update?
Each AI-graded question uses a specified model version; it does not move automatically when a provider releases a new model. Stemble may move questions to a new model, for example when a provider retires a version. We evaluate new models before they are adopted.
How does my institution evaluate whether Stemble fits its requirements?
Through your institution’s standard procurement, security, privacy, accessibility, and legal review. We can provide detailed documentation to support that process.
What if a student has a question or concern about AI-assisted grading?
For general questions about how Stemble and/or AI are being used in their course, students should contact their instructor. If you are using Stemble and have technical questions or issues, navigate to the support page in the Stemble app and create your request there.