AI Principles

Correctness is checked by science, not by a model's confidence.

Structures, equations, units, stoichiometry and mechanisms are validated with scientific libraries and deterministic checks before any AI-written explanation is shown. Where a model and a chemistry engine disagree, the engine wins and the discrepancy is logged for review.

AI should support practice, feedback and explanation rather than remove the thinking a course is intended to assess.

Student-facing AI gives hints, explains a result, or answers a question about the textbook. It does not complete assignments, and hints are scaffolded so the next step is still the student's.

Where a course wants no student-facing AI at all, it turns it off and keeps every other feature, including the Supplemental Instruction materials in the higher packages.

Teachers choose when AI is available and what it may do.

Every AI feature in a course is a setting the instructor sees: on or off, which assignments, which capabilities. Course-building suggestions from the assistant are proposals the instructor approves; nothing is applied silently.

Use answer keys, parsers, chemistry engines, calculations and deterministic checks rather than asking a model to invent a grade.

Numeric, multiple-choice, short-answer, structure and mechanism questions are scored by rules and engines that give the same answer every time. A model may explain why an answer was wrong; it does not decide that it was.

For subjective feedback or grading assistance, instructors remain responsible for official assessment decisions.

Where AI drafts feedback on open-ended work, it is labeled as a draft, kept apart from the attempt record, and released only by the instructor. Chat is separate from grades.

Training on student work is off by default.

Homework attempts, notebook entries and assistant conversations are not used to train models, ours or anyone else's. Providers are used under API terms that prohibit training. Any future opt-in would be written into the Privacy & Student Data page before it existed.

Students and instructors should be able to tell which features use AI.

AI-assisted features are labeled where they appear, say which provider is in use, and show what was sent. Generated text is marked as generated until a person has reviewed it.

A course can use CourseHouse's books, homework and tools without making generative AI the educational experience.

The base Homework package has no student-facing AI at all. AI is priced low into the packages that include it and is never required to read the book, do the homework or receive feedback.

Sensitive educational content gets more restrictive treatment than generic public content.

Requests carrying student work or identifiers go to providers and settings with no-training, limited-retention terms. Public textbook text can use a wider set of models. Institutions can restrict routing further in their agreement.

Use AI to draft descriptions, transcripts and alternative representations that a person then reviews — so accessible materials arrive with the course, not after it.

AI drafts alt text and long descriptions for figures, transcripts for video, and plain-language readings of structures and data; instructors review and approve them before students see them. Accessibility artifacts travel with the course into EPUB, PDF and LMS delivery.