For its 2023 offering, Harvard’s introductory computer science course CS50, taught by David Malan, built its own AI tutor and folded it into the course’s official policy. Known as cs50.ai or the CS50 Duck, the tool is a “rubber duck debugger” - a nod to the practice of explaining a bug aloud to a rubber duck - that the course says “leverages the tools of Azure and OpenAI, along with our own vector database” holding content from recent lectures.
The course’s policy, on the official CS50 notes page cited here, draws a sharp line: “In CS50, we do not allow the use of ChatGPT,” but CS50’s own duck at cs50.ai is permitted. The design intent is that the Duck “behave more like a good tutor,” leading students toward answers rather than spoiling them, in effect offering “office hours” 24 hours a day. The course also warns students that the tool “is an AI and that it is experimental” and should not be blindly trusted.
CS50’s approach is notable because it neither banned AI outright nor allowed any chatbot. Instead the course supplied a constrained, course-aware tutor and prohibited general models, distinguishing a tool tuned to teach from one happy to do the work for the student.
Why business readers should care: CS50 models a middle path between banning and freely permitting AI - provide a sanctioned, purpose-built tool and restrict the general ones. That same “approved internal assistant, restricted public ones” pattern is how many organizations now govern generative AI use.