Research program

Learning Copilot

Learning Copilot is a learner-facing AI tutor. It hints, asks, and retrieves a definition, and it stops short of handing over a finished answer.

Learning Copilot compared with A recorded lesson
A recorded lessonLearning Copilot
Who the lesson is forEvery student hears the same explanation.The student explores inside a topic the teacher already set.
What the student leaves withThe teacher’s explanation, played back.A sharper question, a revision, or the next step they still have to take.
When the answer appearsThe video gives it.The agent withholds it while the student is still doing the work.

What this agent is for

Learning Copilot holds a goal for a study session, looks at the learner's work, and takes a short sequence of steps: ask, hint, retrieve a definition, or wait. The product of those steps is a better question from the learner, not a completed assignment.

An agent is a poor tutor when every turn ends in the answer. It is a useful tutor when it can tell a reminder from a struggle that should continue, and when it can use a diagram, a spoken explanation, or a worked step without taking the thinking away.

Planning, tools, and restraint

The technical questions are the ones any education agent has to answer. What is the agent allowed to do on this turn? Which tool is appropriate: a hint, a retrieval from course material, or no tool at all? What should it remember about this learner, and what should it forget when the session ends?

Restraint is a design constraint. A copilot that only answers trains students to stop asking. A copilot that helps them sharpen a question trains a skill they can use when the agent is closed.

What it is not

Learning Copilot is not an answer engine with a tutoring costume, and it is not a claim that any one model already does this well. Essays on the learning theory and the technical limits are published in Writing as they are ready. NexEd describes who does this work on the About page.

Writing on this program