Articles · AI in education · Teaching

Tutor, coach, collaborator, never ghost: teachers in charge of AI

The same AI can tutor, coach, collaborate, or ghostwrite. How schools can name those roles, keep the thinking with students, and keep teachers in charge.

By Dr. William Hickman · 5 min read

Open almost any AI tool and you will find a single empty box. Type into it, and the same machine can play four very different roles. It can be a Tutor that helps a student learn to do something. A Coach that helps her do it better herself. A Collaborator that works alongside her while she stays the author. Or a Ghost that does the work in her name.

There is no switch labeled Tutor or Ghost. The words decide which one you get, and most general-purpose tools are tuned to be as helpful as possible, which in practice means handing over the finished thing. Ask for help with an essay and you often get an essay. That is Ghost mode by default.

In a classroom, the choice of role is too important to leave to a default. Here is how I would put teachers back in charge of it.

Four roles, one machine

  • The Tutor helps you learn to do it. It asks questions, gives hints, checks understanding, and makes you do the work. Its goal is to make itself unnecessary.
  • The Coach helps you do it better yourself. You did the work first. The Coach gives feedback, and you decide what to change.
  • The Collaborator works alongside you while you stay the author. It drafts pieces, generates options, or gathers research while you direct, decide, and own the result.
  • The Ghost does it in your name. What comes back goes out as yours, with little or no change.

None of these is evil. A Ghost is fine for pure grind with no learning goal and no personal meaning. The question is never only “Is this allowed?” It is “What does this task exist to build, and which role protects that?”

Why the Ghost is the risk for students

For a student, most schoolwork exists to build something in her head. Hand that work to a Ghost and the product looks great while the learning quietly disappears.

The research shows both sides clearly. In a randomized trial at Harvard published in 2025, Greg Kestin, Kelly Miller, and colleagues built an AI tutor around good teaching practice. It nudged students through physics problems instead of solving them. Students learned more than twice as much as students in a well-taught active learning class covering the same material, and in less time. That is the Tutor at its best.

Now the Ghost. In a study of about 1,000 high school math students in Turkey, Bastani and colleagues (2025) found that students with unrestricted chatbot access did better on practice and then scored about 17 percent lower on the exam than students who never had it. A version with teacher-designed guardrails avoided that harm.

Same kind of technology. Different role. Opposite outcome.

Put the roles in writing for students

A vague policy (“use AI responsibly”) leaves every student to decide alone, at night, with a deadline. Name the roles instead, by type of task.

  • Practice and homework help: Tutor. The tool explains, quizzes, and hints. It does not answer.
  • Writing the student drafted: Coach. Write first, then ask for feedback, not a rewrite.
  • Research and projects: Collaborator, with disclosure. The student sets the direction and can defend every part.
  • Anything meant to show what the student learned: never Ghost.

Then make it visible. Add a three-line cover sheet to every assignment where AI is allowed: what I used, what I asked it, what I changed. And teach students to name the hat at the top of a prompt: “Tutor me on this.” “Coach me; don’t rewrite.” Saying “Ghost” out loud makes a student hesitate on exactly the tasks where she should.

Put the roles in writing for teachers

Teachers deserve the same clarity about their own use.

Ghost is reasonable for grind: reformatting a unit calendar, turning notes into a supply list, a first pass at routine logistics. Coach fits the words that should sound like you, such as feedback to a student or a message to a family. Write first, then ask for help tightening it. Collaborator fits planning, where the tool proposes and you hold the standards.

Some calls belong only to a person. When AI shows up inside platforms as recommendations about students, I use a simple boundary. An AI tool may recommend, and must explain its evidence: a next task, a regrouping queued for your review, a flag that deserves a human look. It may never decide on its own that a student mastered a standard or earned a grade, change a placement no adult confirmed, alter an accommodation or an IEP or 504 plan, contact a family, open a discipline record, or put a label like “at risk” in a permanent record. And student data is never used to evaluate a teacher.

When you override a recommendation, write one line saying why. Over a semester, those lines teach your team, and the tool, what the data could not see.

Protect the reps

Every expert was once a beginner who did the groundwork. The groundwork looked tedious, and it was also the training: thousands of small attempts, small mistakes, and small corrections. When AI does the groundwork instead of the learner, it can saw off the bottom rung of the ladder the learner was supposed to climb.

Aviation figured this out first. In 2013, the Federal Aviation Administration issued a safety alert warning that constant use of automated flight systems does not keep a pilot’s manual flying skills sharp, and it advised giving pilots regular chances to fly by hand. Students need the same thing. Not on every task, but regularly, they should do the core work without AI, then use the tool to check and extend it.

And change the question you ask about finished work. Instead of only “Is this right?”, ask “Walk me through how you got here. What did the AI miss?” If a student cannot answer, the work is not done, no matter how good it looks. A coach does not judge a player by the highlight reel. She watches the footwork.

What to do this week

  1. Label three assignments with the role AI may play: Tutor, Coach, Collaborator, or none.
  2. Add the cover sheet: what I used, what I asked it, what I changed.
  3. Try “name the hat” with your class for one week, and do it yourself.
  4. Check your platform’s defaults. Does your main AI tool act like a Tutor or a Ghost for students? Ask whether that setting can be changed, and whether students can switch the guardrail off.
  5. Hold five “walk me through it” conversations about finished student work.

The four roles, and the Four Keeps behind them, are at the heart of The Outsourced Mind.

Dr. William Hickman (Dr. Philip William Van Hickman) is an inventor, former superintendent, and the founder of MindAble Solutions. He is the author of eleven books, including The Intelligent Classroom. About him · Official site