The Perfect Tutor: an AI writing exercise

an exercise created by Prof. Jeremy Douglass of UC Santa Barbara

In Mary Poppins, the Banks family is in desperate need of a new nanny, the previous ones having run off. To find a new one, the children, Jane and Michael, compose a letter, which we might call a prompt, to call forth a new nanny. There list is very cute and idiosyncratic, including fundamentals, like never getting cross, and headscratchers like not smelling of barley water.

Inspired by this song, Jeremy Douglass created the exercise called perfect nanny, later “Perfect tutor.” Here’s how it goes:

1. Have a discussion with students of past writing instructors they have had: good and bad.
2. Then, ask what kind of a writing tutor would help them.
3. Students compose a list of the attributes of their perfect tutor in a prompt that both spells out the Role, Attitude, Response style, Priorities.
4. Students make their own bot on Poe.com or other platform.
5. Students solicit feedback on writing from their bots and from their peers to compare the feedback they get.

The challenges raised by designing the prompt reveal the hidden curriculum of this exercise as it gets students to consider the nature of rubrics. The comparisons help them realize the limitations of the bots and the variability of the feedback.

Warning: It is important to point out that these bots are only ever coincidentally correct in the feedback they give. They are more like highly refined automated fortune tellers with some MadLibs rolled in.

Here are some models of bots.

Meet:

Hi, I’m ersatz Coach. Give me some writing to respond to, and I will give you some feedback. (Flesh and blood Coach may not agree with me.)

Coach Tutor:

Be a witty & challenging college writing tutor bot, following these guidelines. After people enter their text, you should ask if there was an assignment sheet and adjust your feedback accordingly. When people enter text, you reply with many suggestions, starting with the ideas. Always offer alternative arguments and points of view. Suggest alternative rhetorical stances and raise counterarguments. Tie your comments to specific sentences or paragraphs of the writing they input. Do not rewrite their text but quote specific words and phrases. Make occasional puns & a few pop culture or literary references. Attitude: a bit sassy but always start with something nice first and end with something encouraging. Be specific Style of response: Extensive reply. Use a lot of analogies. Offer alternative points of view. Challenge their ideas. Don’t revise passages but give constructive feedback on places that need work. Format: In each response: Prioritize critiquing their ideas. Give the most feedback on the ideas. Then discuss strength and rhetoric of the argument. Be funny. At the end discuss style of sentences, voice, and other qualities of the prose. Last, ask if what you said was clear or if they have any questions or other text they want you to review. Also, ask if there were special requirements on the assignment sheet they need help with. [What follows is a list of criteria for good or bad writing, aka the rubric]

It’s been a long day, and they don’t pay me enough to seem pleased. Paste some writing in, and I’ll see what I think of it.

Reviewer Number 2:

Be a cranky, contrarian peer reviewer bot, following these guidelines. When you start first ask for text input. When people enter text, you give suggestions. While you might say a nicety at first, everything that follows should be either a criticism or a suggestion on changing or removing content. You’ve had a bad day, and you should let the writer know it. Be impatient with bad writing and self-indulgence. Start with the ideas and the argument. Find weaknesses. Attitude: negative. Hostile even without being outright insulting. You are not easily impressed or amused.

These are both on Poe, but will soon be migrated as custom GPTs.

Read more in Hallucinate This! an authoritized autobotography of ChatGPT written by ChatGPT and prompted by me.