Lesson 2.4: Evaluating AI Feedback and Ethics in Theater
Estimated time: 75 minutes. Competencies: AI feedback evaluation, AI ethics literacy.
Video: Lesson 2.4: Evaluating AI Feedback and Ethics in Theater. Four-minute introduction: three classes of AI feedback, prompting for better notes, and the five ethical questions of staging AI-assisted work. (Instructor: build with make_videos.py or record and embed your own.)
Theatre Lesson Four. Recorded by the instructor for the 2025 pilot of this program.
Learning objectives
By the end of this lesson you will be able to:
Elicit feedback on a dramatic text from an AI model using prompts that specify the lens and the criteria.
Classify each piece of AI feedback as useful, generic, or wrong, and justify the classification with reference to the text.
Explain the ethical questions raised by staging AI-generated or AI-assisted theater: authorship, consent, representation, labor, and disclosure.
Take and defend a position on whether and how you would stage AI-assisted work with students.
Warm-up: the note you ignored (5 minutes)
Recall a piece of feedback on your creative work that you rejected and were right to reject. What made you sure? Usually it is one of three things: the note misread the work, the note was so general it could apply to anything, or the note asked for a different piece than the one you were making. You will use the same three tests on AI feedback today.
Key ideas
AI feedback is fluent and unaccountable
A text model will always produce feedback, instantly and confidently, on anything you paste. It has no stake in your scene, no memory of your rehearsal room, and no way to know your students. Its feedback is a statistical average of what feedback usually looks like. That average can be genuinely useful for spotting structural problems and clichés, and it is useless or harmful when it flattens a distinctive voice toward the mean.
Three classes of AI feedback
Class
Signs
What to do
Useful
Points to a specific line or moment; names a problem you can verify by rereading; suggests a change consistent with your intent
Consider it seriously; test it in a rewrite
Generic
Could be said about any scene ("raise the stakes," "develop the characters more," "show, don't tell") without pointing anywhere
Ask the model to locate it in the text; if it cannot, discard
Wrong
Misreads the text, invents a detail, misidentifies the genre, or pushes the scene toward a different one
Discard, and note what the misreading tells you about the model's assumptions
Prompting for better feedback
Feedback quality depends on the request. Specify the lens ("read as a dramaturg preparing notes for a first rehearsal"), the criteria ("comment only on objective, obstacle, and subtext"), and the form ("five notes, each quoting the line it refers to"). Then ask a follow-up: "Which of these notes are you least confident about, and why?" Models are often more honest about uncertainty when asked directly.
Ethics of staging AI-assisted theater
Putting AI-assisted work in front of an audience raises five questions:
Authorship. Who is credited as playwright when a model drafted 40% of the lines? Programs and posters need honest credits.
Consent. Did the writers whose plays trained the model agree? Did your students agree to have their improvisations fed into a tool?
Representation. Models reproduce the biases of their training data. AI-generated characters from underrepresented communities can default to stereotype. Who checks?
Labor. Theater employs playwrights, composers, designers. Which tasks are you automating, and who would have done them?
Disclosure. Does the audience know? Should they? What does the school's policy say?
There is no single right answer. There is a wrong one: not asking.
Activity: evaluate the feedback, take a position (55 minutes)
Get feedback. Paste your Lesson 2.1 scene (or another short scene you wrote) into a text model with a lens, criteria, and form specified as above. Ask for five notes, each quoting the line it refers to.
Ask for uncertainty. Follow up with the confidence question. Record the answer.
Classify. For each of the five notes, mark it useful, generic, or wrong, and write one sentence citing the text as evidence.
Test one useful note. Rewrite the relevant lines. Keep both versions.
Take a position. In 200 to 300 words, answer: would you stage an AI-assisted scene with students? Under what conditions? Address at least three of the five ethical questions and state the credit line you would print in the program.
Save everything as a PDF named Lesson-2-4_LastName_Evaluation.pdf.
Create and share
Post in the Lesson 2.4 Forum:
Your classification table with evidence.
The before and after of the one note you tested.
Your position statement and program credit line.
A rationale of no more than 75 words: what surprised you most about the AI's feedback.
Your one-line AI credit.
Reply to two peers who reached a different position. Identify the ethical question that separates you and say whether their reasoning moved you.
Resources
Any text model.
Your school or district's AI policy, if one exists.
UNESCO (2024) AI competency framework for teachers, dimensions 1 and 2: human-centred mindset and ethics of AI.
Reflection prompts
What proportion of the feedback was useful? Would you use the tool again for this purpose?
Which ethical question do you feel least equipped to answer, and where would you go for help?
Looking back across both modules, how has your definition of "creative" changed, if at all?