The AI Critic Tank

Overview

When students use AI on a project, it too often writes the idea and the student decorates it. This session flips the roles. Teams of teachers, working as students, choose a problem in the fictional town of Cedar Bend, understand it, and design a solution entirely on their own. Only then do they ask an approved chatbot to act as a skeptical critic. They log every critique, accept, reject, or modify each one with a reason, revise, write an honest disclosure of how AI was used, and pitch to a "Tank" panel. The debrief turns the structure into a problem-solving project teachers can run with students.

Objectives

• Participants will work through a problem-solving process (understand the problem, identify causes and stakeholders, design, test, revise) before using any AI tool.
• Participants will use AI as a critic rather than an author, and justify accepting, rejecting, or modifying each AI critique.
• Participants will write a clear AI-use disclosure statement and explain why transparency matters for academic integrity.
• Participants will plan a student project that uses the same human-first, AI-as-critic structure and names the evidence of student problem solving.

Step by step

1. 10–5 min

   Hook

   Author or critic?

   Ask: "If a student uses AI to write the project idea and then does all the slides, who solved the problem?" Take two answers. Then say: "Today AI gets one job: critic. You're the authors. It can question you, but it can't write for you. And at the end, you'll tell the panel exactly what it did."

   Facilitator noteSet the rule clearly: no AI until the Apply step. Some teams will want to jump ahead. The waiting is part of the lesson.

2. 25–20 min

   Explore

   Understand the problem

   Each team picks a problem card. On chart paper, they map it: Who is affected? (at least four stakeholders, including someone easy to overlook). What are the likely causes? (ask "why?" at least three times). What do we not know yet, and who would we ask? Teams end by writing a one-sentence problem statement: "___ need a way to ___ because ___."

   Facilitator noteThe facilitator key on each card lists stakeholders teams often miss. Offer one as a hint only if a team is stuck.

3. 320–40 min

   Create

   Design it by hand

   Teams brainstorm at least five possible solutions, choose one using two criteria they set themselves (for example, cost and who benefits), and sketch it on chart paper: what it is, how it works, who does what, and how they'll know it worked. No devices. The facilitator circulates and asks, "What would the person most affected by this problem say about your idea?"

   Facilitator noteThis is the evidence of problem solving: the brainstorm list, the criteria, and the reasoning behind the choice. Have teams photograph or keep the chart; it becomes the baseline for comparing their revision.

4. 440–55 min

   Apply

   Call in the critic

   Project and model the critic prompt: "You are a skeptical review panel for a community project. Here is our problem statement and our solution. Do not propose your own solution or rewrite ours. Ask us the five hardest questions a panel would ask, point out the three biggest weaknesses or risks, and name one group of people we may have overlooked." Teams type their problem statement and solution (no real names or personal information) and log each critique on Handout B: accept, reject, or modify, with a reason. If the chatbot starts writing solutions, teams redirect it and note that in the log.

   Facilitator noteLook for teams that reject a critique with a good reason, for example "it assumes families have cars; our stakeholder map shows many don't." Rejecting an AI critique thoughtfully is strong evidence of human judgment.

5. 555–65 min

   Create

   Revise and disclose

   Teams revise the solution on the chart in a second color so changes are visible. Then they write an AI disclosure statement at the bottom of Handout B, for example: "We designed this solution ourselves. We used a district-approved AI chatbot as a critic: it asked us five questions and named three weaknesses. We accepted two critiques (added a Spanish-language sign-up and a rainy-day plan), rejected one (it assumed families drive), and modified one. The AI did not write any part of our solution or pitch."

   Facilitator noteMention that major style guides, such as MLA and APA, have published guidance on citing generative AI; students should follow whichever your school uses. A disclosure statement explains the role AI played, which a citation alone may not.

6. 665–77 min

   Practice

   The Tank

   Each team gets 2 minutes to pitch and 1 minute for panel questions. Pitches must include the problem statement, the solution, one change made because of the AI critic and one critique they rejected, and the disclosure statement. The panel scores with Handout C. For more than six teams, run two panels at once.

   Facilitator notePanelists should ask at least one question about the rejected critique. That is where teams show their reasoning most clearly.

7. 777–85 min

   Debrief

   As teachers: where was the thinking?

   Step out of role. Ask: "Where in the last hour did the real problem solving happen? What did the AI add that you wouldn't have gotten from a peer team? What did it get wrong? How would your students' work change if AI were the critic instead of the author?" Chart two columns: "Evidence of student thinking" and "Where AI helped or got in the way."

   Facilitator noteHonest answers help here. Some teams will say a peer critic would have been just as good. That's a legitimate finding: use AI only where it adds something.

8. 885–90 min

   Transfer

   Plan the student version

   Each person writes a project plan for their students on the back of Handout C: the problem (from the school or community), when AI enters and in what role, what students log, and how they'll disclose. They name the one piece of evidence they'll grade most heavily. Suggestion: the critique log.

   Facilitator noteFor students under 13, the teacher runs the critic prompt on a projector with the class, and the class decides together which critiques to accept. No student accounts are needed.

Adaptations

Grades 6–8 teachers

Use school-based problems (the lunch line, lost-and-found, hallway crowding) and have the teacher run the critic prompt on the projector for each team. The class votes on which critiques to accept, with reasons.

Higher Ed faculty

Use the structure for a capstone or design project and require a methods note describing the AI critique process, with the log as an appendix, alongside the citation style your discipline uses.

Librarians

Add a research step between Explore and Create in which teams find one reliable local source (city data, a news story, an interview) about their problem. Teams often find it changes their solution more than the AI critic does.

Shorter session (60 minutes)

Give each team a pre-written solution to critique instead of designing one, and skip to the Apply step. You lose some of the authorship, so note that in the debrief.

Pairs well with

• Generative AI Adoption ChecklistCheck your chatbot and classroom routines before running this with students.
• AI Custom Instructions LibraryIdeas for setting up a chatbot to act as a critic rather than an author.
• Augmented K–16 AI Learning FrameworkWhere human-AI collaboration fits in a K–16 progression.