AI Critique Exercises That Actually Teach Critical Thinking
Tried something different this term that’s working well enough to share.
Instead of banning AI or ignoring it: I give students an AI-generated response to a question and ask them to critique it. Find the errors. Find the unsupported claims. Find where the reasoning is weak. Improve it.
Result: students engage more critically with the AI output than they would with a textbook passage, because they’ve been conditioned to trust AI. When it’s wrong, they’re genuinely surprised. That surprise is the teaching moment.
Two additional benefits: students learn to use AI better (by understanding its failure modes), and I can see their critical thinking in action through the critique rather than trying to detect if their final essay was human.
this works better than detection. it also teaches something useful.
5 Replies
Join the discussion.
Log In to Reply
This approach aligns with what curriculum research consistently shows about metacognitive skill development. when students must evaluate and critique rather than produce, they engage with the underlying concepts more deeply. The fact that AI is the subject of critique also builds AI literacy, which our provincial curriculum explicitly identifies as a learning outcome. Worth documenting as a formal pedagogical approach.
can you share more about how you set this up? which AI are you using to generate the response, and how do you make sure the errors are meaningful rather than just obvious? want to try this with my science class next term.
saving this thread. haven't had time to design AI-specific lessons and these examples are practical enough to adapt directly. the bias audit idea especially.
the "find the AI error" exercise is one I've used with my grade 10s and the results were genuinely surprising. students who couldn't articulate why an essay was weak suddenly had the vocabulary for it when AI produced a clearly plausible-but-wrong version. the errors made the standards visible in a way direct instruction didn't.
tried this based on a similar idea I read about last month. worked really well for chemistry. generated a GPT response to a mechanism question with a subtle error in the electron movement - students caught it in 10 minutes. they were so much more engaged than with a textbook problem.