An ethical-AI-literacy program I designed, built, and ran as a feasibility pilot. I scoped it to a real cohort, built the assessment tool myself, and used it to watch people decide — in the moment — whether their own AI use was responsible.
Institutions reacted to generative AI with bans and fear, not instruction — leaving a generation improvising, with real stakes for their learning, integrity, and early careers.
We were treating AI like a cheating tool instead of teaching people to use it responsibly. The problem wasn't enforcement — it was literacy.
The cohort were adults 18+ already using AI — students and early-career professionals. The goal was never to restrict access. It was to build judgment.
They reach for AI daily, but lack any framework for when it helps learning versus when it replaces it.
They're anxious about an invisible integrity line, and want to defend their choices to a professor or employer.
They want proof of responsible AI competency they can carry into coursework and job applications.
A capstone semester is unpredictable, so I designed for resilience: a full framework, a tool I could ship, and a pilot small enough to actually run and measure.
Reframed the problem from "prevent AI" to "build judgment," which set every later decision.
Built the assessment as a working flashcard tool, so the literacy check was hands-on, not a paper quiz.
Cut an 8-week framework to a 3-session feasibility pilot with a 4-person cohort, adapting methods to match.
Ran the cohort, captured the data, and wrote it up as a 19-page feasibility study.
I assessed literacy with a flashcard tool I designed and built — GhostSquirrelNinja. Participants worked through the AI-literacy material hands-on: tap a card, reveal, self-check. But the deeper choice was built into the tool itself — it's privacy-first by design, so the user decides what's kept.
Why a built tool, not a worksheet? A flashcard format makes recall active — tap to reveal, self-check, move on. I loaded the literacy material as decks so participants practiced their judgment hands-on instead of reading about it.
The design models the message. A program about responsible, transparent tech use shouldn't measure people with a tool that quietly hoards their data. So I built for data sovereignty: guest mode, on-device storage, export anytime — or close out and it's gone. The instrument lives the ethic the curriculum teaches.
I designed a full eight-module framework, then made the call to test a condensed version with a real cohort — because a working three-session pilot teaches more than an unrun eight-week plan.
The eight-module framework I designed (condensed for the pilot)
The interesting part of any project is what I decided — and what I deliberately gave up to get there.
Teach, don't ban. Rejecting the enforcement framing made the program something users wanted, not something done to them.
Run small, learn real. Cutting to a 3-session, 4-person feasibility pilot meant I could actually finish, gather data, and learn — instead of shipping an unrun plan. The design degraded gracefully.
The tool lives the lesson. I built the flashcard instrument to respect data sovereignty — keep it or wipe it, your call. Measuring responsible AI use with a tool that itself respects the user kept the whole project honest.
With a small cohort, the goal wasn't statistical proof — it was evidence about whether the format, the tool, and the methods were viable enough to build on.
Can a short, remote, hands-on program actually move participants toward more responsible AI decisions?
Does a flashcard instrument surface real judgment — and hold up when actual users touch it?
What did the cohort reveal about pacing, scope, and where the design needs to evolve before scaling?