Fletcher Dubin
Co-author · college sophomore
One half of the pair. The people this paper studies are his classmates, his group-project partners, and the friends who text him at 2am about something an AI told them.
About the project
Two sophomores.
One question.
Fletcher and Maxwell Dubin are twins, college sophomores, and the authors of a research paper on how their generation actually uses AI, and how it doesn't. Most of what's written about students and AI is written by people who aren't students. This project is the other way around. The test on this site is how they're asking.
The researchers
Maxwell Dubin
Co-author · college sophomore
The other half. Twins, same year, two sets of eyes on the same generation: close enough to the subject to ask honest questions, and to get honest answers back.
They're writing the paper to better understand how their peers are using AI, and how many aren't. If you're a student who took the test, you're already part of it. Thank you.
The question
Not whether.
How.
Almost every conversation about students and AI collapses into one question: are they cheating? It's the least interesting question available, and it produces the least honest answers. Nobody tells the truth to a survey that's deciding whether they're a cheater.
The project asks something different. Not whether you use AI, or how much, but how: whether you lead it or follow it, trust it or check it, use it to learn or to finish, let it into your work, your messages, your feelings, or keep it at arm's length. Those are relationships, and relationships have patterns. The patterns are what the paper is about.
The nine dimensions the test measures
- AgencyWho starts the work: you, or the model?
- VerificationDo you check it, or trust it?
- MotiveAre you there to understand, or to be done?
- ScopeOne domain, or everything?
- FluencyCan you actually make it perform?
- IntegrationA tool you pick up, or part of how you think?
- VulnerabilityDoes it get the polished you, or the real one?
- HustleAre you building something with it?
- SociabilityIs your AI life private, or something you talk about?
The instrument
A test people actually finish.
A forty-question survey about AI habits gets abandoned halfway and answered politely until then. So the instrument is built like something you'd send a friend: forty-one swipes, about three minutes, a result worth posting. The fun is the method. It's what gets honest answers at the scale a paper needs.
Three kinds of card
Pick-one scenarios reveal the core relationship. Yep-or-nope statements catch behavior: what you did this week, what you'd never tell a professor. Three prompt-judgment rounds have a right answer and measure whether you can make AI perform, separately from how you feel about it.
Swipes, not scales
A swipe takes about three seconds and gives the rationalizing part of the brain less time to tidy the answer. Response times are logged too; hesitation is data.
Forty archetypes
The result is a vocabulary, not a verdict: sixteen core relationships, eight pattern types, twelve behavioral specialties, four outliers. Each comes with a shadow side and one thing to try. Nobody is "bad at AI." Everybody is somewhere.
Honest margins
The three-minute version is calibrated to roughly ±6 points on a hundred-point index. A hundred-question version, in development, narrows that and adds peer percentiles. The archetype labels, rarity figures, and thresholds are being recalibrated against real responses as they come in.
Your data
Anonymous by design.
The project needs patterns, not people. The test is built so that it can't identify you even if it wanted to.
What the test records
- A random ID that lives in your browser (so a retake on the same phone isn't counted twice)
- Each answer and how long it took
- Your result: score, archetype, badges, dimension scores
- The optional context taps: age range, year, field, which AI you use
- A group tag, if you came in through a class or club link
- An email address only if you join the waitlist for the full test, kept separately from your answers
What it doesn't
- Your name, login, or account of any kind
- Your location, contacts, or anything else on your device
- Which school you attend, unless you tell us in a group tag
- Anything that trains an AI model. Your answers inform the paper and the next version of the test. That's all.
Results are reported in aggregate only: how many people land where, how dimensions relate, how groups differ. No individual result is ever published or shared. You can stop the test at any point, and nothing is recorded as a result until you reach the end.
Take part
Three ways to help.
1
Take it
Three minutes. Answer like nobody's watching, because nobody is.
Start swiping →
2
Share your result
3
Bring your class or club
Every share link greets the next person with your archetype and asks what they are. That's how the sample grows.
The share button is at the end of the testA tagged link files every result under your group. If you teach, advise, or run something on campus and want your people in the study, get in touch.
Email the project →