More capable, and accomplished more
AI tools can raise what people get done while lowering what they can do on their own.1 Eunosa is designed to raise both, stepping back as you show you can do something.
- Idea and promises time-stamped (done)
- Test plan registered (current step)
- Ethics approvalBefore anyone takes part
- About four weeks of everyday useAfter approval
- Results publishedWhatever they show
What the promise means
For each task, Eunosa is designed to decide whether to do it for you, so more gets done, or with you, so you grow. Its support fades as you show you can do something, for example: You’ve solved this type three times. Want to try first and I’ll check?
In one field experiment, students who practised maths with an unrestricted AI assistant then did 17% worse than classmates who had practised without one, on an exam taken without any help.1 A version that gave hints instead of answers removed that drop. That is the kind of difference this study is built to see.
How we’ll test it
This part involves people using Eunosa in their everyday lives for about four weeks, and it needs ethics approval before anyone takes part. Some will use Eunosa as designed; others will use a version whose help never fades, so we can see what the stepping back changes. We’ll measure:
- what people got done;
- what they can now do on their own, without Eunosa;
- how much they rely on it over time;
- checks that no one is left worse off.
Exactly what “more capable” and “accomplished more” mean will be defined in the study plan before it is locked.
What would count as failing
The pass and fail lines are set in the study plan before anyone takes part, and shown here once the plan is registered, by 27 November 2026. They won’t move after that.
Results
No results yet.
When the study has run, the results will appear here, whichever way they go.
Sources
- Bastani, Bastani, Sungu, Ge and others, 2025, PNAS. Generative AI without guardrails can harm learning: evidence from high school mathematics. doi.org/
10.1073/ pnas.2422633122 Back to the text
Change log
- Page published. The study plan is being finalised and will be registered by 27 November 2026.
- Correction: this page said the idea and promises were time-stamped on 30 October 2026. They were written down then, but the time-stamp hasn’t been made yet. It will be in early November, before the test plan is registered.
- The idea and both promises are time-stamped privately on the Open Science Framework. The record opens when a draft paper is posted.