Teach AI new ways to approach a problem.
KeepRI plans to work directly with AI companies to develop reasoning games and environments around the capabilities they want to improve. These environments would capture how people form hypotheses, revise assumptions, and make conceptual leaps.
Competitive gameplay is designed to bring participants back, allowing us to observe how their reasoning develops across attempts. Follow-up challenges would test whether their insights transfer to unfamiliar problems.
AI partners would receive structured human discovery data and executable environments tailored to their training and evaluation needs.
Explore the proposed deliverablesDiscover the rule.
Test your thinking.
Explore unfamiliar problems, experiment with different approaches, and learn from the results.
KeepRI plans to combine reasoning games, feedback, and competition to give people a place to exercise independent judgment. Take on the next challenge to test whether an insight carries over.
Challenges would follow players across attempts. Progression and competition are intended to make this an experience people want to return to.
A research model built around participant choice.
Free practice would remain available without research participation.
Under the planned model, entry into cash-prize research competitions would require explicit agreement to data collection and commercial use.
We intend for research partnerships and licensing revenue to help fund new games, free access, and competitions. Identity and prize-payment information would stay separate from research deliveries.
Research enrollment and cash-prize events are in development.
Our philosophies.
Independent human thinkingIndependent judgment
Independent judgment is essential to a society shaped by AI.
People must be able to evaluate evidence, question recommendations and take responsibility for decisions. KeepRI plans to create challenges in which people form, test and revise ideas for themselves.
Discovery and transfer
A useful insight should be tested beyond the problem that revealed it.
Our planned challenges would ask players to form, test and revise ideas. Follow-up tasks would examine whether those ideas transfer. Decisions and stated hypotheses provide evidence, not direct access to thought or a measure of broad intelligence.
Participant choice
Free practice would remain available without research participation.
Under the planned model, entry into cash-prize research competitions would require explicit agreement to data collection and commercial use. Research enrollment and cash-prize events are in development.
“The limits of my language mean the limits of my world.”Ludwig Wittgenstein
Tractatus Logico-Philosophicus, 5.6 ↗
For KeepRI, this is an inspiration to explore how discovering a new concept can expand the problems we are able to solve. We want to create experiences that exercise independent human thinking while helping researchers investigate how AI might learn from human discovery.
Built with AI teams.
We plan to co-design games with AI companies and license consented discovery data, runnable tasks, and separate tests.
- Discovery sequences
- Executable environments
- Evaluation packages
From attempts to insight.
- Observe
- Propose
- Test
- Revise
- Transfer
Record what players see, their stated hypotheses, actions and feedback. Follow-up tasks would test whether an insight transfers to an unfamiliar problem.
Does discovery help AI?
Can discovery sequences help models solve new problems more creatively and efficiently than final answers or synthetic examples alone?
Tests would measure hypothesis formation, creative problem-solving and conceptual transfer. Model improvements remain unproven.