Working paper Working paper, draft 1
Aporia
Why learning starts where your reasoning breaks
- Author
- Cogniloop
- Date
- November 2026
- Length
- 28 pages

Abstract
Most people believe AI is bad for learning. They are right about how it is mostly used. A tool that hands over the answer in seconds meets a student who would rather not spend hours, and the student takes the answer. Studying feels like understanding, and the exam shows that it was not. A large field experiment measured exactly this: students who practised maths with plain GPT-4 scored 48 per cent higher in practice and 17 per cent lower on the exam that followed.
I think the problem is how AI is applied, not AI itself. Learning starts at the moment your own reasoning breaks: the impasse that Socrates' questions led to, which the Greeks called aporia. A tool that removes that moment removes the learning with it. A tool that finds it, every time, could make study shorter and make it hold.
Cogniloop is built on that idea. You try first. Under the tutor, an engine reads your working line by line and finds where it stopped holding. It keeps what you can do alone apart from what you did with help, schedules each topic to come back before it is likely to fade, and remembers you across weeks. It also checks its own predictions about you, so it finds out when it is wrong.
Its first self-check did find it wrong. On my own study, it expected me to solve two-thirds of my problems alone, and I solved one-third. It had started new topics too high, and it had counted help as progress: the same illusion it was built to break. Both were corrected.
Nothing in this paper shows that Cogniloop improves learning. It sets out the test that would: whether learners who use it can solve more on their own a week and a month later than learners who use a chatbot or study the usual way. If they cannot, the approach is wrong, and I will say so.
About this draft
Draft 1. The design, the reasoning and the engine's records are real. Real data comes from one learner, the founder, and every projection and illustration is labelled as one.
Contents
- 01Executive summaryDraft
- 02IntroductionDraft
- 03Aporia: where learning startsDraft
- 04What AI changesDraft
- 05An engine under the tutorDraft
- 06Reading the work, line by lineDraft
- 07What the engine knows about youDraft
- 08Remembering youDraft
- 09Deciding what comes nextDraft
- 10The engine checks itselfDraft
- 11What would prove this wrongDraft
- 12Who it is forDraft
- 13LimitationsDraft
- 14ConclusionDraft
Cite
@techreport{cogniloop2026aporia,
title = {Aporia: Why learning starts where your reasoning breaks},
author = {Cogniloop},
year = {2026},
month = nov,
type = {Working paper},
institution = {Cogniloop},
note = {Working paper, draft 1},
url = {https://cogniloop.app/paper}
}