The science of how humans retain information has been well understood for over a century. Ebbinghaus's forgetting curve, Leitner's card system, the spacing effect — these are robust, replicable findings. The problem was never the theory. It was the tooling.
Flashcard apps got the spacing cadence right. What they missed is everything else: *why* something is hard for a specific person, whether the explanation is the right format for how they think, and whether the prompt is testing recognition or genuine recall.
Where AI changes the equation
Modern language models can generate alternative explanations on the fly — if a concept isn't sticking, reframe it. They can detect from response patterns (latency, partial answers, confident wrong answers) what type of confusion is happening. And they can vary question format — definition, application, counterexample — so that memorization doesn't masquerade as understanding.
This isn't hypothetical. Systems running adaptive LLM-in-the-loop review are already showing measurable gains in retention benchmarks over static spaced repetition alone, particularly for conceptually dense material like legal frameworks, medical pathways, and financial modeling.
The equity angle
The education gap isn't primarily a content problem — Khan Academy solved content a decade ago. It's an *engagement* and *calibration* problem. Students who can afford tutors get real-time feedback and re-explanation. Students who can't get a static video they may or may not rewatch.
Adaptive AI closes this gap more directly than any previous technology because it doesn't require a human teacher to scale. The same calibration loop that works for a student in Austin works for one in Accra, at the same cost per session.
Seggie is building in this direction: not just smarter scheduling, but a system that models what you know, how you know it, and where the seams are — and uses that model to decide what to surface next.