AI Tutoring and Adaptive Learning: Generative Tutors, Scaffolds, and Evaluation
Quick Answer
This topic gathers richards.ai's research on generative AI tutors and personalized adaptive learning systems — the lineage from intelligent tutoring systems, the emerging RCT evidence base, and the pedagogical-safety design patterns that distinguish a tutor from a solver. The page offers reading paths for learning-systems engineers, instructional designers, and learning leaders weighing where AI tutoring belongs in real deployments.
AI Tutoring and Adaptive Learning: Generative Tutors, Scaffolds, and Evaluation
This topic clusters richards.ai's research on generative AI tutors and personalized adaptive learning systems around a single question: how to build AI-mediated instruction that produces durable learning rather than just immediate task completion. The originating source is the Generative AI Tutors and Personalized Adaptive Learning Systems paper; the surrounding artifacts translate its architecture, evidence, and design constraints into working frames for engineers, designers, and learning leaders.
What this topic covers
The cluster connects three threads: the pre-LLM lineage of intelligent tutoring systems and what it got right about decomposition; the new and uneven RCT evidence base on generative AI tutors, including both gains and learning-loss results; and the pedagogical-safety patterns — help fading, hint ladders, retrieval grounding, evaluation hierarchies — that distinguish a tutor from a solver. Mechanism, findings, and definitions live in the member artifacts. The hub stays editorial.
How to read this page
The practitioner who wants the design frame should start with What Is a Generative AI Tutor?, then read the productive-struggle and solver–tutor-gap glossary entries, then work through the tutor design checklist. Engineers building an adaptive system should start with the paper, then the checklist, then the ITS glossary entry to ground the reference architecture. Learning leaders making a deployment call should start with the executive brief, then the paper, then the explainer.
Where this topic sits
This is the inaugural learning-pillar hub on richards.ai and the entry point into the broader catalog of learning artifacts. It runs parallel in structure to the security-pillar agentic AI security hub: an editorial frame around a coherent body of work, with the substantive arguments held in the member artifacts. Some checklist and brief detail is held to a yellow risk level pending wider review of deployment guidance.
Papers
1 memberLearn
1 memberGlossary
3 membersIntelligent Tutoring System (ITS)
An intelligent tutoring system (ITS) is a computer-based instructional system that adapts to an individual learner by maintaining an…
NoteThe pre-LLM lineage. Defines the domain-model / learner-model / pedagogical-policy decomposition the paper builds on.
Productive Struggle
Productive struggle is the band of cognitive effort where a learner is challenged enough to build durable schema but not so overwhelmed…
NoteThe cognitive-load construct that explains why help fading matters and why over-helpful tutors can suppress learning.
Solver–Tutor Gap
The solver–tutor gap is the empirically observed divergence between a language model's ability to solve a domain problem and its ability…
NoteNames the central failure mode: a model that solves problems is not automatically a tutor. Frames evaluation benchmarks.