Mastery tracking.
Without understanding mastery.
AI tutors track correct answers, hint usage, and completion rates. They miss the 45-second hesitation that separates a student who is thinking carefully from one who has stopped understanding. VanLehn showed this gap in 2011. It remains accurate today.
AI tutors are good. Human tutors are better. This is why.
VanLehn's 2011 meta-analysis found that intelligent tutoring systems outperform conventional instruction but remain less effective than human tutors. The effect size gap is approximately 0.4 standard deviations. That gap has not closed significantly since the paper was published.
The reason is not processing power or curriculum quality. Human tutors read cognitive and emotional states in real time - they can tell when a student is confused but too embarrassed to ask, when they are bored and need a harder problem, when they are anxious and need encouragement before content. AI tutors cannot. They see the answer, not the state that produced it.
ABSD proposes a framework for closing this gap - not by adding sensors or cameras, but by interpreting the interaction data that AI tutoring systems already collect.
EdTech systems reach Layer 2. Layer 3 is the frontier.
EdTech is ahead of most CX platforms on the ABSD ladder. Mastery models track behavioural patterns - attempts, hint usage, time-on-task - and use them to calibrate difficulty. That is genuine Layer 2 capability. But the cognitive-emotional state beneath the behaviour is invisible. The 45-second pause, the repeated failure, the avoidance of worked examples - all logged but not interpreted as psychological signals.
Based on VanLehn (2011), Li et al. (2020), Paquette et al. (2014). Indicative of sector average.
Seven signals in an EdTech context.
If EdTech AI reached Layer 3 - what changes.
The mastery model stops confusing performance with understanding. A student who gets the right answer after 45 seconds of hesitation and one who answers in 4 seconds both show mastery at Layer 1. At Layer 3, the hesitation is a signal - the first student understood but needed cognitive processing time, the second answered reflexively. One is consolidating, one is recalling. They need different next steps.
Anxiety becomes visible before it becomes failure. Avoidance patterns are the earliest detectable signal of learning anxiety. A student who consistently skips worked examples is showing anxiety at Layer 2. At Layer 3, that anxiety is named. The intervention changes from "you haven\'t completed this section" to "here is a lower-stakes way in."
Flow becomes the optimisation target, not mastery. Pekrun et al. (2017) confirmed that sustained engagement produces better long-term outcomes than maximised short-term performance. A system that detects flow and adjusts challenge to maintain it - rather than maximising difficulty at every step - produces better learners over a semester, not just better test scores this week.
Competence drive shapes how the system scaffolds, not just what it teaches. Two students both reach a frustration state. One persists and one disengages. The drive beneath the state determines the correct intervention. SDT provides the theoretical foundation. ABSD provides the architecture to apply it in production.