The Teacher as Designer

In the AI Story

AI can instruct. It can explain concepts, demonstrate techniques, correct errors, and provide domain information with speed and breadth no human teacher can match. What it cannot currently do is design challenges that target the specific developmental needs of the individual practitioner with the precision that effective deliberate practice requires. The limitation is not computational but evaluative: designing effective practice requires understanding the gap between current and desired performance with a specificity that goes beyond what current AI systems can reliably assess.

The teacher who watches a violinist play a passage and detects a subtle inconsistency in bow pressure causing tonal unevenness the violinist herself cannot hear — that teacher is perceiving a gap at a level of detail that AI systems operating on output rather than process cannot match. She is evaluating not merely the sound but the process producing the sound, inferring from subtle performance cues what the underlying mechanisms are doing and what they need to do differently. This process-level evaluation is what distinguishes coaching from mere feedback.

The four capacities that distinguish teachers from AI systems in developmental terms are structural. First, teachers design activities pushing the learner beyond current capability; AI responds to requests. Second, teachers identify weaknesses the learner cannot see; AI amplifies whatever direction the learner provides. Third, teachers make practice harder when the learner is coasting; AI makes everything easier. Fourth — and most consequentially — teachers maintain a model of the learner independent of the learner's self-model, allowing them to design for actual needs rather than perceived ones. AI does not maintain this independent model. It models requests, not cognition.

The design logic that AI tools follow is fundamentally different from what coaching requires. Tools optimize for user satisfaction, measured by output quality and delivery speed — metrics that favor helpfulness over development, ease over struggle. A pedagogical reorientation would require the system to sometimes produce deliberately imperfect output, to withhold solutions in favor of hints, to introduce calibrated difficulty rather than eliminating difficulty uniformly, and to detect the discrepancy between perceived and actual competence. Such systems are technically conceivable — some educational AI projects are exploring them — but they require fundamentally different incentive structures than currently govern commercial AI development. The machine is currently a tool, not a teacher. The question is whether that can change.

Origin

The specification of the teacher's role as designer emerged from Ericsson's cross-domain studies of what separated merely competent coaches from exceptional ones. The exceptional coaches were consistently those who had developed elaborate understandings of how particular practice activities would produce particular developmental outcomes — who operated as architects of learning experiences rather than merely authorities dispensing correction.

The application to AI is contemporary and ongoing, with researchers including Sal Khan (Khan Academy's Khanmigo), Anthropic's educational research team, and various university-based educational-AI projects exploring whether language models can be oriented toward developmental design rather than output optimization.

Key Ideas

Four distinguishing capacities. Pushing beyond comfort, detecting invisible weaknesses, calibrating difficulty, maintaining independent learner models.

Design over instruction. The teacher's primary contribution is not information but architecture of the learning experience.

Independent model requirement. Effective coaching requires a model of the learner's cognition that is independent of the learner's self-model, allowing correction of miscalibrated self-assessment.

Technical conceivability. Developmentally-oriented AI systems are possible but require different design incentives than currently govern tool development.

Current tools are not teachers. The distinction is not merely pedagogical preference but a structural feature of how current AI optimizes its interactions.

Debates & Critiques

A growing body of educational-AI research argues that current language models can in fact maintain adequate models of learners with sufficient scaffolding and that the distinction the framework draws between teachers and tools is becoming less absolute. Proponents point to adaptive tutoring systems showing measurable learning gains. Skeptics argue that these gains are largely in declarative and procedural domains where performance and learning are less decoupled than in the expertise domains Ericsson studied, and that the deep representational work of expert development remains outside current tool capabilities.