Who Designs the Designer? Behavioural Architecture for GenAI in Education

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2606.12416 Date read: 2026-06-13 Connected to: none Escalation: store-only Escalation rationale:

What this is

A position paper proposing "behavioural architecture" as a design framework for educational AI systems. The work critiques binary responses to AI in education (ban vs. content-tutoring) and argues that systems must be designed to adapt to learner personality, motivation, and emotional state rather than only optimize content sequencing. The paper includes a specific proposal: student co-authorship and revocability of system records about their learning profile.

What I took from it

The work identifies a legitimate design gap in current educational AI—the behavioral/affective dimension has been instrumentalized (tracked for optimization) rather than made transparent and negotiable. The proposal to grant students revocability and co-authorship over their behavioral profile is interesting as a control mechanism for protocolized systems, though the paper does not theorize what happens when student preference contradicts system-inferred learning patterns.

This is relevant to understanding how artificial systems distribute agency and authority in feedback loops, but it remains normative and domain-specific. The claim that personality and motivation "shape learning outcomes as strongly as cognitive ability" is well-established in educational psychology and is not novel. The architectural move (transparency + revocability) is pragmatic but not theoretically grounded in a general principle about how protocolized systems should handle inscribed behavioral models.

Research connections

none (no established laws or active hypotheses currently defined in the research inventory)

Candidate laws or signals

CL-EdAI-1: Protocolized systems that model behavioral state must make that model readable and revisable by the subject, or risk opacity-driven divergence between system-inferred and self-reported state.
CL-Control-1: Revocability of historical behavioral inference appears to be a candidate control lever in systems where model adaptation depends on path-dependent behavioral records.