Date
Publisher
arXiv
The integration of Large Language Models into Intelligent Tutoring Systems pre-sents significant challenges in aligning with diverse and often conflicting values from students, parents, teachers, and institutions. Existing architectures lack for-mal mechanisms for negotiating these multi-stakeholder tensions, creating risks in accountability and bias. This paper introduces the Advisory Governance Layer (AGL), a non-intrusive, multi-agent framework designed to enable distributed stakeholder participation in AI governance. The AGL employs specialized agents representing stakeholder groups to evaluate pedagogical actions against their spe-cific policies in a privacy-preserving manner, anticipating future advances in per-sonal assistant technology that will enhance stakeholder value expression. Through a novel policy taxonomy and conflict-resolution protocols, the frame-work provides structured, auditable governance advice to the ITS without altering its core pedagogical decision-making. This work contributes a reference architec-ture and technical specifications for aligning educational AI with multi-stakeholder values, bridging the gap between high-level ethical principles and practical implementation.
Study at a glance
- Question: How can an advisory multi-agent governance layer translate natural-language stakeholder policies into real-time oversight for intelligent tutoring systems?
- Setting and data: Not stated in the paper
- Method: Conceptual architecture design and framework specification
Why it matters
Educational AI creates content autonomously, but lacks formal mechanisms to negotiate conflicting values among students, parents, teachers, and institutions. Existing systems lack practical architectures to translate ethical principles into real-time governance without disrupting pedagogical decisions.
Key findings
- The Advisory Governance Layer architecture separates governance advisement from pedagogical logic across four specialized agent types, providing non-intrusive oversight without altering core tutoring decisions.
- A four-part policy taxonomy—comprising hard constraints, soft preferences, temporal rules, and hierarchical rules—operationalizes natural-language stakeholder requirements into machine-executable formats.
- Distributed, federated voting enables stakeholder agents to evaluate tutoring actions locally and maintain privacy, exchanging only structured votes rather than revealing underlying policy rules.
Summary drafted with AI from the full paper and may contain errors. See the full paper for details.
What is the application?
Who is the user?
Who age?
Why use AI?
Study design
