The Life Cycle of Large Language Models: A Review of Biases in Education
Abstract
Large Language Models (LLMs) are increasingly adopted in educational contexts to provide personalized support to students and teachers. The unprecedented capacity of LLM-based applications to understand and generate natural language can potentially improve instructional effectiveness and learning outcomes, but the integration of LLMs in education technology has renewed concerns over algorithmic bias which may exacerbate educational inequities.
Summary
Relevance to Cognitive Load Theory
This paper examines how LLMs are being integrated into education to provide personalized learning support. This directly relates to cognitive load theory because:
-
Personalization and Load Management: LLMs can potentially adapt content difficulty to individual learners, helping manage intrinsic cognitive load.
-
Instructional Design Implications: The paper maps the LLM life cycle from development to deployment, which has implications for how these systems can be designed to minimize extraneous cognitive load.
-
Bias and Learning Equity: Algorithmic bias could inadvertently increase cognitive load for certain student populations by providing inappropriate scaffolding or feedback.
Key Insights for AI Tutoring
The paper highlights that traditional bias metrics from ML fail to transfer to LLM-generated educational content because:
- Text is high-dimensional
- Multiple correct responses exist
- Tailoring responses may be pedagogically desirable rather than unfair
This nuance is critical for designing AI tutors that appropriately scaffold learning without overloading or underloading students.