Research Summary: Positioning Generative Artificial Intelligence in STEM Assessment: When to Require, Scaffold, or Restrict Its Use

A recent research paper from arXiv addresses the complex governance challenge posed by Generative Artificial Intelligence (GenAI) in STEM assessment. The paper proposes a student-focused framework, based on Evidence Centered Design (ECD), to guide decisions on when to restrict, scaffold, or require GenAI use. This framework aims to provide decision rules linking assessment constructs, evidence requirements, and task characteristics to appropriate GenAI governance regimes, acknowledging the limitations of both unrestricted access and blanket prohibitions.

Why it matters

The integration of GenAI into educational and professional domains necessitates clear governance strategies to ensure fair and effective assessment. This framework offers a structured approach to managing GenAI use, preventing both the devaluation of assessment outcomes and the unpreparedness of future professionals for modern work environments.

Key insights

• GenAI in STEM assessment presents a significant governance challenge.
• Unrestricted GenAI access can compromise the validity of traditional assessments by enabling task outsourcing.
• Blanket prohibitions on GenAI are difficult to enforce and may lead to covert use.
• Prohibiting GenAI fails to prepare individuals for environments where GenAI-supported workflows are increasingly standard.
• A proposed framework, grounded in Evidence Centered Design (ECD), specifies when to restrict, scaffold, or require GenAI use.
• The framework provides decision rules based on target constructs, evidence requirements, and task characteristics to inform governance.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.07475