International Open Access Journal Platform

AI as the Opponent: Constructing a Critical Thinking Training Model Driven by Large Language Models

Tao Wang¹², Liu Jiexian¹²

Guide to Education Innovation / 2026,6(2): 283-296 / 2026-07-02 17 13

View Download PDF

• Information:

  1. Tongling University, Tongling;
  2. Anhui Provincial Philosophy and Social Sciences Key Laboratory of Intelligent Decision Making in Copper Industry Development, Tongling

• Keywords:
  Large language models; Critical thinking; Opponent model; Functional alignment; Adversariality
• Abstract: The proliferation of large language models (LLMs) has deeply embedded AI-generated content into the information environment, presenting critical thinking education with a dual challenge: expanding the objects of critique from human-authored works to AI-generated content, and upgrading training methods from static text analysis to dynamic adversarial dialogue. This paper proposes a critical thinking training model termed “AI as the Opponent” and offers a systematic theoretical construction of this model. Drawing on Facione’s critical thinking framework, we demonstrate a functional alignment between three technical affordances of LLMs — simulating argumentative opponents, dynamically generating variants, and providing real-time feedback and probing questions — and the core dimensions of critical thinking competence. On this basis, we construct a four-stage training model: Identification — Questioning — Verification — Evaluation. Building upon this model, we distill a five-step operational loop: Role-based Prompting — Simulated Text Generation — Adversarial Inquiry — Cross-Verification — Structured Output, and integrate these theoretical elements into a unified “CST-ACTS framework”. The core theoretical proposition of this paper is that the unique pedagogical value of AI in critical thinking instruction lies in its “adversariality” — its inherent capacity to serve as an engageable opponent. Furthermore, we refine a transdisciplinary methodology of “functional alignment”, thereby providing a replicable analytical framework for other disciplines to construct their own discipline-specific “opponent” training models.
• DOI: 10.35534/gei.0602026 (registering DOI)
• Cite: Wang, T., & Liu, J. X. (2026). AI as the Opponent: Constructing a Critical Thinking Training Model Driven by Large Language Models. Guide to Education Innovation, 6(2), 283−296.