Toward an AI Learning Environment: Vygotskian Scaffolding Approaches to Generative Artificial Intelligence

This article traces the trajectory from Vygotskian cognitive scaffolding to its contemporary elaborations within the ecology of the Extended Mind Theory (EMT) and the recently proposed concept of System 0. We conceptualize artificial intelligence (AI) as a dynamic cognitive infrastructure that functions both as a support system and as an invisible filter shaping our everyday cognitive ecosystem. Within this framework, we compare three modes of AI-mediated learning: OpenAI-ChatGPT Study Mode, Google-Gemini Guided Learning, and Anthropic-Claude Learning Mode. Each was analyzed across eight core dimensions: diagnostic capacity, problem decomposition, dialectical questioning, graduality of suggestions, progressive fading of support, metacognitive promotion, verification strategies, and multimodal integration. Findings reveal a nuanced landscape: ChatGPT emerges as an adaptive tutor of excellence in the textual domain; Gemini distinguishes itself as an orchestrator of multiple expressive modalities; Claude positions itself as a mentor fostering critical thinking and intellectual autonomy. The article concludes with concrete design recommendations, including explicit diagnostic mechanisms and continuous adaptivity, a balanced integration of text and media, transparent fading processes, sustained metacognitive promotion, age-sensitive differentiation, and strong safeguards for formative integrity. The core thesis is clear: AI should not replace the teacher but can significantly expand the learner’s Zone of Proximal Development (ZPD), provided it is designed to foster active participation and critical engagement rather than diminish them.