Work

Trust, Critical Thinking, and Psychological Safety in Human–AI Interaction: A Framework for Calibrated Reliance and Intellectual Autonomy

Year: 2026
Type: preprint
Abstract: The growing integration of generative artificial intelligence (GenAI) into education, professional practice, research, and everyday decision-making has transformed the psychological conditions under which individuals acquire information, evaluate evidence, and formulate judgments. Although conversational AI systems can improve access to knowledge, facilitate reflection, and support problem-solving, their fluent, personalized, and often socially responsive outputs introduce distinctive risks of overreliance, automation bias, confirmation bias, and reduced intellectual autonomy. This conceptual paper examines trust, critical thinking, and psychological safety as interrelated dimensions of human–AI interaction. Drawing on research concerning trust in automation, metacognitive regulation, algorithmic reliance, AI sycophancy, and the psychosocial consequences of conversational systems, the paper argues that effective human–AI interaction requires neither unquestioning confidence nor generalized distrust, but context-sensitive and evidence-calibrated reliance. The paper introduces the Critical Trust and Psychological Safety Framework (CTPSF), an integrative conceptual model comprising four interdependent dimensions: epistemic reliability, critical engagement, decisional autonomy, and relational psychological safety. A complementary five-stage TRUST protocol is proposed to operationalize the framework through task-risk assessment, evidence evaluation, uncertainty recognition, adversarial examination, and preservation of human judgment. Particular attention is given to distinguishing supportive AI interactions from uncritical validation, identifying indicators of excessive cognitive and emotional dependence, and developing practical strategies that preserve users' capacity to disagree, reconsider, verify, and act independently. The framework provides a theoretical foundation for educational interventions, professional AI literacy programs, interface design, and future empirical studies. Its effectiveness remains to be established through controlled research
Language: English
Open Access status: green
OpenAlex ID: W7235904255
Cites:
Cited by:
Related to: 10