摘要
BEYOND AUTOMATION: INVESTIGATING THE INFLUENCE OF COGNITIVE BIASES ON HUMAN INTERPRETATION OF EXPLAINABLE ARTIFICIAL INTELLIGENCE PROF(DR.) VIJAYKUMAR B GADHAVI PROFESSOR AND HEAD , SWARRNIM INSTITUTE OF TECHNOLOGY, SWARRNIM START-UP AND INNOVATION UNIVERSITY - GANDHINAGAR GUJARAT INDIA ORCID: 0009-0005-6313-5900, E-MAIL: DRVIJAYKUMAR@SWARRNIM.EDU.IN PARMAR JIGAR JAYSUKHBHAI B. TECH. CE STUDENT, SWARRNIM INSTITUTE OF TECHNOLOGY SWARRNIM STARTUP & INNOVATION UNIVERSITY, GANDHINAGAR, GUJARAT, INDIA ORCID: 0009-0008-7651-9913, EMAIL: JIGARAHIR410@GMAIL.COM Abstract Explainable artificial intelligence (XAI) is intended to make algorithmic outputs more understandable, auditable, and contestable. Yet an explanation does not reach a neutral observer: people interpret it through prior beliefs, cognitive shortcuts, professional experience, and assumptions about machine competence. This conceptual paper examines how cognitive biases shape the interpretation of XAI and why greater transparency does not necessarily produce better decisions. Drawing on research on automation bias, judgment and decision-making, explainability, and human–AI collaboration, the paper develops an integrative framework linking explanation characteristics, user cognition, task context, and reliance behavior. It argues that anchoring, confirmation bias, availability, authority effects, and cognitive miserliness can transform explanations into persuasive cues rather than evidence for independent evaluation. Explanations may improve mental models and error detection when they are faithful, appropriately scoped, and matched to users’ needs; however, they may also legitimize an incorrect prediction, increase cognitive load, or create an illusion of understanding. The paper proposes design principles centered on calibrated trust, contrastive and uncertainty-aware explanations, independent judgment, verification prompts, and evaluation of decision quality rather than satisfaction alone. Because this is a conceptual synthesis rather than an empirical study, its propositions require experimental testing across domains and user groups. The central conclusion is that responsible XAI must design not only for model transparency but also for the cognitive conditions under which people interpret explanations. Keywords: explainable artificial intelligence, cognitive bias, automation bias, human–AI interaction, trust calibration, decision-making, algorithmic transparency