This study investigates how undergraduate translation students perceive and evaluate AIgenerated translations of scientific and technical texts in comparison to their own humanproduced versions. Thirty third-year students in Iran participated in a classroom-based task where each translated an English scientific text into Persian and generated two additional versions using AI tools (ChatGPT and Copilot). Students assessed the three translations— student-generated, AI Tool 1 (ChatGPT), and AI Tool 2 (Copilot)—using a structured questionnaire based on five quality criteria: accuracy, fluency, terminology, readability, and reliability. They also responded to two open-ended questions about their preferences and evaluations. Quantitative findings revealed that students consistently rated their own translations highest across all criteria, followed by ChatGPT and then Copilot, with statistically significant differences confirmed by Friedman test. Thematic analysis of open-ended responses showed a strong preference for human translations, mainly attributed to greater perceived accuracy, terminological precision, natural tone, and academic appropriateness. While AI tools were acknowledged for fluency and terminological strengths, they were criticized for inconsistency, lack of coherence, and mechanical language. The findings suggest that while students acknowledge the potential of AI tools, they still trust human translation more in academic and technical contexts. The study highlights the need to integrate AI literacy into translator training, encouraging students to critically evaluate and refine machine-generated output.
