This study aims to examine how secondary science teachers perceive,
interpret, and respond to students’ use of generative artificial
intelligence(generative AI) in science performance assessments, and how
these judgments are enacted in actual classr...
This study aims to examine how secondary science teachers perceive,
interpret, and respond to students’ use of generative artificial
intelligence(generative AI) in science performance assessments, and how
these judgments are enacted in actual classroom instruction and assessment
practices. As generative AI has rapidly permeated educational contexts,
teachers’practical responses to students’AI use in science performance
assessments were analyzed through the lens of the Perception–Interpretation
–Decision(PID) model and Visible Practice.
The participants consisted of 35 high school science teachers working in
metropolitan areas, representing a wide range of teaching experience. The
data were collected through a survey focusing on teachers’ experiences
and cases of unintended student use of generative AI, their positions toward
the adoption of generative AI and the reasons for those positions, their
experiences or intentions regarding the transformation of performance
assessments involving generative AI, the student competencies that should
be emphasized in future science performance assessments, and open-ended
comments. The collected data were analyzed using qualitative content
analysis.
Teachers’ responses were segmented into meaning units and categorized
into Perception, Interpretation, and Decision according to the PID
framework. Through iterative reading and thematic analysis, common
meaning units within each category were identified, resulting in five themes
for Perception, eight themes for Interpretation, and six themes for Decision.
The Perception category included themes such as awareness of students’
generative AI use, difficulty in verifying AI use, detection cases, perceived
effects of generative AI, and recognition of the widespread diffusion of
generative AI in society. The Interpretation category encompassed themes
related to assessment validity and fairness, emphasis on thinking and inquiry
competencies, emphasis on generative AI literacy, perceived educational
usefulness and reliability of generative AI, concerns about overuse and
dependency, limitations of prohibition and restriction, and institutional
insufficiencies. The Decision category comprised themes such as simple
allowance, caution and guidance, strict control and sanctions, active
adoption, separation of usage domains, and assessment redesign.
The analysis revealed that teachers did not perceive students’ use of
generative AI in a uniform manner. Even when confronted with similar
phenomena, teachers made different judgments and enacted different
practices based on divergent perceptions and interpretations. At the
perception stage, some teachers viewed the spread of generative AI as an
uncontrollable reality, while others perceived it as a threat to assessment
credibility. At the interpretation stage, these perceptions were variously
framed as issues of prohibition and control, guidance and pedagogical
transition, or opportunities to reconstruct assessment goals. At the decision
stage, teachers demonstrated diverse practical responses, including explicit
prohibition of AI use, conditional allowance with rules, structural
transformation of performance assessment tasks, and reinforcement of
process-oriented feedback.
Notably, even among teachers who selected outwardly similar practices,
the underlying PID reasoning pathways differed. For instance, some
teachers who restricted or prohibited AI use grounded their decisions in
concerns about assessment fairness and controllability, whereas others
emphasized the educational goal of protecting students’ thinking processes.
These findings suggest that teachers’ responses to generative AI cannot
be reduced to a simple dichotomy of approval versus opposition or adoption
versus non-adoption, highlighting the importance of analyzing the underlying
judgment logics that inform similar decisions.