Who Is Solving the Challenge? The Use of ChatGPT in Mathematics and Biology Courses Using Challenge-Based Learning

Synthesis: Elizondo-García, Hernández-De la Cerda, Benavides-García, Caratozzolo, and Membrillo-Hernández (2025) report a pilot study using ChatGPT within challenge-based learning (CBL) in two higher-education digital courses — Fundamentals of Biological Systems (biology) and Mathematics and Data Science for Decision Making. Students solved activities with ChatGPT, then verified output quality against high-quality traditional sources. Surveys and NLP topic modeling revealed that students valued ChatGPT for its immediacy, ease, and Accessibility while raising concerns about the veracity and depth of its output, whether it would supplant the teacher, and whether it would erode skill and competency development. The authors call for updated academic-integrity codes and new AI-use ethics, and note that teachers should be prepared to use AI and that AI-text detectors should be available for evaluation.

Design and context

• Setting: Tecnológico de Monterrey, two digital (distance-learning) higher-education courses: biology (Fundamentals of Biological Systems) and mathematics/data science.
• Pedagogies and Teaching Strategies: challenge-based learning (CBL) under the Tec21 educational model — students address real challenges, here using ChatGPT to solve assigned activities.
• Verification step: students were explicitly instructed to validate ChatGPT's outputs against formal, high-academic-quality sources.
• Method: surveys (Likert-scale items) plus NLP analysis of open-ended responses — sentiment analysis, Latent Dirichlet Allocation (LDA), and Latent Semantic Analysis (LSA) topic modeling.

Student perceptions (findings)

• Positive: students appreciated ChatGPT's immediate access to information, ease, availability, and clear/concise explanations; it supported personalized and self-directed learning. LDA revealed five themes: (1) information-retrieval tooling, (2) rapid/dynamic learning with technology, (3) ease and speed of use, (4) research and specific responses, and (5) AI as a support tool.
• Concerns: veracity and depth of topics, the tool's reliability (outputs "simplify research" but sometimes lack precision or current data), potential to supplant the teacher, and risk of excessive dependency that could impair critical thinking, complex reasoning, and Problem Solving.
• Notably neutral: sentiment analysis showed 90% neutral responses, indicating students were still largely unfamiliar with ChatGPT's full capabilities — "we are still at the beginning of using this tool."
• Academic integrity gap: a third of surveyed students did not know about modifications to academic-integrity regulations regarding AI use.

What this means for practice

• Instructors. Build output validation into the activity itself: students here were instructed to check ChatGPT against high-quality formal sources, and they valued its speed while doubting its veracity and depth.
• Instructors. Coach critical interrogation alongside use — students worried the tool would supplant the teacher and weaken critical thinking, complex reasoning, and Problem Solving.
• Administrators. Update academic-integrity codes for AI use and publicize the changes, since a third of surveyed students did not know about the modifications to regulations already made.
• Administrators. Make AI-text detectors and clear ethical-use guidance available for evaluation, and ground them in responsible use, fairness, and transparency.
• Instructors. Position ChatGPT as a support tool for personalized, self-directed learning rather than an answer machine, extending information access without displacing teacher guidance.

Limitations

• The study is a pilot that measures perceptions — a concluding Likert survey plus NLP of open-ended responses — with no measure of learning outcomes or skill development.
• The two courses differ sharply in size — 94 biology students across six campuses versus 392 mathematics/data-science students — and the authors attribute some cross-course variation to the smaller biology sample.
• Sentiment analysis returned 90% neutral responses, which the authors read as limited familiarity with ChatGPT, so the thematic evidence is dominated by undifferentiated sentiment.
• All data come from two distance-learning courses at a single institution, Tecnológico de Monterrey, so curriculum and context effects cannot be separated from the tool.

Citation

Elizondo-García, M. E., Hernández-De la Cerda, H., Benavides-García, I. G., Caratozzolo, P., & Membrillo-Hernández, J. (2025). Who is solving the challenge? The use of ChatGPT in mathematics and biology courses using challenge-based learning. Frontiers in Education, 10, 1417642.