What the study found
GPT-4o demonstrated some aspects of the nature of science and the nature of GenAI-influenced science. It also showed both strengths and weaknesses in differentiating between those two kinds of scientific understanding.
Why the authors say this matters
The authors say this matters because educational technologists need to fine-tune generative AI so it can communicate its own influence in producing scientific knowledge. The study suggests this could support ethical and responsible use of GenAI in science education and help teachers foster students’ epistemic learning outcomes in science.
What the researchers tested
The researchers examined whether and how GPT-4o, a recent generative AI model, can represent the nature of science and the nature of GenAI-influenced science. They interviewed GPT-4o using the Family Resemblance Approach, which includes categories such as aims and values, methods and methodological rules, knowledge, and practices.
What worked and what didn't
GPT-4o was able to demonstrate some aspects of the nature of science and the nature of GenAI-influenced science. However, it also had both strengths and weaknesses in making epistemic differentiation, meaning distinguishing between how science works and how GenAI influences scientific knowledge production.
What to keep in mind
The abstract does not provide detailed examples of GPT-4o’s responses or specify exactly which aspects were strong or weak. It also does not describe broader limitations beyond the scope of this single-model study.
Key points
- GPT-4o showed some ability to represent the nature of science.
- GPT-4o also represented the nature of GenAI-influenced science to some extent.
- The model showed both strengths and weaknesses in distinguishing between science and GenAI-influenced science.
- The authors link the findings to the design of classroom instruction and students’ epistemic learning outcomes.
- The study used the Family Resemblance Approach and interviewed GPT-4o about aims, values, methods, knowledge, and practices.
Disclosure
- Research title:
- GPT-4o shows partial understanding of science and GenAI-influenced science
- Authors:
- Kason Ka Ching Cheung, Wenhao Zhang
- Institutions:
- Education University of Hong Kong, Chinese University of Hong Kong
- Publication date:
- 2026-02-11
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.