I wrote recently about the Intelligent-TPACK framework and the argument that teachers need pedagogical judgment and ethical assessment, not just technical skills, to use AI effectively in their classrooms (Celik, 2023). That post dealt with the theoretical model. This one looks at what happens when the theory meets a real classroom.
Park, Teo, Teo, Chang, Huang, and Koo (2023) followed a group of science teachers in Singapore as they integrated machine learning into a Grade 7 science unit on models. The study captures something that’s hard to find in the AI-in-education literature: an honest, detailed account of what teachers actually experience when they try to bring AI into their existing curriculum.
AI Integration in Science Through the Concept of Modeling
The lesson package connected AI to science through a shared concept: model-building. Students used the Orange software platform to train machine learning models using logistic regression, working with a Mars Rover activity that involved 32,000 data points.
What’s interesting is that several teachers initially saw AI and science as completely separate domains. That perception shifted once they engaged with the lesson materials. One teacher described the moment of recognition:
“Actually, I viewed AI as quite separate from AI and science… it’s only after I got involved… then I realized. Actually, there’s the link, and for us as a teacher, the most obvious link is to the team of models… You put some data in. It creates a model, and that model can be used to solve problems” (p. 9).
That shift is significant. AI didn’t enter the curriculum as an isolated technology topic. It entered through a scientific concept that students were already learning. The modeling framework gave teachers a conceptual anchor, a way to make AI feel like a natural extension of what they were already teaching.
Science Frames the Problem, AI Supports the Prediction
The teachers in this study were clear about the relationship between AI and science. They saw the two as complementary, each contributing something the other couldn’t. Science provides the conceptual grounding and the judgment to decide which variables are meaningful. AI provides computational power and pattern detection.
One teacher captured this well: “Scientists’ jobs are really to help us to see the data that’s useful and the data that are not… I think that’s the value of the scientific model… and AI is a tool, right, to help us to supplement that” (p. 12).
I appreciate how clearly this positions AI as a support tool within disciplinary thinking. The science knowledge tells you what to look for. AI helps you find patterns in the data once you know what questions to ask. That framing echoes what Cheng et al. (2025) found in their writing study: students who brought their own questions to AI got more out of the interaction. The same logic applies here. Teachers who bring disciplinary understanding to AI integration create richer learning experiences.
The Black Box Was Acknowledged, Not Avoided
The teachers acknowledged that students lacked the mathematical background to fully understand logistic regression. They treated parts of the algorithm as a functional black box and focused on the logic of training and testing models: 70% of the data for training, 30% for testing.
One teacher explained the approach: “We don’t actually teach them about how the algorithms work… they just have to do it, except that this AI program… is learning… then at the end of it, it makes a prediction” (p. 10).
This is a pragmatic and honest instructional decision. Teachers drew parallels with how science already handles complexity. Students use instruments and apply formulas without fully understanding every underlying mechanism. The goal at this level is procedural understanding and conceptual awareness, not algorithmic mastery.
And the Orange platform helped. Its visual flow diagrams let students follow the sequence of steps in training and testing a model, even if the mathematical details remained abstract. As one teacher reflected: “It provides the visual part. Visual understanding of the flow of how the AI creates the model” (p. 14).
Teacher Confidence Was the Biggest Barrier to AI Integration in Science
The strongest theme across the findings wasn’t about student readiness or curriculum fit. It was about teacher confidence. Multiple participants reported feeling underprepared to explain AI concepts to their students, even after engaging with the lesson materials.
One teacher was candid: “I might not have like a lot of confidence in explaining it to my students. If I have no background in AI and then suddenly I need to teach them… I also would have some trouble with that” (p. 15).
Another described how preparation time was consumed by personal learning: “I was not confident about my content knowledge of AI to teach it, and therefore most of my initial time was spent on developing my content knowledge rather than how to teach it” (p. 15).
This finding lands differently when you read it alongside the Intelligent-TPACK research. Celik (2023) showed that technical knowledge alone doesn’t predict effective AI integration, that pedagogical and ethical knowledge are the stronger predictors. Park et al. show what that gap looks like from the inside: teachers spending most of their energy trying to understand the content, with little left over for thinking about how to teach it well.
The confidence issue also mirrors what Bilbao-Eraña and Arroyo-Sagasta (2025) found in their pre-service teacher study. Short-term training can improve awareness and attitudes, but building the deeper confidence required for sustained integration takes more time and more targeted support.
AI as Supplement, Not Core
The teachers in this study positioned AI as supplementary to science. They recommended placing AI-integrated lessons within enrichment programs or after-school settings, especially for students who already have strong science interest. They raised practical concerns about curriculum time, student readiness, and buy-in from colleagues.
One teacher put it simply: “I would say AI is a supplement. It’s like an extra thing, adding on” ( p. 15).
That framing reflects where many schools are right now. AI integration is happening at the edges, carried by individual teachers who are motivated enough to try it. Scaling it into core curriculum will require the kind of structured teacher preparation and institutional support that Celik’s framework describes.
The study concludes with a recommendation I fully agree with: teacher AI literacy must come first. Without targeted professional development that builds both content knowledge and pedagogical confidence, AI integration will remain fragile and limited to motivated individuals working in enrichment contexts.
Reference
Bilbao-Eraña, A., & Arroyo-Sagasta, A. (2025). Fostering AI literacy in pre-service teachers: Impact of a training intervention on awareness, attitude and trust in AI. Frontiers in Education, 10, 1668078.https://doi.org/10.3389/feduc.2025.1668078
Celik, I. (2023). Towards Intelligent-TPACK: An empirical study on teachers’ professional knowledge to ethically integrate artificial intelligence (AI)-based tools into education. Computers in Human Behavior, 138, 107468. https://doi.org/10.1016/j.chb.2022.107468
Cheng, Y., Fan, Y., Li, X., Chen, G., Gašević, D., & Swiecki, Z. (2025). Asking generative artificial intelligence the right questions improves writing performance. Computers and Education: Artificial Intelligence, 8, 100374. https://doi.org/10.1016/j.caeai.2025.100374
Park, J., Teo, T. W., Teo, A., Chang, J., Huang, J. S., & Koo, S. (2023). Integrating artificial intelligence into science lessons: Teachers’ experiences and views. International Journal of STEM Education, 10, 61. https://doi.org/10.1186/s40594-023-00454-3