Questioning the Answer: Science Education in the Age of AI

Artificial intelligence can produce a scientific explanation in seconds. It can summarise research, analyse information and present an answer that sounds confident and convincing. The challenge is that a convincing answer is not always a correct one.

A recent Australian report by Learning First, drawing on responses from nearly 3,400 teachers and more than 750 school leaders, highlights the impact AI is already having on student learning and assessment. The report warns of cognitive outsourcing, where students hand the thinking itself over to AI and bypass the productive effort required for genuine learning. It also draws attention to the “mirage of false mastery”, where a polished piece of work may create the appearance of understanding without providing trustworthy evidence of what the student knows and can do.

This is where Science education is becoming more important than ever. Learning scientific knowledge remains essential, but what happens in a Science classroom is about much more than remembering the parts of a plant. It teaches students how to ask worthwhile questions, analyse data, recognise patterns and evaluate the quality of evidence. Students learn to consider alternative explanations and decide whether a conclusion is genuinely supported by the information available.

These skills have always mattered and extend well beyond the science classroom. In a world where students can access an immediate and convincing answer to almost any question, they need the knowledge and confidence to pause, question it and decide whether it can be trusted.

Headmaster Chris Ivey recently spoke about the importance of helping students embrace AI in their learning without allowing it to do the work for them. Within Science at Somerset, we are considering what this looks like in practice.

We do not want to ignore AI or simply tell students not to use it. It is already part of the world in which they are being educated, and it will almost certainly be part of their future workplaces. Our students need to learn how to use AI effectively and responsibly. At the same time, they need enough scientific knowledge to recognise when an AI response is inaccurate, incomplete or unsupported.

We also need to protect the opportunities students have to develop their own critical thinking. When a polished answer is always available at their fingertips, it can be tempting to skip the difficult thinking that leads to genuine understanding.

This year, our Year 10 science students have been completing Critical AI Prompt Engineering lessons. These lessons have two connected aims. The first is to show students how the quality and specificity of a prompt affects the responses produced by AI. Students compare different prompts and evaluate how factors, such as scientific detail, intended audience and required format, influence the accuracy and effectiveness of the explanation they receive. They then use what they have learned to construct a purposeful scientific prompt of their own.

The second, and more important aim, is to teach students that generating a response is not the end of the process. Once AI has produced an explanation, students critically analyse it. They look for scientific misconceptions, missing information, unsupported claims and language that sounds impressive but does not actually explain the science. Through this process, students learn that a well-written response is not necessarily an accurate one.

Prompting is only one part of the process. Asking AI a better question may improve its response, but it does not remove the student’s responsibility to check the answer. We want students to ask: Is this scientifically accurate? What evidence supports it? Has anything important been left out? Could this explanation create a misconception? Most importantly, do I understand the science well enough to make that judgement?

The same Year 10 cohort is also completing its Research Investigation through a safe exam browser. This controlled environment restricts access to generative AI while students complete the task in class. It gives them the opportunity to analyse evidence, develop an argument and communicate their own scientific reasoning without an AI tool generating the response for them.

Protecting assessment integrity should not mean abandoning rich tasks such as research investigations. Our challenge is to preserve these valuable learning experiences while also gathering trustworthy evidence of what each student knows and can do.

These approaches are intended to work together. At some points, students are taught how to use AI carefully, purposefully and critically. At others, they work without it and demonstrate what they can do independently. Students need experience with both. They need to understand when AI can support their learning, but they also need the confidence and ability to think without it.

Year 10 was deliberately selected for this approach because the assessment is completed and graded internally, with common conditions across the Somerset cohort. This allows us to strengthen students’ research, analysis and scientific writing before they enter the senior years, without affecting externally comparable results or their future pathways.

Although Year 10 was chosen for the controlled assessment approach, our response to AI is not limited to restricting its use. Recent research from Learning First argues that discussions about AI in education must consider who is using it, for what purpose and in which educational context. This has guided the different approaches we have taken across year levels.

On the Student Free Day, all Somerset staff attended an AI conference where Professor Kai Riemer discussed the potential value of using AI as a feedback tool when it is guided by a carefully constructed prompt. Following this, Year 12 Physics students were provided with a pre-constructed prompt to help them self-assess their completed Research Investigation draft before final submission.

The prompt was deliberately designed to keep the student responsible for the thinking. It could not write or rewrite any part of the report. Instead, it asked students to reconsider their reasoning, identify areas requiring further attention and check whether they had addressed the assessment criteria. Students then had to decide for themselves whether a change was needed and how that change should be made.

This approach also responds to the concern raised by Learning First about “cognitive outsourcing”, where students hand the thinking itself over to AI. The prompt was most useful for students who had already completed a thoughtful draft. It could help them identify gaps and question their decisions, but it could not replace the research, scientific understanding or effort required to produce the report.

We cannot know which technologies will have the greatest impact, or what careers will exist, when our younger students leave school. What we can do is make careful decisions about when AI may support learning and when students need to work without it. New technologies will continue to produce exciting claims about what might soon be possible. Our aim is for students to leave Somerset science able to approach those claims with interest, while still asking for the evidence.