As generative AI becomes increasingly embedded in academic practice, it presents both new risks and new opportunities for research reproducibility. This project explores how AI-human collaboration can be utilised in teaching to promote reproducibility as a core research skill.
The reproducibility crisis in science has underscored the urgent need for better research practices and more transparent methodologies. The ‘Exploring AI-Human Triangulation for Research Reproducibility’ project set out to investigate how generative AI could be used not just to automate reproducibility checks, but to actively teach students the critical thinking skills needed to evaluate research themselves.
Rather than positioning AI as a replacement for human judgement, the project aimed to embed human critique into every stage - using AI to scaffold the process, not take it over. The goal was to create a learning experience where AI supports rigorous analysis while ensuring students remain the ultimate decision-makers.
With technical support from Xavier Laurent, Training Manager, and Haseeb Ahmad, Senior Research Software Engineer, in the AI Competency Centre, the team developed the ‘Reproducibility Analyser’: a web-based application that combines large language models with a structured checklist of reproducibility standards. The tool guides users through evaluating research manuscripts, presenting both a PDF viewer and the AI’s step-by-step reasoning for each checklist item.
A codesign workshop with students in March 2025 provided valuable feedback. Participants appreciated the transparency of the AI’s reasoning and suggested improvements such as breaking down checklist items into smaller, more manageable components to facilitate more precise evaluation.
Key lessons emerged. Transparency and explain-ability are not optional features but essential pedagogical tools. Involving students as co-designers strengthens both the tool and the learning experience. While rapid prototyping proved useful, it also highlighted the need for a simple, intuitive interface when the goal is education rather than experimentation.
Ultimately, the project reinforced that integrating AI into education should not be about handing over decisions to machines, but about creating reflective, supportive systems that deepen human insight.