3.07 Case study: AI-enhanced citizen science platforms

This case study delves deeper into the practical application of the AI-powered citizen science models introduced in Module 3, focusing on the two leading platforms: iNaturalist and Zooniverse. These platforms are not just data collection tools; they are dynamic ecosystems where AI and collective human intelligence interact to create a self-improving cycle of learning and discovery.

iNaturalist has successfully integrated a sophisticated computer vision AI into its core user experience. When a user uploads a photograph of a plant or animal, the AI instantly suggests a list of possible species identifications. This AI model was not built in a lab in isolation; it was trained on the millions of observations previously identified and verified by the human community on the platform. This creates a powerful feedback loop: the community’s collective intelligence trains the AI. In turn, the AI’s suggestions make it easier for new users to participate and learn, which increases engagement and generates more data. This new data is then used to further refine the AI, making it more accurate and expanding its species coverage. This synergy enhances the learning experience for individuals while simultaneously creating a more detailed and valuable dataset for scientific research on biodiversity.

Zooniverse, on the other hand, often uses AI in a different but equally collaborative way. Many Zooniverse projects involve tasks too nuanced for current AI, such as identifying faint galaxies or transcribing complex historical handwriting. Here, the “wisdom of the crowd” is paramount. However, Zooniverse has been experimenting with integrating AI to optimise, not replace, the work of its human volunteers. In the ‘Gravity Spy‘ project, a deep learning algorithm creates a personalised training experience for each volunteer, assessing their skill in identifying different types of data ‘glitches’ and presenting them with tasks suited to their ability level. This AI-driven personalisation dramatically increased classification accuracy from 54% to 90% and improved volunteer retention. In other projects, volunteer classifications are used to train a machine learning model, which can then handle the most straightforward classifications, allowing human experts to focus their attention on the most interesting, unusual, or ambiguous cases.

Both platforms illustrate the core insight from Module 3: AI-powered citizen science is not about automation replacing people. It is about creating a socio-technical system where human and machine intelligence work in concert, each augmenting the capabilities of the other to achieve results that would be impossible for either to accomplish alone.