Using deep learning to personalise and improve training of volunteers on citizen science projects
Online citizen science relies on the contributions of hundreds, sometimes thousands, of amateur researchers to a shared project. Two of the main challenges in citizen science are ensuring consistent quality of contributions and sustaining engagement. These challenges require sufficient training of citizen scientists on project-specific tasks and understanding participants’ motivations, respectively. While projects choose to address these issues in a variety of ways, most have to settle for broad-strokes solutions that cater to the majority, rather than being able to adapt to the unique needs of individual volunteers.
Gravity Spy is one of the most popular projects on the Zooniverse citizen science platform, where 12,000 registered citizen scientists help astronomers label known and novel categories of ‘glitches’ in the images generated by LIGO , an observatory that measures cosmic gravitational waves.
The human eye is still the best tool for distinguishing between sources of noise and novel features on images, which makes volunteer contributions vital to the research. The project has recently introduced AI into its volunteer training pipeline to improve the performance of tasks and the overall experience of volunteers. It uses a convolutional neural network (deep learning) algorithm to assess each volunteer’s ability to make classifications of different types of ‘glitches’ according to five levels of difficulty. The AI model uses this information to adapt to each individual volunteer and create a personalised training experience. This means that citizen scientists progress through different levels of the project’s workflow based on their individual ability.
The project illustrates the importance of uniquely human contributions and optimising the experience and performance of the individual citizen scientists taking part through AI.
By introducing AI, Gravity Spy project raised the quality of classifications from 54 per cent to 90 per cent accuracy and improved volunteer retention. This is just one of the many ways that the Zooniverse platform has been experimenting with integrating AI into the platform. As the largest online platform for crowd-based citizen science, Zooniverse has remained mindful of the impact of automation on their volunteer community. Gravity Spy is an example of using AI to improve the skills of volunteers so that their individual contributions to the shared goal of the project can be of better quality.