Interdisciplinary and inclusive collaboration

Interdisciplinary AI approaches in the context of environmental research and climate change studies are crucial. In environmental research, interdisciplinarity has been proven to accelerate the adoption of AI solutions from one sector to another — removing the need to start from scratch, avoiding massive reinvestment into challenges that are already addressed, and above all, providing faster solutions for climate change response through collaboration. Collaboration in this context is not a one-time intervention but is embedded throughout the project lifecycle. The success of all processes, from open communication to stakeholder engagement, and from guiding the co-development of project outputs to managing access to all research outputs ultimately depends on inclusive approaches to decision-making. Achieving these, nonetheless, can be extremely challenging, especially among distributed groups of stakeholders representing different objectives from their individual sectors and the problems their communities face.

Efforts across academia, government, private sectors and grassroots often remain disconnected, leading to ad-hoc development of tools for specific applications that cannot be easily adopted by others. Hence, it is important to engage all stakeholders through different modes of collaborative approaches that allow them to explore problems through multiple lenses, ask relevant questions, raise concerns, and collaboratively assess the needs of data-based solutions. The Turing Institute is uniquely placed to bring together partners and contributors from academia, industry, government and public sectors, as well as to engage members of the public in building modern, innovative and state-of-the-art technology. The Turing has specifically championed infrastructure roles in the Research Engineering Group (REG) and Research Application Management (RAM, a product manager equivalent at the Turing) team, each leading on specific collaborative aspects of big-team science (Forscher, Patrick S., et al., 2020). Members in these roles are contributing to building and deploying toolkits and examples from the real world and successfully implementing collaboration models that support inclusive and equitable interdisciplinary research. Research engineers and research applications managers typically work across multiple initiatives, creating new opportunities to connect projects and exchange practices. For example, Scivision is connected to 6 more existing projects at the Turing, all of which have collaborators from academia, industry and government beyond the institute. The Scivision platform enables researchers and algorithm developers from these projects to quickly change between algorithms or datasets, thus decreasing the time needed to adapt methods to datasets or vice versa, leaving greater time for scientific experimentation. In the near future, Scivision will integrate the DeepSensor codebase for increasing spatial granularity for real-time multimodal environmental monitoring adding a new challenge to Scivision. REG members assisted with internal practices for code sharing across projects (Scivision, Living with Machine).

One of Scivision’s data models comes from the MapReader (Hosseini, McDonough et al., 2021) developed under the ASG’s sibling in humanities, Living with Machines, a large-scale collaborative effort studying the impact of technology on people during the industrial revolution. MapReader was originally developed as an open source tool to analyse large collections of historical maps. Developed as a generalisable computer vision pipeline, it has found its use in a wide variety of domains, including classifying plant patches in geospatial images that are too large to closely investigate. Scivision’s adoption of MapReader has made it easy for users to run its model with the Scivision tool. Centre for Environment, Fisheries and Aquaculture Science (CEFAS) adoption of Scivision forms a striking example of both interdisciplinarity and the multi-domain use of generalisable tools. This interaction of CEFAS with ASG research work was led by the RAM team under the Turing’s Data Study Group, a collaborative hackathon event which brings together multi-disciplinary researchers to explore data analysis on practical challenges. In less than 6 months after the Data Study Group event, the RAM team catalysed Scivision’s deployment on the CEFAS research vessel – having improved plankton classification accuracy rates to over 90%.

These accelerated routes for solving real-world problems demonstrate the bigger roles of interdisciplinarity, infrastructure roles and cross-domain solutions in speeding up the deployment of research-based technology in different contexts. Future development work of the Turing research will take inspiration from these success stories to better inform the foundational development of AI/ML technology.