We live in an information age. We saw the benefits of this era during Covid when the 10,000+ articles published in 2020 sped up the process of discovery and innovation. But ironically this explosion of knowledge and data can make decisions even harder. Even experts struggle to identify and interpret the most relevant evidence amongst the noise. AI has been heralded as a solution but it has many limitations, particularly in domains such as medicine where bad decisions cost lives, or for complex problems such as climate adaptation where there isn’t one right answer. These situations require human judgement and reasoning. An alternative is using a combination of AI and other technology to augment collective human intelligence – ensuring that the best of diverse human expertise and insight can shine. This can help get us to the holy grail of better decisions in the face of increasing complexity, uncertainty and inequality.
In the HACID project we’re developing new technology that can support better decision making in health and climate adaptation. Our approach combines three key methodological innovations:
- crowdsourcing key insights from diverse groups of experts
- artificial intelligence to streamline processing of different sources of data
- knowledge graph technology
Knowledge graphs store data and information differently from traditional databases. They make it easier to discover new patterns and relationships between data – helping to overcome information silos. Many tools we use every day are powered by knowledge graphs, including Google Search and Amazon.
There are also more specialised domain-specific uses of knowledge graphs. For example, the life sciences company AstraZeneca built a knowledge graph containing many different sources of biomedical data such as genes, proteins, compounds and diseases. They use the graph to predict new disease targets for drug discovery.
The tools we’ll develop through HACID will support professionals working in two domains: medical diagnosis and climate adaptation. We’ll test whether technology that uses our approach helps experts in these fields make better decisions.
For example, our technology might help:
- medical students test their understanding of a particular disease, its causes, treatments and symptoms
- clinicians identify the latest insights and peer feedback when they’re trying to diagnose unusual cases
- climate scientists combine the most relevant data sources and models to predict future climate scenarios based on input provided by other domain experts
- city-level decision-makers, eg, Greater London Authority or Transport for London, develop strategies to adapt to potential extreme weather events like flooding and heat shock.
Our role within the project is twofold:
- Our designers are helping to map out the user requirements for both of the prototypes, working closely with a range of stakeholders including Met Office climate scientists, organisations from the London Climate Change Partnership and clinicians using the HumanDx platform. This will help us make sure that our tools meet the needs of decision makers.
- Our researchers are overseeing the design and evaluation of the participatory methods we’ll use to create the tools. We’re interested in understanding if a participatory approach impacts on perceived trust, efficacy and uptake of the tool in addition to a more technical evaluation of performance.
Three research questions are guiding our work.
- Are decision-making tools designed in a participatory way more trusted, better understood and more likely to be useful to professionals in medicine and climate services?
- Do tools that use both collective human intelligence and AI lead to better decisions than alternative approaches?
- What are the most appropriate benchmarks and measures of success for evaluating these tools, especially in domains where there isn’t one “right answer” (ground truth)?
At Nesta’s Centre for Collective Intelligence Design we create new technologies, processes and systems that make the most of human collective intelligence, or the enhanced capacity that’s created when diverse groups of people work together.
We first started thinking about how to best combine human collective intelligence and AI in our report, The Future of Minds & Machines. We also imagined future scenarios that brought together the best of both in our series on Minds & Machines. Our research shows that there is great potential for technology to enhance how we coordinate, communicate and understand complex information to create new solutions or make decisions in groups.
But new tools are often limited by a lack of explainability and transparency. And technology development in industry is driven by commercial priorities that focus on the low-hanging fruit of automating simple tasks rather than focusing on improving decision-making in the face of big societal challenges. It’s rare that domain experts and other affected stakeholders are involved throughout the design and development process. This can result in tools that aren’t fit for purpose. Research shows that reactions amongst professionals faced with using AI-enabled tools in their work can range from blind faith, when experts don’t critically interrogate outputs of tools or blind ignorance when experts don’t trust the tools and fail to integrate them into their workflows.
We think that involving domain experts in the design of new decision making tools can help to improve their performance and utility, as well as help professionals to build new skills and understanding of AI. Engaging wider stakeholders in the design process can also help identify potential unintended consequences and opportunities to address these. We call this approach Participatory AI.
Last year, we worked with the International Committee for the Red Cross to develop two humanitarian AI prototypes using our participatory approach. Through HACID we plan to continue refining our approach and understanding of participatory AI.
Over the course of the HACID project we’re aiming to:
- create two new decision-making tools to support a) healthcare professionals and medical students making clinical diagnoses and b) climate service providers such as the Met Office and organisations preparing for climate change like the London Climate Change Partnership
- test out new methods for involving end-users and stakeholders in the design and evaluation of technology, developing our Participatory AI methodology
- generate evidence about the potential value of combining AI and collective intelligence for complex high-stakes decisions
- generate evidence about the potential value of participatory AI
- identify new opportunities to apply a similar approach in other policy domains.
Nesta’s Centre for Collective Intelligence Design focuses on new ways to bring people, data and technology together to harness their collective intelligence, solve problems that matter, and strengthen collaboration between citizens and institutions. UK Research and Innovation (UKRI) funds the Nesta and Met Office contributions to the HACID project.
Other partners are funded by the European Union's Horizon Europe research and innovation programme under Grant Agreement No. 101070588.
We’ll be sharing what we learn along the way so look out for project updates. In the meantime, if you’d like to learn more about this project get in touch at [email protected] using the subject line HACID.