In Nesta’s recent report The Future of Minds and Machines, we looked at the the different ways AI can enhance our collective intelligence. While the study helped us understand what this emerging field of practice looks like, it also highlighted a number of gaps and future opportunities for research. This blog sets out the exciting new research at the frontline of this emerging field and introduces our new project to map the research that falls within and between AI and CI.

In The future of minds and machines, we mapped more than 100 case studies that made use of both AI and collective intelligence (CI) in sectors from health to agriculture and urban planning. We analysed 50 of these in detail in order to better understand the opportunities offered by combining human and machine intelligence for solving social challenges. Even though our analysis revealed that AI is already being applied across CI contexts, ranging from citizen science to crowd predictions (see the figure above and Making the most of the CI opportunity for more details), we found that there was not much variety or imagination in the ways that AI is used.

A large proportion of projects relied on either Computer Vision or Natural Language Processing which use machine-learning to process vast amounts of data, text and images. Given that many CI projects collect citizen-generated images, videos and text from participants, it is unsurprising that these AI methods dominated our sample. After all, a large amount of unstructured data is an ideal match for a data-hungry machine-learning algorithm. Current integration of AI is, therefore, predominantly aimed at the "low-hanging fruit" of overcoming data challenges faced by CI, leading to gains in efficiency and scale.

However, new investment and public sector experimentation could help to encourage methodological innovation in AI and CI over the next 5–10 years. In the future, we hope to see more imaginative uses of AI for collective benefit. This might come from improvements to existing technology or the transfer of current methods to new contexts. We discuss some of the potential sources of innovation below.

A group of methods referred to as distributed artificial intelligence (DAI)[1] have been gaining attention in recent years. Like many AI techniques, DAI methods are not new but using them in combination with machine-learning has resulted in new opportunities for better models of complex systems.

DAI is a class of algorithms based on the activities of many individual autonomous agents, each of which generates solutions to small parts of the overall problem. DAI is inspired by collective behaviour in the natural world, such as the swarming of bees or the interaction of individuals in social networks, and is often used to model complex systems with many parts. What all DAI approaches have in common is that no individual agent has enough information to get to the overall solution by itself.

DAI models are still more frequent in research settings. For example, conservation and ecology researchers have used DAI to explore the management of forests, the timing of animal migrations, and the population pressures on endangered species but there is increasing interest in using them to visualise the complexity of systemic public sector problems in scenarios ranging from urban waste management and forecasting the spread of infectious disease to the co‑ordination of emergency responses.

[1] Agent-based modelling (ABM), Multi-agent systems (MAS) and Swarm intelligence (SI) are different methodological approaches to DAI.

Developments in conversational AI and more widespread use of AI agents in group contexts are other areas that are likely to impact on CI in the coming years. Nesta’s Centre for Collective Intelligence Design has funded multiple experiments that investigate how AI agents could work in closed-loop systems with human groups to improve collective decision-making. These experiments are helping to advance our understanding of how to design group interactions that minimise the effects of social bias during problem-solving or helping participants to navigate collective risk dilemmas such as the climate crisis. Similarly, research by Anita Woolley is starting to demonstrate some of the potential productivity gains of integrating AI agents into teams as a facilitator who helps individuals make the most of their skills.

Finally, generative AI has the potential for more creative and collaborative interactions between groups of citizens or communities and institutions. For example, using existing generative design technology from engineering and product design in public sector planning could help transform the design of future cities, public buildings and even public services.

As well as the rise of new AI methods, the next 5–10 years are likely to bring changes to how these methods interact with data. There is a growing demand to ease the AI sector’s reliance on costly, labour-intensive labelling and vast datasets. Some of the changes will be methodological. New approaches to machine-learning that can train on little data are already gaining traction in the field. Some AI and CI projects that have difficulty generating huge labelled datasets, such as the smaller citizen science projects on Zooniverse, have started to experiment with these new approaches. Taking the opposite approach, the Syrian Archive project used computer-aided design to generate synthetic datasets. This helped the project to create a large enough training dataset for the data-hungry computer vision model that it needed to power its search engine.

Other likely changes include the way that projects work with data in response to growing concerns about data privacy and the exploitation of users’ data. For example, edge AI moves the source of computation towards the field of application, rather than relying on centralised servers. Similarly, federated learning is a machine-learning approach that enables sensitive data to stay with users on their devices. These techniques are particularly relevant to AI and CI projects where personally identifiable or sensitive data is collected from volunteer communities or citizens, such as humanitarian, health or digital democracy projects.

There is also a growing emphasis on deploying AI methods in combination with other technologies, including blockchain, quantum computing and the Internet of Things. Most of these convergence technologies are still being tested in research and development contexts and are far from being applied to high-stake problems in the real world.

Many of the newest methodological advances are less studied and so introducing them into the public sector setting too early poses a risk. The creation of controlled, smaller-scale testbeds for public problem‑solving could help to mitigate this risk while allowing the public sector and civil society to benefit from the latest innovations in the field.

In the following months, we'll be carrying out a large-scale analysis of the collective intelligence research landscape. We'll look at how scientific interests have evolved since 2000 and track the geography of research at the country, city and institution level to shed light on the dynamics of the CI ecosystem. In the network diagram below, the colours represent different "communities" of words that are highly interconnected. The font size of the label is proportional to the number of connections each word has in the network.

We'll also build on our existing research into the intersection of CI with AI, to better understand how the fields are connected and where they diverge in academia. As part of this work, we'll build a model to help us identify some of the key predictive characteristics of CI and AI research.

We believe that this project will help to build an evidence base on how CI research intersects with AI in order to decrease information asymmetries between practitioners and help them spot gaps and opportunities in their field. Coupled with Nesta’s Collective Intelligence Design Playbook and research on mapping the AI sector, we hope that our work will help to demystify CI and AI for policymakers and the civil society so that they can design new approaches to innovation that make the most of the complementary capabilities of groups of people and machines.