The complementary strengths of AI and human intelligence

Exploring AI-crowd interaction

The first group of grantees looked at ways in which AI can complement human crowd intelligence. This theme included experiments such as one by medtech Spotlab, which used serious games to train AI models for medical diagnosis. It asked if citizens playing online games could be as effective as physicians in training AI models for diagnosing neglected tropical diseases, such as malaria.

The experiment found that AI models trained on images annotated by both adults and schoolchildren can obtain similar results to ones trained on physician-based annotations, of around 93% accuracy.

Meanwhile, AI startup Samurai Labs grappled with how humans and AIs might work together to moderate online spaces and reduce cyber violence – and discovered that sometimes machine creativity can outperform the human variety.

Making better collective decisions

This second group of grantees explored how technology can be used to improve decision-making within groups by communicating opinion diversity. The University of Bristol built a swarm of robots to test whether they could help a crowd to reach an inclusive and informed consensus on challenging (and often polarising) topics such as climate change.

Participants input their responses to a question into the robot, which the robot then displayed for other participants to see before they responded to the same question. It showed that robot swarms can be used to engage people on challenging topics, to diffuse and influence opinions, serve as a prompt to launch conversations, and empower introverts to share their opinions in large group of people.

Understanding the dynamics of collective behaviour

The third group of grantees looked into ways in which positive behaviours and collective action can spread, and be encouraged, within and between groups. Urban design studio Umbrellium tested whether residents in a London borough could be connected together to collectively make environmentally friendly lifestyle changes to reduce air pollution – even though the direct individual effects of these actions might not be immediately noticeable.

It found that communication and collaboration among citizens led to the sustained adoption of these actions to decrease air pollution. The experimental groups saved, on average, four times more carbon dioxide emissions than the control groups, who were not allowed to communicate and collaborate within their groups.

Meanwhile, Nottingham University’s team explored ways in which improving inter-group communication might mitigate the tragedy of the commons (overusing of shared resources). The participants took part in a task that mimicked a common resource dilemma – participants (members of the public) in groups could individually claim from a common resource, but faced the prospect of receiving nothing if the total of their claims exceeded a certain threshold.

It found that communication (strongly) and connectivity (modestly) enhanced sustainable resource use, and that these two factors mutually reinforce each other. This has potential future applications for problems such as over-fishing, water table depletion, or even the fuel crisis witnessed in the UK last year.

Gathering better data

The fourth and final group focused on ways in which collective intelligence and AI can be applied to data. This included an experiment that took place in Bolivia, where Swisscontact, a charity that ​​focuses on international development projects, turned the usual development model upside-down by bringing Andean farmers into the world of meteorology.

Instead of imposing new technologies on them, their crowdsourced reports of micro-climate changes and pest outbreaks helped to power better forecasts across the region. Farmers who crowdsourced reports on pests and disease outbreaks were more likely to act on the information they received to protect their crops, such as by irrigating their crops more and applying more pest repellent, compared with farmers who were not involved.

The breadth of the experiments supported through the grants programme illustrates the diverse applications of collective intelligence and its potential to solve complex societal challenges.​​ It is vital that we continue to develop and test new and unique collective approaches, and collaborate in new ways in order to accelerate learning in the field.

Despite its potential, collective intelligence design is still a nascent area for research funding and is dwarfed by investments in AI, so funding for experimentation is crucial to help push the boundaries of existing practice and knowledge. Our hope is that these experiments, which were backed by Nesta, Wellcome, the Omidyar Network and the Patrick J McGovern Foundation, will prompt other funders to direct their resources to collective intelligence.

Funders who want to solve the complex global challenges of our time, from climate change to misinformation, will recognise the urgency of making progress in how we, as humans, understand, think and act together. Collective intelligence is about combining the best of human and machine intelligence to do that. It’s the best way to empower communities to act on the problems that matter to them and create the futures they want and deserve.

Kathy Peach is the director at Nesta’s Centre for Collective Intelligence Design