Combining Crowds and Machines: Experiments in collective intelligence design 1.0
New technologies, including artificial intelligence (AI), allow us to mobilise human intelligence in new ways and at greater scale. The COVID-19 pandemic has demonstrated the many uses of collective intelligence: from symptom-tracking apps to open-source production of medical equipment. But despite its potential, collective intelligence is still a nascent area for research funding and is dwarfed by investments in AI.
In spite of many emerging opportunities, we still know relatively little about what works and how to do it well. Through our Collective Intelligence Grants programme – the first of its kind – we supported 12 diverse organisations worldwide to conduct experiments that increase our understanding of how to make the most of the new technologies available to help with collective thinking and acting.
Analysis of the first round of experiments has provided new insights into how we can improve our decision-making, enable effective cooperation, make better use of citizen-generated data and increase the effectiveness of participation in collective intelligence initiatives.
Some of the findings provide the basis for further research, whereas others will be directly applicable in practice. Below are seven key insights:
- To make better decisions, delegate to AI – Passing responsibility to autonomous agents increased cooperation and coordination between groups.
- Want happier voters? Let them swarm! – Politically polarised British voters were happier with decisions made through ‘swarming’ and Borda count methods; majority voting consistently produced the least satisfactory decisions.
- Use AI to stop people following the herd – Mediating group decisions through an AI system reduced the tendency of people to go along with the group majority and led to more accurate outcomes.
- When fast action is needed, let the crowd self-organise – In time-critical scenarios, such as food rescue efforts, effective coordination among different actors is key. Decentralising coordination through a collective intelligence platform allowed volunteers to adapt to a changing situation and led to a significant increase in the amount of food that was rescued.
- To make digital democracy work better, use AI to help people find similar ideas – Natural language processing could improve the effectiveness of citizen participation on digital platforms, mainly by reducing the time it takes to find similar proposals.
- Offering better rewards or more varied tasks doesn’t get better results from crowdworkers – When analysing Twitter data for disaster recovery, higher pay seemed to have an adverse effect on labelling accuracy.
- AI recommendations can increase the engagement of citizen scientists by helping them discover less popular projects – Recommendation algorithms can better match users to projects they are interested in, increasing the activity of volunteers.
The first major funder to put £10 million into this field has the opportunity to make a lasting impact on it. We hope that the funding gap can be filled by UK Research and Innovation and foundations, as well as by public funders integrating collective intelligence into existing AI funding programmes.
We are very grateful to Aarathi Krishnan, Artemis Skarlatidou, Beth Noveck, Georgia Ward Dyer, Joysy John and Juan Mateos-Garcia for taking the time to advise us on the shortlisted applications and supporting us in the selection process.