Swarm AI

www.nesta.org.uk/feature/ai-and-collective-intelligence-case-studies/swarm-ai/
Skip to content

The challenge

Bringing together diverse groups of people to make decisions or predictions is a key goal of many collective intelligence projects and a core principle of the wisdom of crowds. However, the interactions between people who have different experiences and viewpoints can be difficult to co‑ordinate. Conflicting viewpoints, interference from social biases and failure to tap into the expertise in the group can all stand in the way of an optimal consensus being reached. We therefore see many attempts to mobilise networked groups of individuals (both online and offline) leading to the opposite outcome, with incendiary behaviours in response between those who disagree and the rising polarisation of views.

The AI and CI solution

The Swarm AI platform, developed by Unanimous AI, is a rare example of distributed AI and human groups working together on a task in real time.

Swarm AI is inspired by the collective behaviour of natural systems, such as flocks of birds and swarms of bees. Swarm intelligence algorithms moderate the interaction of a group of individuals who are deciding between a set number of options. Each person can log into the online platform at their location. The algorithms are trained on data about behavioural dynamics of groups, rather than on the subjects they are debating. Individuals connect with each other and AI agents to form a closed-loop system where both the machine and individuals can react based on the behaviour displayed by others to change or maintain their preference. In a second step, a neural network model trained with supervised machine‑learning uses the interaction dynamics of the participants to generate a conviction index. This index estimates the group’s confidence in the final outcome.

So what?

The Swarm AI platform has increased the accuracy of group decisions across a wide variety of tasks, from health diagnostics to forecasting political polls. For example, diagnostic accuracy of a small group of networked radiologists working as a real-time swarm intelligence system reduced errors by 33 per cent in comparison to the individuals on their own, and by 22 per cent compared to an AI-only solution. Unanimous AI has claimed that the Swarm AI system navigates the group towards optimal consensus decisions, which result in higher levels of satisfaction in the group. As of January 2020, Swarm AI has been deployed in decision‑making in commercial settings and research environments, but the results show promise for applications in the public sector, such as prioritisation of public policy.