Specialty Digest

Global English editionLearners’ Study Dashboard — in development
The Knowledge Library for All

Collective Intelligence

nounEmerging AI Researchalso group intelligence, wisdom of crowds, swarm intelligence in groups

In one line

Collective intelligence is the enhanced capability that emerges when many individuals, human or artificial, work together, share information and combine judgments, so that the group can solve problems, make predictions or make decisions better than its members acting alone.

Collective intelligence is the ability of a group to do things that its individual members cannot do alone, or cannot do as well. It appears in ant colonies finding the shortest path to food, in scientists building knowledge together over centuries and in online communities solving puzzles. In the field of artificial intelligence, the term is used both for human groups that use technology to coordinate and for systems in which many software agents cooperate. It connects to swarm intelligence, artificial life, AI agents and human-AI collaboration.

Classic examples

### Wisdom of crowds

Francis Galton reported in 1907 that at a country fair the median of 787 guesses of the weight of an ox was within about one percent of the true weight [1]. James Surowiecki popularized the idea in his 2004 book The Wisdom of Crowds, which argued that groups judge well when members are diverse, independent, decentralized and their answers are aggregated sensibly. Prediction markets and polling averages use this principle.

### Social insects and animals

Ants, bees and termites build complex nests and find food without central planning, using simple local rules and signals such as chemical trails. Fish schools and bird flocks move in coordination. These examples inspired algorithms in computing; see swarm intelligence.

### Open collaboration

Wikipedia, launched in 2001, brings together millions of volunteers to write and correct an encyclopedia [3]. Open-source software projects such as Linux and citizen science programs such as Galaxy Zoo, where volunteers classify galaxy images, show how large numbers of contributors can produce high-quality work.

### Scientific communities

Science itself is a collective process: peer review, replication and cumulative citation allow a community to correct errors and build knowledge over time; see AI in scientific research.

What makes groups smart

Research suggests that group performance depends more on how members interact than on how clever individual members are. In a 2010 study in Science, Anita Williams Woolley and colleagues found evidence of a general collective intelligence factor that predicted performance across a range of tasks, and it was correlated with the social sensitivity of members and the equality of conversational turn-taking, but not strongly with the average or maximum intelligence of individuals [2]. Other conditions often cited are:

  • Diversity of perspectives and skills.
  • Independence of judgments, so that errors are not all in the same direction.
  • Good aggregation methods, such as averaging, voting or markets.
  • Clear communication and shared goals.
  • Psychological safety, so people speak up.
  • Incentives and feedback that reward accuracy.

Collective intelligence and AI

### Humans and machines together

Platforms combine human judgment with algorithms for tasks such as moderation, mapping and medical image review. Researchers at institutions such as the MIT Center for Collective Intelligence study how computers can connect people so that the group is smarter, sometimes called superminds. See human-AI collaboration.

### Machine learning built on collective data

Large models learn from text, images and code created by millions of people, so they can be seen as compressing and recombining collective human knowledge, which raises questions about credit and consent; see AI copyright and training data. Crowdsourced labeling services and human feedback also shape model behavior; see human-in-the-loop AI.

### Multi-agent AI systems

Groups of AI agents can divide tasks, debate answers or check each other’s work. Ensembles of models often outperform single models, and agent teams are used for software development and research; see AI agents and agent-to-agent protocol.

### Federated and distributed learning

Many devices can jointly train a model without sharing raw data, which supports privacy; see AI privacy.

Applications

  • Forecasting and prediction markets.
  • Crowdsourced science and mapping, such as disaster response maps.
  • Public policy and deliberation platforms that gather citizen input.
  • Product design, open innovation and bug finding.
  • Medical diagnosis through combined opinions of several experts.
  • Education, where peer learning improves understanding; see AI in education.

Risks and limitations

  • Herding and information cascades. When people copy each other, errors spread and diversity is lost.
  • Manipulation. Bots, fake accounts and coordinated campaigns can distort crowd signals; see AI misuse.
  • Bias and exclusion. Crowds reflect who participates, which may leave out groups; see algorithmic bias.
  • Groupthink. Strong social pressure reduces independent thinking.
  • Quality control. Open systems need moderation and verification.
  • Overreliance on AI aggregates. If people rely on the same models, their errors may become correlated, reducing the diversity that makes crowds wise.
  • Exploitation. Crowd workers who label data are sometimes poorly paid; see AI ethics.

A worked example

A city wants to estimate the number of potholes needing repair. Staff can inspect only a small part of the streets. The city invites residents to report potholes through an app, and uses a model to check photos and remove duplicates. It adjusts for the fact that some neighborhoods report more than others by comparing reports with a random sample of inspections. The combined estimate is far more accurate than either source alone, and repairs are scheduled fairly. The city also shares results publicly, so that residents can see progress and correct errors.

Designing for collective intelligence

Organizations and platforms can improve group performance through deliberate design. Useful practices include collecting individual estimates before discussion so that people are not anchored by the first speaker, rotating roles so that quieter members contribute, using structured methods such as the Delphi technique in which experts revise forecasts over several anonymous rounds and tracking accuracy so that well-calibrated contributors receive more weight. Digital tools can summarize arguments, display where opinions differ and surface minority views, and some deliberation systems use language models to find common ground among large numbers of participants. Designers must watch for hidden bias in summaries and keep humans accountable for decisions; see human oversight.

Further reading and next steps

Readers can continue with swarm intelligence, AI agents and technological singularity to see how groups of minds, natural and artificial, might shape the future.

FAQs

What is collective intelligence? It is the ability of groups, human, animal or artificial, to solve problems and make decisions better than individuals acting alone.

What is the wisdom of crowds? It is the observation that the combined judgment of many independent people can be very accurate.

Does a smarter group need smarter members? Not necessarily. Research suggests that interaction patterns, such as turn-taking and social sensitivity, matter a great deal.

How does AI relate to collective intelligence? AI can connect people, aggregate their input, work as a team of agents and learn from collective data.

What can go wrong? Groups can fall into herding, manipulation, bias or groupthink, which weakens their accuracy.

Key facts

  • Francis Galton reported in 1907 that the median of 787 guesses of the weight of an ox at a country fair was within about one percent of the true weight, an early example of the wisdom of crowds1.
  • Anita Williams Woolley and colleagues reported in 2010 in Science that groups show a general collective intelligence factor that predicts performance across tasks and is correlated with the social sensitivity of members and equality of conversational turn-taking rather than with the average or maximum individual intelligence2.
  • Wikipedia, launched in 2001, is often cited as an example of large-scale collective intelligence, with millions of volunteers creating an encyclopedia without central direction3.

Related concepts

Quick checkWhat is collective intelligence?Show answer

Collective intelligence is the enhanced capability that emerges when many individuals, human or artificial, work together, share information and combine judgments, so that the group can solve problems, make predictions or make decisions better than its members acting alone.

Sources

  1. Galton, Vox Populi, Nature, https://www.nature.com/articles/075450a0
  2. Woolley et al., Evidence for a Collective Intelligence Factor in the Performance of Human Groups, Science, https://www.science.org/doi/10.1126/science.1193147
  3. Wikimedia Foundation, About Wikipedia, https://wikimediafoundation.org/about/

Editorially reviewed by Specialty Digest Editorial TeamLast reviewed November 17, 2026Researched and drafted with AI assistanceReport an issue