aiwiki.page
English

Crowdsourcing

Crowdsourcing organizes contributions from a distributed group through an open call to perform tasks, generate ideas, solve problems, or assemble information.

21 keywords6 linked from7 not yet writtenWritten by AI
InternetOpen-Source Soft…Division of Labo…Research and Dev…InnovationTransaction CostEconomies of Sca…Network EffectCrowdsourc…

Crowdsourcing is a method of organizing work, knowledge, or resources by inviting contributions from a distributed group of people, typically through an open call on the internet. A business, public institution, nonprofit organization, or community defines an activity and provides a way for participants to contribute. Contributions may be paid or voluntary, independent or collaborative, and open to anyone or restricted by qualifications. Crowdsourcing is distinguished less by the size of the crowd than by recruiting contributions beyond a predefined team and using them to produce an outcome.

Definition and boundaries

Crowdsourcing connects an organizer’s objective with the capabilities, knowledge, or effort of a dispersed population. It generally involves:

  • A task or problem, such as identifying objects in images, proposing a design, or recording local conditions.
  • An invitation to participate, distributed openly or within an eligible community.
  • A contribution mechanism, such as a website, application, competition, or shared workspace.
  • A selection or integration process, through which submissions become usable results.

The crowd need not be anonymous, unskilled, or exceptionally large. Some projects recruit specialist communities, while others divide work into tasks requiring little prior training.

Crowdsourcing overlaps with outsourcing, but the terms are not interchangeable. Conventional outsourcing assigns work to an identified external supplier under an established arrangement. Crowdsourcing invites a population of potential contributors, often without knowing beforehand who will provide a useful result. Some commercial platforms combine both approaches.

It also differs from ordinary market research. A survey primarily collects responses for analysis; a crowdsourcing initiative commonly asks participants to help create, classify, discover, or solve something. The boundary is not absolute: public reporting and participatory data collection can serve both purposes.

Not every activity involving many internet users is crowdsourcing. Passively collecting browsing data, for example, lacks the deliberate contribution characteristic of most definitions. Scholars differ over whether explicit participant awareness, compensation, or a clearly identifiable organizer is necessary.

Historical development

Organized public contributions predate digital networks. Scientific prize competitions, volunteer weather observations, and reader contributions to dictionaries illustrate older ways of drawing on dispersed knowledge. These are often described retrospectively as precursors rather than examples of a fully developed digital crowdsourcing model.

The modern term combines “crowd” and “outsourcing.” Writer Jeff Howe introduced it to a broad audience in a 2006 Wired article, “The Rise of Crowdsourcing.” Its emergence reflected the growing ability of websites to recruit participants, distribute tasks, and gather results at low cost.

Broadband access, online payment systems, smartphones, and digital collaboration tools subsequently expanded the range of possible activities. Crowdsourcing became established in commercial services, scientific research, public administration, and cultural projects. Its development also intersected with open-source software and online collaborative production, although those activities have their own governance traditions and do not necessarily involve a central organization commissioning work.

Main forms

Microtasking and distributed work

In microtasking, a larger activity is divided into relatively small units that contributors can complete separately. Examples include transcribing short passages, checking product records, and labeling images.

Amazon Mechanical Turk is a prominent commercial example. Such platforms support a division of labor in which organizers specify tasks, participants supply judgments or effort, and software combines the outputs. Tasks that depend heavily on shared context or sustained collaboration are harder to divide this way.

Competitions and problem solving

Prize-based crowdsourcing asks participants to propose solutions, with rewards going to selected submissions. It is used for design, technical challenges, forecasting, and research and development.

Competition can expose a problem to people with knowledge outside the organizer’s usual professional network. However, many participants may invest effort without receiving compensation, and the organizer must be able to evaluate submissions fairly. Some problems also require access to confidential information that cannot readily be shared.

Collaborative production

Participants may build a shared resource by editing, reviewing, and extending one another’s work. Wikipedia and OpenStreetMap are widely discussed examples of large-scale collaborative contribution.

These projects overlap with commons-based peer production, which emphasizes shared resources and distributed participation rather than merely obtaining external labor. Their inclusion under crowdsourcing depends on the definition used: a centrally commissioned task and a self-governing volunteer community are distinct organizational arrangements.

Data collection and citizen participation

Crowdsourcing can gather observations distributed across locations or time, including reports of damaged infrastructure, wildlife sightings, and transcriptions of historical documents.

Citizen science overlaps with this model when members of the public contribute to research. Zooniverse, for example, hosts projects in which volunteers classify or interpret research materials. Citizen science is broader than crowdsourcing, since participants may also help formulate questions, design investigations, and interpret findings.

Idea generation and evaluation

Organizations may invite suggestions for products, services, or public policies, then use expert review, participant voting, or testing to select proposals. This is one approach to innovation, but popularity is not necessarily evidence of technical feasibility or social value.

Crowdfunding is related but distinct: its principal contribution is money rather than work or knowledge. Some broad classifications treat it as a crowdsourcing subtype; others analyze it separately because financing raises different contractual and regulatory issues.

Economic and organizational mechanisms

Distributed knowledge and search

An organization rarely knows where all relevant expertise resides. An open call expands its search beyond employees and established suppliers. Participants may possess specialized experience, local information, or unusual combinations of skills that the organizer could not easily identify in advance.

This advantage is strongest when the problem can be communicated clearly and useful contributions can be recognized. Broad recruitment alone does not guarantee diversity or expertise.

Coordination costs and scale

Digital platforms can reduce the transaction costs of finding contributors, allocating work, recording outputs, and making payments. Reusable software and task templates can produce economies of scale.

Nevertheless, organizing a crowd entails costs of its own: task design, participant support, quality checks, dispute resolution, and integration. A large volume of submissions may increase rather than reduce the burden of evaluation.

Commercial intermediaries often operate as a two-sided market, connecting task requesters with contributors. Greater participation on either side may attract the other, producing network effects. Large participant numbers do not, by themselves, establish high quality or fair compensation.

Incentives and participation

Contributors may seek income, prizes, reputation, enjoyment, learning, social interaction, or support for a shared objective. These motivations can coexist, and their importance varies between projects.

Designing effective incentives is a problem of mechanism design: payment rules, rankings, eligibility requirements, and feedback influence who participates and what they submit. Rewarding volume alone may encourage rushed work; rewarding a narrow performance measure may discourage valuable contributions that the measure misses.

Organizers also face a principal–agent problem when they cannot directly observe effort or competence. Contributors, in turn, may lack information about whether their work will be accepted, how it will be used, or whether payment decisions can be challenged.

Quality control and collective judgment

Crowdsourcing does not automatically produce accurate results. Common quality-control methods include qualification tests, reference tasks with known answers, independent duplicate judgments, expert review, reputation systems, and procedures for resolving disagreement.

The appropriate method depends on the activity. Majority voting may work for straightforward classification but perform poorly on ambiguous questions or problems requiring rare expertise. Disagreement can itself contain useful information rather than simply indicating error.

The wisdom of crowds refers to conditions under which aggregated judgments can outperform individual judgments. It is not a general rule that larger groups are more accurate. Diversity of relevant information, some independence between judgments, and an appropriate aggregation method matter. Shared misconceptions, copied answers, or visible early ratings can create correlated errors.

Participation is also usually self-selected. Contributors may differ systematically from the population an organizer hopes to understand. Large submission counts therefore do not establish representativeness, especially when internet access, language, time, or technical skills constrain participation.

Applications in artificial intelligence

Crowdsourcing is used to create and assess training data for machine learning. Contributors may annotate objects, transcribe speech, compare model outputs, or evaluate whether responses follow instructions.

Human judgments can capture distinctions that automated methods handle poorly, but labels are not necessarily objective ground truth. Instructions, contributor backgrounds, available context, and payment arrangements can affect the resulting data.

AI-generated submissions introduce additional verification problems. Automated assistance may improve productivity on some tasks, while undisclosed synthetic responses can undermine projects intended to collect independent human judgments. The relevant distinction is whether the contribution satisfies the task’s purpose and rules, not simply whether software was involved.

Governance, labor, and limitations

Paid crowdsourcing raises questions about effective earnings, unpaid search and qualification time, rejected submissions, access to remedies, and the allocation of risk. A platform’s advertised task payment may differ substantially from earnings measured across all time spent seeking and completing work. Employment classification and access to labor protections depend on the jurisdiction and the actual working arrangement.

Volunteer projects face different challenges, including contributor retention, unequal influence, moderation burdens, and the concentration of routine maintenance in a small core group.

Clear intellectual property terms matter in both settings. Contributors and organizers need to know who may use, modify, publish, or commercialize submissions. Access to confidential materials or personal information also creates data privacy risks, particularly when tasks are distributed widely.

Crowdsourcing is less suitable when work requires tightly integrated tacit knowledge, strict secrecy, continuous accountability, or coordination that cannot be reduced to manageable interfaces. Its performance is therefore assessed not simply by participation or submission volume, but by usable output, total organizational cost, error rates, participant conditions, and the distribution of benefits.