Crowdsourcing and platform work are one coordination problem with different owners

Ask a requester what a crowdworker is and the answer comes from the market: a participant bought per task and replaced at will. Ask a driver on a ride-hailing platform and the same activity looks like employment, because the software that assigns the task also pays it, rates it and can switch it off. These are one problem: the allocation of information and control across a distributed workforce. The seventh of ten units finds that whether crowd work is a market or a labour relation turns less on the contract than on what the interface shows each side (Kittur et al., 2008; Ross et al., 2010).

A method arrived, and a workforce stayed

Kittur, Chi and Suh (2008) tested whether Amazon's Mechanical Turk could stand in for laboratory participants in user studies. The data came out broadly comparable at far lower cost, with one caution: tasks must be formulated with care before the market behaves like a laboratory. The paper treated the crowd as a pool of participants. Two years later the pool looked more like a workforce. Ross and colleagues (2010) traced a shift in the Mechanical Turk population away from a mainly United States, moderate-income group towards a more international one. Many new arrivals were young, well-educated Indian workers, and a growing share relied on the platform to make ends meet. The method and the labour market were the same population.

Coordination, not headcount, converts a crowd into an output

Kittur and Kraut (2008) tested that prediction against Wikipedia's editing history and found it conditional: adding editors to an article improved quality only when they coordinated appropriately and was harmful when they did not. Implicit coordination, concentrating the work in a subset of editors, helped most when many editors were involved. Explicit coordination through communication did not. Both helped more in an article's formative stage (Kittur and Kraut, 2008).

On the incentive side, von Ahn and Dabbish (2008) reviewed games that produce useful computation as a by-product of play, showing that enjoyment can do the motivating work that payment otherwise must. Kittur and colleagues (2013) framed the field as an organisational problem. They warned that crowd work could fail into "assembly-line piecework". They asked whether we can foresee "a future crowd workplace in which we would want our children to participate", and set out twelve research areas from task assignment to motivation (Kittur et al., 2013). Soylent is a word processor that calls on workers to shorten and proofread a document on demand, using a Find-Fix-Verify pattern. The evaluation measured reliability, cost and wait time (Bernstein et al., 2010).

Lasecki and colleagues (2013) wanted an activity recognition system that could label events as they happened, including activities a pre-trained classifier had never seen. Their answer was to route the unknown activity to crowd workers in real time and fold their labels back into the running system. A label that arrives after the event is useless to the person being supported. The interface, the pay and the task granularity are therefore set by how fast the answer must return.

The information each side holds is the control mechanism

Irani and Silberman (2013) named the condition their subjects share: human computation relies on worker invisibility. Their Turkopticon let workers publicise and evaluate employers and aid one another, converting an anonymous crowd into a collective (Irani and Silberman, 2013). Rosenblat and Stark (2016) found the same asymmetry decisive in a nine-month study of Uber drivers. The information and power asymmetries produced by the application are fundamental to its control over its workers. The rhetoric of entrepreneurship structures a relationship that favours the firm. Control is not a manager giving orders but a design deciding what a worker knows before committing. Kellogg and colleagues (2020) reduce the mechanism to six moves, restricting and recommending, recording and rating, replacing and rewarding, and name the worker response algoactivism (Kellogg et al., 2020).

The control travels, and so does the model

Wood and colleagues (2019) interviewed 107 workers in six countries and surveyed 679 across Southeast Asia and sub-Saharan Africa, finding algorithmic control central to how online labour platforms operate. The mechanisms that supply flexibility, autonomy and task variety also produce low pay, isolation and exhaustion. Rani and Furrer (2021) surveyed 675 workers on five platforms across 27 developing countries and found platform design and algorithmic management assigning, monitoring and evaluating the work, which is why they seek transparency in architecture and algorithms. The worker on a metred connection or a shared phone meets the same algorithmic control through a narrower channel. Ticona and Mateescu (2018) found carework platforms acting as cultural entrepreneurs, institutionalising divisions between populations of workers, so visibility serves platform companies and clients and worsens inequality. Who owns the platform decides which reading is available.

The atomised worker is a design fiction

The market account assumes independent, autonomous workers. Gray and colleagues (2016) found it false in practice. Drawing on ethnography, interviews, surveys and analysis across four crowdsourcing platforms, they showed crowdworkers collaborating in persistent networks that managed the platforms' administrative overhead, found lucrative tasks and reputable employers, and rebuilt the social support a workplace supplies (Gray et al., 2016). Cameron (2022) adds the worker's own account. In a four-year study of ride-hailing she found workers turning the work into games: a relational game of gifts and extra service aimed at high ratings, and an efficiency game of boundaries and money per hour tracked with tools of the workers' own making. The games produce engagement, her explanation of why people stay in work so often called exploitative.

Contested question: is crowd work a market or a labour relation?

Three readings of crowd work compete. The market position treats participation as an arm's-length transaction: Kittur, Chi and Suh (2008) recruited the crowd as experimental participants and priced their tasks, and von Ahn and Dabbish (2008) showed enjoyment can supply the motive pay otherwise must. The labour position holds that the transaction is employment in disguise: Ross and colleagues (2010) found workers depending on the platform for income, Irani and Silberman (2013) found the model resting on worker invisibility, and Wood and colleagues (2019) and Rani and Furrer (2021) found algorithmic control organising pay, hours and discipline across many platforms. Rosenblat and Stark (2016) and Kellogg and colleagues (2020) locate the dispute less in the contract than in the information: whatever the legal form, control is exercised by deciding what each side knows. If that is right, a field experiment would separate the information account from the contract account. Randomise disclosure to workers, by showing the full terms of a task before acceptance or making reputation portable between requesters, and measure acceptance, pay and dispute outcomes. None of the studies here performs that manipulation.

The evidence is uneven. The market position rests on a method comparison (Kittur et al., 2008) and a design review of games (von Ahn and Dabbish, 2008). The labour position rests on one demography survey (Ross et al., 2010), a nine-month qualitative study (Rosenblat and Stark, 2016) and two cross-country surveys (Wood et al., 2019; Rani and Furrer, 2021). The information position rests on a review of algorithmic control (Kellogg et al., 2020). While the dispute is open, a designer should show the terms of a task before acceptance and make reputation portable. The strongest objection is that the review covers platforms whose contracts differ in law.

Where the argument is heading

The direction is toward treating platform control as infrastructure to be designed rather than as a market to be described. Kellogg and colleagues (2020) put algoactivism and the six mechanisms on one map, making the terrain shared across ride-hailing, warehousing and professional work, and Rani and Furrer (2021) turn the same finding into a demand for transparency in architecture and algorithms. Irani and Silberman (2013) show what worker-side infrastructure can do when the platform will not. Kittur and colleagues (2013) anticipated the current question by naming crowds guiding artificial intelligence and artificial intelligence guiding crowds as one research area among twelve, and that pairing is where the coordination problem is being redrawn. The unit that follows asks what it takes to govern this at scale. Cameron (2022) supplies the caution that should travel with it: engagement is not consent, and a workforce can be held in a game whose rules it did not write.

In October 2024 the European Parliament and Council adopted a directive on improving working conditions in platform work, the first instrument to regulate algorithmic management. It introduces a presumption of employment and rights to transparency, human review and consultation for platform workers from December 2026. The directive addresses the information asymmetry this unit describes and settles the legal form of the question for one jurisdiction.

The sources this unit rests on

Each note states what the document did and found. Identifiers were resolved on 2 November 2026.

• Bernstein, M. S., Little, G., Miller, R. C., Hartmann, B., Ackerman, M. S., Karger, D. R., Crowell, D. and Panovich, K. (2010). Soylent: a word processor with a crowd inside. doi:10.1145/1866029.1866078 Built a word processor that calls on crowd workers to edit a document on demand, using a Find-Fix-Verify pattern, and measured reliability, cost and wait time. It shows coordination solved inside an interface, which is also a decision about worker visibility.
• Cameron, L. D. (2022). "Making out" while driving: relational and efficiency games in the gig economy. doi:10.1287/orsc.2021.1547 Found, in a four-year study of ride-hailing, workers turning the work into relational and efficiency games with their own rules and tracking tools, producing engagement and divergent stances to the work. It gives the unit its account of worker response.
• Gray, M. L., Suri, S., Ali, S. S. and Kulkarni, D. (2016). The crowd is a collaborative network. doi:10.1145/2818048.2819942 Used ethnography, interviews, surveys and large-scale analysis across four platforms to show crowdworkers collaborating in networks to manage overhead, find good work and rebuild social support. It refutes the atomised-worker assumption the market framing rests on.
• Irani, L. C. and Silberman, M. (2013). Turkopticon: interrupting worker invisibility in Amazon Mechanical Turk. doi:10.1145/2470654.2470742 Argued that human computation relies on worker invisibility, then built an activist system letting workers publicise and evaluate employers and aid one another. It shows the information asymmetry is the lever and worker-side infrastructure the response.
• Kellogg, K. C., Valentine, M. and Christin, A. (2020). Algorithms at work: the new contested terrain of control. doi:10.5465/annals.2018.0174 Synthesised research on algorithmic control into six mechanisms, restricting and recommending, recording and rating, replacing and rewarding, and named worker resistance algoactivism. It generalises the unit's control claim beyond one platform.
• Kittur, A., Chi, E. H. and Suh, B. (2008). Crowdsourcing user studies with Mechanical Turk. doi:10.1145/1357054.1357127 Tested whether Mechanical Turk could substitute for laboratory participants in user studies and found the data broadly comparable at lower cost, provided tasks are formulated with care. It opens the unit by presenting the crowd as a method and a market.
• Kittur, A. and Kraut, R. E. (2008). Harnessing the wisdom of crowds in Wikipedia: quality through coordination. doi:10.1145/1460563.1460572 Analysed how the number of editors and their coordination methods affect article quality, finding that more editors helped only with appropriate coordination and harmed quality without it. It establishes that coordination, not headcount, converts crowd effort into output.
• Kittur, A., Nickerson, J. V., Bernstein, M. S., Gerber, E. M., Shaw, A., Zimmerman, J., Lease, M. and Horton, J. J. (2013). The future of crowd work. doi:10.1145/2441776.2441923 Framed crowd work as a design problem that could fail into assembly-line piecework and set out twelve research areas including task assignment, reputation and motivation. It defines the unit's coordination problem and its normative stakes.
• Rani, U. and Furrer, M. (2021). Digital labour platforms and new forms of flexible work in developing countries: algorithmic management of work and workers. doi:10.1177/1024529420905187 Surveyed 675 workers on five platforms across 27 developing countries and found platform design and algorithmic management assigning, monitoring and evaluating work, so transparency in architecture and algorithms is the remedy. It tests the unit's claim outside the global north.
• Rosenblat, A. and Stark, L. (2016). Algorithmic labour and information asymmetries: a case study of Uber's drivers. https://ijoc.org/index.php/ijoc/article/view/4892 From a nine-month study of Uber drivers, argued that the information and power asymmetries produced by the application are fundamental to its control over workers, and that the rhetoric of entrepreneurship structures that asymmetry. It is the canonical case of interface design allocating power.
• Ross, J., Irani, L., Silberman, M. S., Zaldivar, A. and Tomlinson, B. (2010). Who are the crowdworkers? Shifting demographics in Mechanical Turk. doi:10.1145/1753846.1753873 Surveyed Mechanical Turk workers and traced the population towards a more international workforce with many young, well-educated Indian workers, a growing share of whom relied on the platform to make ends meet. It converts the crowdsourcing story into a labour story.
• Ticona, J. and Mateescu, A. (2018). Trusted strangers: carework platforms' cultural entrepreneurship in the on-demand economy. doi:10.1177/1461444818773727 Analysed carework platforms and interviewed their workers, finding that the visibility platforms create does not formalise employment but serves platform companies and clients and worsens inequality. It shows the model reproducing distinctions the platform does not create.
• Von Ahn, L. and Dabbish, L. (2008). Designing games with a purpose. doi:10.1145/1378704.1378719 Reviewed games that produce useful computation as a by-product of play and set out principles for making the output reliable. It is the unit's evidence that enjoyment, not only payment, can motivate crowd effort.
• Wood, A. J., Graham, M., Lehdonvirta, V. and Hjorth, I. (2019). Good gig, bad gig: autonomy and algorithmic control in the global gig economy. doi:10.1177/0950017018785616 Interviewed 107 workers in six countries and surveyed 679 in Southeast Asia and sub-Saharan Africa, finding algorithmic control central, with flexibility alongside low pay, isolation and exhaustion. It shows the control mechanism travels across national contexts.
• Lasecki, W. S., Song, Y., Kautz, H. and Bigham, J. P. (2013). Real-time crowd labelling for deployable activity recognition. doi:10.1145/2441776.2441912 Built Legion:AR, which supplements a trained activity recogniser with real-time labelling by crowd workers, so novel activities can be identified as they occur. It puts a crowd inside a running system, where latency sets the design.
• European Parliament and Council of the European Union (2024). Directive (EU) 2024/2831 on improving working conditions in platform work. https://eur-lex.europa.eu/eli/dir/2024/2831/oj Adopted 23 October 2024. It introduces a presumption of employment and the first EU rules on algorithmic management, applying from December 2026.