microtask

A microtask is a type of crowdsourcing task designed to be completed quickly and by a large number of workers. Each task is small, discrete, and typically requires minimal training, allowing many tasks to be finished in a short period. Microtasks are a common component of data annotation workflows used to generate labeled data for machine learning, natural language processing, computer vision, and related AI applications.

Common microtask types include data labeling (such as image, video, or text tagging), transcription, simple categorization, ranking or relevance judgments, verification of information, and metadata extraction. These tasks are usually designed to be objective and finite, with clear acceptance criteria to enable automated aggregation of results.

The typical workflow involves distributing tasks to a pool of workers through an online platform. Tasks are often assigned redundantly to multiple workers to improve accuracy, and results are aggregated via majority voting, consensus scoring, or weighted inputs. Quality control may incorporate gold-standard checks, calibration tasks, and performance-based pay adjustments to maintain reliability.

Microtasks underpin many applications, notably the rapid creation of large labeled datasets for AI training and evaluation, content moderation, data cleaning, and quality assurance in various digital services. Platforms such as Amazon Mechanical Turk, Appen, Figure Eight, and similar services facilitate microtask creation, assignment, and aggregation, sometimes with integrated quality controls and incentive structures.

Ethical and practical considerations include fair compensation, worker privacy, and the potential for worker exploitation. The concept emerged with the broader rise of crowdsourcing in the 2000s, with MTurk popularizing the use of small, scalable tasks to support larger computational workflows.