Mechanical Turk shutting down September 30
Amazon’s decision to shut down its Mechanical Turk crowdsourcing platform on September 30 is seen as the end of an era for low-paid, microtask-based online work. Commenters link the closure to the rise of AI and LLMs, which can now perform many of the “unskilled” data-labeling and survey tasks MTurk specialized in, combined with growing fraud, quality problems, and arbitrage by bots and intermediaries. While some note this as evidence of AI progress, others highlight the loss of income for workers in poorer countries and the impact on researchers and pollsters who relied on access to large pools of human respondents.
Platform trajectory and internal context
- A major long-time requester reports MTurk was effectively orphaned after its program manager moved to other AWS AI products; only billing was modernized.
- Commenters note MTurk had become an outlier within AWS and its shutdown isn’t surprising.
- Earlier decision to stop accepting new customers in July is referenced as precursor.
Why MTurk is shutting down (per thread)
- Many argue MTurk’s core “unskilled microtask” niche has been largely eaten by LLMs and other AI tools.
- Open, low-friction marketplaces are now easily flooded by bots and task-arbitrage schemes, degrading data quality.
- Maintaining a “high trust, high verification” human market would push prices up and shrink its usefulness.
Quality, fraud, and verification problems
- Long-time requesters describe pervasive fraud and misaligned incentives: buyers understate survey length; workers speed through or random-answer for pennies.
- Residential proxies and VPNs are used to bypass geographic targeting, particularly in political polling.
- Some built elaborate systems: multiple responses per item, consensus checks, banning low-quality workers — still only 85–95% accuracy.
- One user’s trial of MTurk via Bedrock “ground truth” for simple digit labeling produced near-random accuracy.
Economic and labor perspectives
- Several describe the work as monotonous and mentally draining, with very low pay; some call the business model exploitative and are glad to see it go.
- Others emphasize it provided meaningful supplemental income, especially in lower-income regions, and predict the shutdown will hurt those workers.
- Observations that price mainly controlled speed of completion, not quality.
Alternatives and future of human microtasks
- Discussion points to specialized data-labeling/RLHF firms and in-house teams as the new norm, particularly for domain-specific tasks (e.g., medical).
- Some foresee human-in-the-loop roles for robotics (remote assistance for robots) but note latency, safety, and expertise constraints.
- Other freelance platforms (Upwork, Fiverr) are reported as already struggling with AI-generated work and misrepresentation.
Notable use cases and anecdotes
- MTurk was used for: biomedical annotation, market research, doctoral paper summarization, psychology studies, PDF data extraction, transcription of personal letters, SEO manipulation, and high-profile search efforts (e.g., missing aviator).
- Mixed nostalgia: “end of an era” sentiment alongside critiques of poor data reliability and worker treatment.