Amazon Mechanical Turk: New customers barred, is this the end?

Amazon decided to block access for new customers to its long-standing Mechanical Turk platform. This move, announced for July 5, 2024, hints at a slow farewell to a service that provided "human intelligence" for digital tasks for years.

Mechanical Turk, launched by Amazon in 2005, allowed companies to outsource repetitive, simple jobs. We're talking about things like identifying objects in photos or transcribing audio, mundane tasks for us but tough for computers back then.

It was sort of the "grunt work" of AI, performed by a global and often underpaid workforce. Without those small human contributions, many early artificial intelligence systems would never have learned to distinguish a cat from a dog. Amazon announced it will stop accepting new customers for Mechanical Turk starting July 5, 2024.

The official reason isn't super clear, but it's likely the market has changed dramatically. With AI advancing and more specific services emerging, Mechanical Turk's old model might not be as profitable or strategic for Amazon anymore.

Today, more sophisticated and specialized alternatives for data labeling and crowdsourcing exist. Platforms like Scale AI or Appen offer targeted services, often with AI tools assisting workers, making MTurk feel a bit outdated. Who knew AI would put the very people who raised it out to pasture?

Plus, AI itself has gotten much better at tasks that once required a human. Think image recognition or transcription: what used to be a HIT is now often an integrated function in large language or visual models. Platforms such as Scale AI and Appen have emerged as specialized alternatives to Mechanical Turk for data labeling.

For the average person, probably not much, unless they were a "Turker" or a company using it. For the AI industry, it's symbolic: the era of massive, generalist crowdsourcing for training is giving way to more refined solutions, or AI training itself.

Many developers and researchers used MTurk for years to collect data or validate experiments. Now they'll have to look elsewhere, or perhaps use their own models to generate the data they need. Many developers and researchers utilized Mechanical Turk for data collection and experiment validation for nearly two decades.

It's a bit like when dinosaurs made way for mammals: a slow but unstoppable shift. The future of AI training will likely be less "human-dependent" in the strict sense, or require much more specialized skills.