The paradox of the Mechanical Turk

The original Mechanical Turk was an elaborate fraud from 1770: supposedly a chess-playing machine, but that actually had a human player within.

Centuries later, Amazon launched Amazon Mechanical Turk, whereby humans volunteer to perform difficult-to-automate services or “micro-jobs” known as Human Intelligence Tasks or HITs, charging generally very small amounts per unit processed. These types of jobs, such as writing video descriptions, answering surveys, sorting images, etc, provide so-called turkers, with a modest income.

At least that was the case until OpenAI launched ChatGPT in late November. Since then, a paradoxical phenomenon has occurred: a study shows that between 33% and 46% of turkers are using ChatGPT and similar algorithms. A job that was supposed to be difficult or impossible to automate and is consequently offered to humans, who in turn decide that it is better to automate it and carry it out by means of an algorithm.

The contradiction is obvious: these tasks were difficult to automate in the days before ChatGPT and generative algorithms, but now, given their maturity and easy access to these tools, the results obtained are comparatively very good.

The system’s days are now numbered: as the companies who need these jobs doing see that they can be carried out using algorithms, they will stop commissioning Amazon Mechanical Turk, highlighting the contradictions that the advance of technology often generates, while prompting some thoughts on the nature of work and the future.

Have you asked yourself what part of your daily work could be carried out by a Mechanical Turk? And by an algorithm?

(En español, aquí)