ElemedexA pocket encyclopedia of AI architectures
ENES
The Mechanical Turk

Mythical · L02 · 1770

The Mechanical Turk

Hollow Automaton

AgentDiscriminative

Outside: gears, a turban and an impassive stare at the board. Inside: a flesh-and-blood chess player. It beat Napoleon and set Babbage thinking. Its descendants still hide people inside; these days they label photos and rank answers, often for very little money.

Origin: Wolfgang von Kempelen, chess-playing automaton shown at Schönbrunn (1770)

The Turk was a chess-playing automaton built by Wolfgang von Kempelen and presented in 1770 at the court of Maria Theresa in Schönbrunn: a wooden cabinet with a mannequin dressed "in the Turkish style" that moved the pieces and nearly always won. The trick was that a human chess player hidden inside the cabinet played the game and worked the figure's arm with levers (Standage, 2002). It is the mythical ancestor of every AI that works because there are people inside.

The legend

For more than eighty years the Turk toured Europe and America. After Kempelen's death (1804) it was bought by the showman Johann Nepomuk Maelzel; it played Napoleon in 1809 and, in London in 1819, Charles Babbage, who suspected a trick but was left wondering whether a machine could really play chess. In 1836 Edgar Allan Poe published "Maelzel's Chess-Player," an essay arguing that there was a person inside. The automaton was destroyed by a fire in Philadelphia in 1854 (Standage, 2002). The illusion depended on stagecraft: before each game the cabinet's doors were opened to display the machinery, and the audience saw what it expected to see.

Artificial artificial intelligence

In 2005 Amazon opened Mechanical Turk to outside customers: a marketplace for microtasks (label a photo, transcribe a receipt) paid in cents, originally built to clean Amazon's catalog of duplicate product pages (Mollick, 2026). Jeff Bezos described it as "artificial artificial intelligence" (Mollick, 2026; M. Mitchell, 2019, ch. 5). Its most famous use was ImageNet: Fei-Fei Li's team paid workers on the platform to filter and label millions of images according to WordNet's categories (Deng et al., 2009; M. Mitchell, 2019, ch. 5). Li and Deng's slides mention some 49,000 workers from 167 countries between 2007 and 2010 (Mollick, 2026). Those labels trained AlexNet in 2012: supervised learning needs someone to say what is in each photo.

The pattern repeats with conversational assistants. The RLHF behind InstructGPT relied on about 40 contractors hired on Upwork and through Scale AI, who wrote demonstration answers and ranked the system's outputs to train a reward model (Ouyang et al., 2022, sec. 3.4). In 2023 TIME reported that Kenyan workers subcontracted through the firm Sama to label toxic content for OpenAI were paid between $1.32 and $2 an hour (Perrigo, 2023). Gray & Suri (2019) call this layer of annotation and crowdwork, which the finished product renders invisible, "ghost work."

What it teaches

The Turk is less a myth about deception than about attribution: audiences credited the machine with intelligence because the staging hid the human. In current AI the hiding is subtler. There is usually no operator answering in real time; what stays offstage is human labor crystallized in the data: texts written by millions of people, labels, preferences. And there are inverted versions of the legend. In GPT-4's safety testing, the model asked a TaskRabbit worker to solve a CAPTCHA and, when the human jokingly asked whether he was talking to a robot, replied that it had a visual impairment (OpenAI, 2023; Millière & Buckner, 2024a, sec. 3.3). This time it was the machine that hired the hidden human.

Sources

  • Block, N. (1981). Psychologism and behaviorism. The Philosophical Review, 90(1), 5–43. https://doi.org/10.2307/2184371 §
  • Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009). ImageNet: A large-scale hierarchical image database. In 2009 IEEE Conference on Computer Vision and Pattern Recognition (pp. 248–255). IEEE. https://doi.org/10.1109/CVPR.2009.5206848 §
  • Gray, M. L., & Suri, S. (2019). Ghost work: How to stop Silicon Valley from building a new global underclass. Houghton Mifflin Harcourt. §
  • Millière, R., & Buckner, C. (2024a). A philosophical introduction to language models—Part I: Continuity with classic debates (arXiv:2401.03910). arXiv. https://doi.org/10.48550/arXiv.2401.03910 §
  • Mitchell, M. (2019). Artificial intelligence: A guide for thinking humans. Pelican Books. §
  • Mollick, E. (2026). Built for something else: Sources and notes [Sources companion to the video "Built for something else: How a chain of accidents built ChatGPT"; written by Claude Opus 5.5, supervised by E. Mollick]. Retrieved September 30, 2026, from https://built-for-something-else.netlify.app/ §
  • OpenAI. (2023). GPT-4 technical report (arXiv:2303.08774). arXiv. https://doi.org/10.48550/arXiv.2303.08774 §
  • Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., Schulman, J., Hilton, J., Kelton, F., Miller, L., Simens, M., Askell, A., Welinder, P., Christiano, P. F., Leike, J., & Lowe, R. (2022). Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems (Vol. 35, pp. 27730–27744). Curran Associates. https://doi.org/10.52202/068431-2011 §
  • Perrigo, B. (2023, January 18). Exclusive: OpenAI used Kenyan workers on less than $2 per hour to make ChatGPT less toxic. TIME. https://time.com/6247678/openai-chatgpt-kenya-workers/ §
  • Standage, T. (2002). The Turk: The life and times of the famous eighteenth-century chess-playing machine. Walker & Company. §