A Modern Mechanical Turk Is Revealed: Lessons from the Past about the Limits of Artificial Intelligence

The Original Illusion: A Machine That Conquered Kings

A strange machine showed up in the royal court of Vienna in 1770. Wolfgang von Kempelen, a Hungarian engineer, went to a magic show for Empress Maria Theresa of Austria and boldly told her that he would come back with an invention that would be better than anything she had ever seen. He delivered six months later.

He showed off a life-sized wooden figure dressed like an Ottoman with a turban and sitting behind a big cabinet with a chessboard on top. It had a long smoking pipe in its left hand, and its right hand was close to the chess pieces. Before each show, Kempelen would ceremoniously open the cabinet's doors one by one, even shining a candle inside to show off a complex maze of gears, rods, and clockwork. This assured the audience that nothing was hidden inside. He would then use a key to wind the machine, and the automaton would come to life. It would turn its mechanical head from side to side as if it were looking at the board, then reach across to grab a piece and make its first move.

People quickly fell in love with the Mechanical Turk. It beat almost every opponent who dared to sit across from it, and for an amazing 84 years, it was shown in Europe and the Americas. Its list of defeated foes reads like a who's who of history: Benjamin Franklin played it in Paris in 1783 while he was the American ambassador to France and lost. The Turk beat Napoleon Bonaparte in the Wagram campaign in Vienna in 1809. The famous chess master François-André Philidor, who is widely thought to have been the best player of his time, was only able to beat the machine after what his son called his father's most tiring game of chess ever.

Of course, there were a lot of skeptics. In 1784, British writer Philip Thicknesse wrote a passionate pamphlet saying that the cabinet must be hiding a child chess genius. Some people thought about magnets, hidden wires, or even trained monkeys. In 1836, Edgar Allan Poe wrote an essay that tried to break down the illusion. But while the machine was still around, no one could prove for sure how it worked.

As it turned out, the truth was very simple. A human chess master was hiding in the cabinet, crammed into a cleverly designed sliding compartment that kept him hidden no matter which door the showman opened. He kept track of the game on a small chessboard, watched the other player's moves through magnets, and moved the Turk's arm with a series of levers. After Kempelen died in 1804, the Bavarian showman Johann Nepomuk Maelzel bought the machine. He added theatrical touches, like a mechanical voice box that let the Turk say "Check!" Over the years, some of the best chess players in Europe, like Johann Allgaier, Aaron Alexandre, William Lewis, and William Schlumberger, secretly ran the machine from the inside.

The Turk was finally put to rest and given to the Chinese Museum of Charles Willson Peale in Philadelphia. It burned down in 1854. It wasn't until 1857 that the whole secret came out. The son of one of the last owners wrote a detailed article about it in The Chess Monthly.

But the Mechanical Turk's legacy lived on after it was gone. Charles Babbage, an English mathematician, saw the automaton play in 1819. He quickly saw through the trick, but the meeting made him think about whether it was possible to make a machine that could play chess by itself. Babbage went on to make the Difference Engine, which is often thought to be the first computer. The biggest fraud in chess history, in a roundabout way, helped start the real development of artificial intelligence.

Two hundred and fifty years Later: The Turk Comes Back

Two and a half centuries after Kempelen's invention wowed the courts of Europe, the same trick was done again, but this time on our screens instead of in palaces. The same trick was performed using lines of code instead of chess pieces. This time, the name of the machine was Builder.ai.

British-Indian entrepreneur Sachin Dev Duggal, who called himself "Chief Wizard," and co-founder Saurabh Dhoot started the London-based startup in 2016 under the name Engineer.ai. The company promised to make software development more accessible to everyone. The pitch was too good to pass up: anyone, no matter how tech-savvy, could make a custom app just by telling "Natasha," the company's AI-powered virtual project manager, what they wanted. The company said that ordering the product was like ordering a pizza: you pick the features, confirm the order, and the AI puts it together.

The tech world was very interested. I got to meet the Builder.ai team at the Web Summit in Lisbon in 2022. It was Europe's biggest tech show, and I was impressed, just like so many others. The product demos were smooth, the messages were sure, and the support was amazing. By 2023, Builder.ai had raised more than $450 million from a group of big-name investors. Microsoft made a strategic investment and planned to add Natasha to Microsoft Teams, which would give Builder.ai access to 280 million active users every month. The Qatar Investment Authority led a $250 million Series D round. Other investors included SoftBank's DeepCore, Insight Partners, Jungle Ventures, and the International Finance Corporation (IFC). The company's value skyrocketed to around $1.5 billion, making it one of Europe's most famous AI unicorns. According to Fast Company magazine, it is the third most innovative AI company.

But, like the first Turk, there were people behind the impressive facade.

When Intelligence Isn't Fake

As early as August 2019, The Wall Street Journal published an investigation that showed that Engineer.ai (as it was then known) relied mostly on human engineers instead of AI for its coding work. This was the first sign that the illusion was starting to break. In February 2019, Robert Holdheim, a former Chief Business Officer who was hired in late 2018, sued the company and Duggal, saying the operation was "smoke and mirrors" and that he had been fired for pointing out the problems. He said that Duggal told investors that the AI platform was 80% finished when it was really just starting to be built.

Duggal didn't deal with the lie that was at the heart of the problem; instead, he rebranded. The company changed its name from Engineer.ai to Builder.ai and started using the euphemism "Human-Assisted AI." This name didn't do justice to the fact that the company's core operations didn't actually have any real artificial intelligence. This semantic pivot was successful. Even after the Wall Street Journal story came out, Builder.ai went on to raise hundreds of millions of dollars in the years that followed.

It wasn't until the company went bankrupt in 2025 that the full extent of the fraud became clear. Bloomberg and the Financial Times reported that internal documents and testimonies from former employees painted a very bad picture. There were about 700 engineers behind Natasha's polished chatbot interface, most of whom worked in offices in Noida and Bangalore, India. They wrote the code for each and every client project by hand. The "AI" was basically a chatbot that gathered requirements. It got information from the customer, and then people did all the work.

Former employees talked about a complicated theater that was built to keep up the illusion. Engineers were told to never tell anyone where they were or use Indian English phrases when talking to people. They made sure that their updates happened during UK business hours so that clients would think they were dealing with a real-time automated system. Employees at the company made a dark joke about Natasha's real identity. They called the AI assistant "A Guy Instead."

Investors saw presentations that suggested full automation. Customers paid a lot for services that were, in the end, the same as traditional outsourced software development. Later, internal Slack messages showed that executives told employees to make sure that human labor wasn't very visible in materials for investors. An internal memo from 2022 reportedly told employees that the story should focus on proprietary AI and not on human labor.

Accounting tricks and billions of dollars in fake income

It turned out that the technological lie was only the beginning. Builder.ai wasn't just pretending to have AI; it was also pretending to have money.

When Duggal asked the board for more money in late 2024, after the company had already gotten an emergency loan from the creditor group Viola Credit, people started to worry. The board ordered a round of due diligence that changed the revenue numbers by a lot. Sales estimates for 2024 that were previously $220 million were cut to about $55 million, which is a shocking 300% increase. The $180 million figure for 2023 was also changed to just $45 million.

The internal audit also found a complex "round-tripping" scheme involving India's VerSe Innovation, the parent company of the news aggregator Dailyhunt. The two companies sent each other almost the same invoices for millions of dollars' worth of services that were either never done or couldn't be checked between 2021 and 2024. Builder.ai would charge VerSe for development services and then pay VerSe back the same amount for marketing services. This circular flow of money let Builder.ai make its revenue look higher than it really was, making it look like it was growing quickly and attracting new investment.

The investigation also raised questions about the legitimacy of sales made through resellers, especially in the Middle East. Auditors couldn't be sure that a lot of these resellers were real businesses, and in some cases they couldn't even find out who the end customers were. There was money on the books for sales that were never made.

The company made things worse by not having a Chief Financial Officer for ten months, which was the most important time for growth and fundraising. This was a clear failure of governance that let financial problems go unpunished. The company also hired auditors who were friends with CEO Duggal, which made it even harder for independent oversight to happen.

The board had seen enough by February 2025. Manpreet Ratia, an executive from investor Jungle Ventures, took over as CEO after Duggal was pushed out. But it was too late to save the business. An emergency funding round of $75 million did not go through. In April 2025, Bloomberg published a shocking report about the round-tripping scheme. Then, in early May, Viola Credit, the main lender that had given Builder.ai a $50 million loan in 2023, took $37 million from the company's accounts. This left the company with only $5 million in restricted funds, which made it impossible for it to do business.

A few days later, the US Attorney's Office for the Southern District of New York sent out subpoenas asking for customer lists, internal documents, and accounting records. The company's General Counsel told everyone to keep all of their papers. Builder.ai officially filed for bankruptcy by the middle of May 2025. About 1,000 workers, or about 80% of the company's global workforce, lost their jobs. Customers in the UK, US, UAE, Singapore, and India were left with products that were not finished or that had been turned off.

The fallout kept getting worse. Builder.ai had to pay Amazon Web Services $85 million and Microsoft $30 million. The Qatar Investment Authority lost most of its $250 million investment. In August 2025, FBI agents gave a subpoena to the company's former CFO, Andres Elizondo, at an airport in the Dallas area. The subpoena asked for communications related to possible violations of wire fraud, securities fraud, and conspiracy laws. The SEC also started looking into whether Builder.ai lied to investors about what its technology could do. The Enforcement Directorate was also looking into Duggal, the founder of the company, for possible money laundering related to the 2018 bankruptcy of the electronics giant Videocon. Duggal had moved to Dubai after being fired.

A Lesson from an Old Story for Today

The story of Builder.ai is a powerful modern parable that reminds us that history tends to repeat itself, especially when greed, hype, and the desire to believe in magic come together.

The first Mechanical Turk took advantage of the 18th century's interest in automatons and the Enlightenment's belief in the power of machines. Builder.ai took advantage of the 21st century's version of the gold rush: the crazy rush for artificial intelligence, where the two letters "A.I." at the end of a company name can open doors, unlock funds, and stop important scrutiny. In the business world, this is known as "AI washing," which means marketing old technology services as cutting-edge AI to get more money and higher prices.

Builder.ai wasn't the only one. Industry experts say that a large number of companies that say they can use AI are greatly overstating what their technology can really do. But Builder.ai is probably the most extreme and important example so far. The biggest AI startup failure since ChatGPT's launch in November 2022 set off the current investment frenzy.

The similarities to the Mechanical Turk are almost scary. In both cases, a charismatic showman (Kempelen, then Duggal) put on a complicated show to make people believe in a machine that could work on its own. In both cases, the secret was hidden labor: people were hidden behind the machines, doing the real work while the audience thought they were seeing technological magic. Skeptics sounded the alarm early in both cases (Thicknesse in 1784 and The Wall Street Journal in 2019), but those warnings were mostly ignored. And in both cases, the truth came out in the end, not through one big revelation, but through a slow buildup of evidence that the lie couldn't stand up to.

The Turkish robot beat Napoleon, Benjamin Franklin, and many others before it was finally found out. Before its box was opened—not with a physical key, but with a flood of whistleblower reports, unpaid debts, restated accounts, and criminal investigations across three continents—Builder.ai tricked Microsoft, the Qatar sovereign wealth fund, SoftBank, and thousands of customers.

There was no real AI behind either of the machines. There were people who were invisible, tired, and often working in terrible conditions, but they were necessary. AI still can't do everything, even though it has made amazing progress in the last ten years. And if someone says it can, it might be a good idea to ask, "Who's inside the box?"