Stock photo of the New York Stock Exchange. Photo / 123rf

Babak Hodjat believes humans are too emotional for the stock market. So he’s started one of the first hedge funds run completely by artificial intelligence.

“Humans have bias and sensitivities, conscious and unconscious,” says Hodjat, a computer scientist who helped lay the groundwork for Apple’s Siri. “It’s well documented we humans make mistakes. For me, it’s scarier to be relying on those human-based intuitions and justifications than relying on purely what the data and statistics are telling you.”

Hodjat, with 21 patents to his name, is co-founder and top scientist of Sentient Technologies Inc., a startup that has spent nearly a decade-largely in secret-training an AI system that can scour billions of pieces of data, spot trends, adapt as it learns and make money trading stocks.

The team of technology-industry vets is betting that software responsible for teaching computers to drive cars, beat the world’s best poker players and translate languages will give their hedge fund an edge on Wall Street pros.

The walls of Sentient’s San Francisco office are dotted with posters for robots-come-alive movies such as “Terminator.” Inside a small windowless trading room, the only light emanates from computer screens and a virtual fire on a big-screen TV. Two guys are quietly monitoring the machine’s trades-just in case the system needs to be shut down.

“If all hell breaks loose,” Hodjat says, “there is a red button.”

Sentient won’t disclose its performance or many details about the technology, and the jury is out on the wisdom of handing off trading to a machine. While traditional hedge funds including Bridgewater Associates, Point72 and Renaissance Technologies have poured money into advanced technology, many use artificial intelligence to generate ideas-not to control their entire trading operations.

All the same, Sentient, which currently trades only its own money, is being closely watched by the finance and AI communities. The venture capital firm owned by Hong Kong’s richest man, Li Ka-shing, and India’s biggest conglomerate, Tata Group, are among backers who have given the company $143 million (NZ$199m). (Beyond trading, Sentient’s AI system is being applied to a separate e-commerce product.)

Trading is “one of the top 10 places that AI can make a difference,” says Nello Cristianini, a professor of artificial intelligence at the University of Bristol who has been advising Sentient. “A trading algorithm can look at the data, make a decision, act and repeat-you can have full autonomy.”

Sentient’s team includes veterans of Amazon, Apple, Google, Microsoft and other technology companies. They’re part of a small group in Silicon Valley using expertise in data science and the field of artificial intelligence known as machine learning to try and disrupt financial markets.

AI scientists typically have no interest in working for a hedge fund, says Richard Craib, who started the AI hedge fund Numerai. “But they may want to mess around with data sets.” Numerai’s system makes trades by aggregating trading algorithms submitted by anonymous contributors who participate in a weekly tournament where prizes are awarded in Bitcoin.

It recently raised US$6m from investors including Howard Morgan, the co-founder of the quant investment management firm Renaissance Technologies. “It’s entirely a data science problem,” Craib says.

Another company, called Emma, started a hedge fund last year based on an artificial intelligence system that can write news articles.

Hodjat of Sentient spent much of his career focused on the language-detection technology behind smartphone digital assistants. Several employees from his previous company, Dejima, went on to create Apple’s Siri. Rather than join, he chose to focus on advances in artificial intelligence. His career goals didn’t include finance, but he sees markets as one of the most promising applications for the technology. The vast amounts of publicly available data, along with stronger computers to analyse it for patterns, make the field an ideal fit. “That is the fuel for AI,” he says.

Sentient’s system is inspired by evolution. According to patents, Sentient has thousands of machines running simultaneously around the world, algorithmically creating what are essentially trillions of virtual traders that it calls “genes.”

These genes are tested by giving them hypothetical sums of money to trade in simulated situations created from historical data. The genes that are unsuccessful die off, while those that make money are spliced together with others to create the next generation. Thanks to increases in computing power, Sentient can squeeze 1,800 simulated trading days into a few minutes.

An acceptable trading gene takes a few days and then is used in live trading. Employees set goals such as returns to achieve, risk level and time horizon, and then let the machines go to work. The AI system evolves autonomously as it gains more experiences.

Sentient typically owns a wide-ranging batch of US stocks, trading hundreds of times per day and holding positions for days or weeks. “We didn’t impose that on the system,” says Jeff Holman, the company’s chief investment officer. “The artificial intelligence seems to agree with what you get from human intelligence that it’s better to spread your bets and have a more diversified portfolio.”

As impressive as Sentient’s technology appears, it’s hard to know if it works. The company says the AI system is beating internal benchmarks, but won’t disclose what those are. It shares little about the data used for the AI’s decision-making and isn’t profitable. The company plans to bring in outside investors later this year. Holman, a Wall Street veteran who joined last year, said the company is limited on what it can say by US Securities Exchange Commission rules restricting marketing by hedge funds that are raising money. “The platform is solid,” he says. “It doesn’t look like any other strategy I’ve seen.”

Anthony Ledford, the chief scientist at the US$19 billion hedge fund Man AHL in London, warns of putting too much faith in this branch of artificial intelligence without more evidence. Man AHL uses machine learning for a portion of its clients money, and Ledford is encouraged by the results. While the company is exploring a standalone machine-learning strategy, he says it’s too early to declare success. “There’s a lot of hype and promise,” Ledford says. “But when you actually ask people how many hundreds of millions dollars they are trading, many of them don’t come back with much at all.”

Little performance data is available about AI-focused hedge funds. One index that tracks 12 pools that utilize AI as part of its core strategies, called Eurekahedge AI Hedge Fund Index, returned 5 percent last year. That’s slightly better than the average hedge fund, but trailed the S&P 500.

Tristan Fletcher, who wrote his doctoral thesis on machine learning in financial markets and works for a hedge fund, says investors may be reluctant to turn over their money completely to a machine. “I know how conservative investors are and I know of no one who would put their money in a system that’s fully systematic,” says Fletcher. “Machine learning isn’t a panacea for everything. You need people who have literal thinking.”

Chen Chen worked for Sentient until 2012 as an analyst. In August 2011, while the company was still known as Genetic Finance and operating in secret, the system was trading when Standard & Poor’s downgraded the US’s credit rating for the first time, sending markets into turmoil. Chen, now a trader in Hong Kong, went back to review the system’s performance for the day and saw it had outperformed the market and made money. He notes that there was no precedent for a US downgrade and wonders how Sentient’s AI system made its decisions. “That was a big surprise,” he says.