How OpenAI, Founded to Keep Powerful A.I. Out of Corporate Hands, Got Into Bed With Microsoft
OpenAI was founded less than four years ago as a valiant stand against the idea that powerful new technology must inevitably be controlled by giant corporations.
The then non-profit’s goal: Develop artificial intelligence “the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return.”
Earlier this week, a very different OpenAI accepted a $1 billion investment from Microsoft, among the world’s most valuable technology companies. The deal gives OpenAI, which now has a for-profit arm, a huge war chest to compete against well-funded rivals like Google and Facebook, along with Chinese tech titans Tencent and Baidu.
As for that anti-corporate founding philosophy? It’s now a little more nuanced. And what about benefitting humanity? Well, yes, that’s still OpenAI’s mission, but Microsoft may also now make some money in the process.
As a non-profit, OpenAI had been endowed with $1 billion from a group of wealthy backers that included, most prominently, Tesla CEO Elon Musk. He was concerned about the potential for A.I. run amok, including “a fleet of artificial intelligence-enhanced robots capable of destroying mankind,” according to Bloomberg writer Ashlee Vance’s biography of Musk.
Musk also worried about one company owning the intellectual property for artificial general intelligence, or A.G.I. for short. A.G.I. means technology that can excel at a wide-range of disparate tasks like humans can, as opposed to today’s narrow A.I. that can do only one thing—play chess or recognize faces, for instance.
In its inaugural blog post, OpenAI echoed Musk’s worldview, saying: “As a non-profit, our aim is to build value for everyone rather than shareholders. We believe AI should be an extension of individual human wills and, in the spirit of liberty, as broadly and evenly distributed as possible.”
Despite OpenAI’s feel-good philosophy, some people are skeptical about whether the organization is the proper vehicle for democratizing the benefits of A.I. Ben Recht, a University of California at Berkeley professor who specializes in machine learning algorithms, is among those critics.
He’s concerned about the key role played by Musk and Sam Altman, the former president of famed startup incubator Y Combinator who co-founded OpenAI and now serves as the organization’s chief executive.
“Anyone who believed that Elon Musk and Sam Altman were forming something for the good of mankind with a non-profit were fooling themselves,” says Recht. “It’s this weird Silicon Valley vanity project.”
Since its founding, OpenAI has racked up some impressive achievements. It created open-source software that lets other researchers train and benchmark their A.I. systems. It invented a new class of software that learns from trial-and-error rather than from pre-existing data. In one of its best-known feats, it created five separate A.I.s capable of playing a complex video game called Dota2 as a team, so well, in fact, that it beat some of the world’s top human teams.
OpenAI also taught a robot hand to manipulate objects with unprecedented dexterity. Most recently, it created an algorithm that takes a sentence or two of human-composed text and then automatically generates long-passages of novel, but completely believable, text in the same style.
These breakthroughs, however, haven’t come cheap. Some have used huge amounts of computing power, costing serious money. (The computing bill for its Dota2 work, for instance, is estimated to have run into the millions.) And OpenAI has said it suspects the way to get to A.G.I. is to keep increasing the volume of data being used in its experiments, requiring ever more servers to train its algorithms.
Also expensive are the human brains behind the artificial ones—the top echelon of machine learning experts command mid to high six- or even seven-figure salaries. (Ilya Sutskever, the brilliant young machine learning expert OpenAI hired as its research director, was paid more than $1.9 million in 2016, according to OpenAI’s tax disclosures.)
Even with a $1 billion endowment, OpenAI felt it needed more money. The non-profit was at a distinct disadvantage compared to the corporate A.I. labs, many of which are part of companies with massive cloud computing businesses and therefore don’t have to pay market rates for server time.
What’s more, Musk, who had been the biggest initial donor, reduced his commitment to OpenAI in early 2018, stepping down from the organization’s board. OpenAI said Tesla’s increasing interest in commercializing A.I. applications presented Musk with a conflict-of-interest.
To compete with big corporations and startups with huge reserves of venture capital funding, OpenAI’s board made a fateful decision: if it couldn’t beat the corporates, it would join them. In March, Altman announced that OpenAI would create a separate for-profit company, called OpenAI LP, that would seek outside capital.
Then, this week, it landed Microsoft’s whopping investment. The exact terms of the deal were not disclosed and the tie-up left some observers scratching their heads. “Microsoft already has one of the most respected and venerable private research institutions in the world, Microsoft Research,” Recht says.
In response to emailed questions, Greg Brockman, the former chief technology officer at payments startup Stripe who is now OpenAI’s chairman, said Microsoft “owns a significant but not majority” stake in the for-profit entity. He added that Microsoft is entitled to a board seat on the non-profit too, but that it had not yet exercised this right.
Microsoft has said its Azure cloud computing infrastructure will become the preferred platform for OpenAI’s research, so some of its $1 billion cash investment may return to Microsoft in the form of cloud purchases. But the two companies also said they would work together on developing new A.I.-specific supercomputers.
Perhaps most importantly, Microsoft said it would become OpenAI’s preferred partner for commercializing any technology short of A.G.I. that OpenAI develops. Brockman says there’s no contradiction between this arrangement and OpenAI’s anti-corporate founding philosophy.
“OpenAI has the option but not the obligation to license any given technology,” he said. “The point of licensing some of our technologies is to fund our efforts to build AGI, so we can fulfill our mission. But if licensing some pre-AGI technology is contrary to the mission, we won’t do it.”
OpenAI has said that its first set of for-profit backers will have their profits capped at 100 times their initial investment, with any excess profit flowing back to the non-profit. It has said subsequent investors will have different profits caps. OpenAI says Microsoft is not part of the initial round, which included Khosla Ventures and a charitable foundation belonging to LinkedIn co-founder Reid Hoffman, who is also a co-founder of OpenAI. But it declined to disclose the level of Microsoft’s cap.
OpenAI says that A.G.I., if it were to ever create it, is certain to be worth many times the cap. So in the end, it says, most of the value would, as required by OpenAI’s charter, accrue to humanity at large. But in the meantime, OpenAI could make Microsoft, already among the world’s richest companies, much richer still.
More must-read stories from Fortune:
—How the government should spend Facebook’s $5 billion fine
—Cloud gaming is big tech’s new street fight
—Should companies bolster their cybersecurity by “hacking back”?
—FaceApp’s Russia link is the latest alarm in an ongoing digital red scare
—Equifax may owe you some money. Here’s how to get it
Catch up with Data Sheet, Fortune‘s daily digest on the business of tech.