Meta, Long an A.I. Leader, Tries Not to Be Left Out of the Boom

SAN FRANCISCO — Two weeks before a chatbot called ChatGPT appeared on the internet in November and wowed the world, Meta, the owner of Facebook, WhatsApp and Instagram, unveiled a chatbot of its own.

Called Galactica, it was designed for scientific research. It could instantly write its own articles, solve math problems, generate computer code and annotate images.

Like ChatGPT, Galactica also played fast and loose with facts, making up mathematical proofs, misstating historical dates and spinning tall tales. One user coaxed the chatbot into talking about the history of bears in space. When asked who runs Silicon Valley, Galactica replied, “Steve Jobs.”

But unlike OpenAI, the tiny San Francisco lab that made ChatGPT, Meta encountered an avalanche of complaints about Galactica’s mishaps. After just three days, the company, which has faced scrutiny for spreading misinformation and hate speech through its social networking apps, removed Galactica from the internet.

“The people who made the demo had to take it down because they just couldn’t take the heat,” Yann LeCun, Meta’s chief artificial intelligence scientist, said last month during an online appearance at Collective[i] Forecast, a gathering of Silicon Valley leaders and thinkers.

For nearly a decade, Meta has spent billions of dollars building new kinds of A.I. Mark Zuckerberg, the chief executive, made it a mission for Meta to become a leader in the field back in 2013. The company hired hundreds of top A.I. researchers, including Dr. LeCun. It spent hundreds of millions of dollars on the large amounts of computing power needed to build A.I. systems.

Yet Meta has been left out now that Silicon Valley is gripped with excitement by “generative A.I.,” the name for technologies that generate text, images and other media on their own. OpenAI has taken center stage, even though Meta and many other companies have built similar technologies.

Others have since jumped headlong into the frenzy. On Monday, Google said it would soon release an experimental chatbot called Bard. And on Tuesday, Microsoft, which has invested $13 billion in OpenAI, is holding an event where it is expected to introduce a version of its Bing search engine that uses chatbot technology.

Meta, however, was hamstrung, in part, by its reputation as a corporate giant that helps spread untruths, Dr. LeCun said last month. And with responsibilities to billions of users, it could not afford to leave online a chatbot that can generate false and biased information.

“OpenAI and other small companies are in a better position to actually get some credit for releasing this kind of thing,” said Chirag Shah, a University of Washington professor who has explored the flaws in technologies like Galactica and ChatGPT. “They are not going to get the same kind of blowback.”

In recent years, Meta has also shifted its focus to another technology area: the immersive online world of the so-called metaverse, which Mr. Zuckerberg has said he believes is the next big thing. In the short term, it is unclear how the company can offer generative A.I. products with its existing services in a way that really captures the public’s attention.

That does not mean it isn’t trying. Meta is fast-tracking its efforts to put A.I.-driven products into customers’ hands, said Irina Kofman, a senior director of product management for generative A.I. who oversees XAI, a new team that aims to help build A.I. products across the company. Mr. Zuckerberg is directly involved in steering the initiatives, holding weekly meetings with product leaders and top A.I. researchers, she said.

In a call last week with investors, Mr. Zuckerberg repeatedly mentioned A.I. He called it “the foundation of our discovery engine and our ads business” and added that it would “enable many new products and additional transformations within our apps.”

Some Meta executives have spoken about the generative A.I. boom with a distinct air of sour grapes. During his online discussion last month, Dr. LeCun described ChatGPT as “not particularly innovative” and “nothing revolutionary” because it relied on technologies developed and deployed by Meta, Google and other companies.

Earlier last year, Meta released a chatbot, BlenderBot, that stretched the state of the art, Dr. LeCun said. But it never caught on, he said, because the company had worked hard to ensure that it would not produce offensive material.

“It was panned by people who tried it,” he said. “They said it was stupid and kind of boring. It was boring because it was made safe.”

Not long ago, Meta was known for rapidly pushing out products that were not fully tested or bulletproof, preferring to figure out the details along the way. One of its mottos — “move fast and break things” — became an anthem of Silicon Valley start-ups in the early 2010s. And Mr. Zuckerberg embodied what is known as “hacker culture,” embracing new technologies and preferring to throw small teams on them to develop them as soon as possible.

But after the media, the public and lawmakers spent years scrutinizing the company for allowing false and inappropriate material on Facebook and its other social platforms, Meta now cannot release a chatbot that generates misinformation without significant criticism.

“People tend to hold large tech companies to a high standard,” said Andrew Ng, a researcher and an entrepreneur who previously oversaw the A.I. labs at Google and the Chinese internet giant Baidu. “It does create an uneven playing field where smaller companies can move faster.”

Mr. Zuckerberg began delving into A.I. in 2013, just after Google made a big bet on A.I. research. He personally recruited top A.I. academics like Dr. LeCun during a hiring frenzy, which culminated with Mr. Zuckerberg flying to Lake Tahoe, Nev., that year to offer millions of dollars in salary and stock to A.I. researchers who had gathered for a conference. For years, many of them sat next to his desk at Facebook’s headquarters in Menlo Park, Calif.

The researchers eventually helped develop A.I. technologies that Meta largely uses today behind the scenes to help target ads, recommend posts and videos to users, and identify misinformation and other problematic content.

Now Meta faces the problem of how to turn A.I. into a product. Facebook offered cutting-edge facial recognition technology in the late 2010s, but removed it after complaints that it compromised people’s privacy. It also offered A.I. technology that could instantly translate social media posts from one language to another, which ended up as a narrow use.

In the call with investors last week, Mr. Zuckerberg said A.I. underpinned many of Facebook’s and Instagram’s most important features, including showing people Reels videos and suggesting other photos, videos and posts that they might not already follow.

Asked at last month’s online appearance how Meta might deploy chatbots and other generative A.I. technologies, Dr. LeCun said they could help small businesses create ads on Facebook. He added that once people moved into the metaverse, they would need generative tools to create virtual items.

The pace of development has taken off internally in recent months, Ms. Kofman said.

Mr. Zuckerberg and Meta’s chief product officer, Chris Cox, and chief technology officer, Andrew Bosworth, have been in weekly team meetings with leaders across A.I. product teams. Over the past few months, they have also spun up teams that specialize in turning A.I. into products, working with others across Meta’s family of apps, Ms. Kofman said.

The goal is to get A.I. powering products faster, and to put those directly into the hands of users. Ultimately, Mr. Zuckerberg hopes to incorporate some of that underlying technology into building out his vision of the metaverse.

How he will bridge the gap between the two important technologies remains unclear. But he has made generative A.I. a top company priority.

“One of my goals for Meta is to build on our research to become a leader in generative A.I.,” Mr. Zuckerberg said on the investor call.