Tens of billions of dollars has been put toward these AI efforts, and spending is expected to continue. There are already AI generated friends, code, and stickers out there, with much more to come.

“A freakishly huge deal for Meta”

Meta CEO Mark Zuckerberg envisions a world in which content on Instagram, for instance, produced by AI rather than just human creators. “There are both people out there who would benefit from being able to talk to an AI version of you, and the creators would benefit from being able to keep your community engaged and service that demand,” he said recently, while noting there will be a version of this sometime in 2024.

Churning out content directly, without relying on creators, is something of a holy grail for Meta, a former employee explained.

“This is a freakishly huge deal for Meta, all the way to the top of the company,” the person said. “Its future ideal state is having a ton of new content, all the time, that people enjoy, and it doesn’t have to pay creators for it.”

Losing a valuable liability shield

There is a wrinkle in this plan, though. By becoming the producers of content, Big Tech companies like Meta and Google may well lose the shield provided by Section 230 of the 1996 Communications Decency Act, which has long protected them from liability for what’s posted on their platforms. No matter how harmful the content, legally speaking these tech companies have just been the uninvolved host or “intermediary” of some of the worst humanity has to offer, so long as they attempt to moderate their platforms. Courts have reinforced this several times.

Generative AI models and tools, however, are developed, owned and operated by Big Tech companies. And these tools will soon be cranking out search results, social media posts, and other content that is absent any other author or creator, no intermediary to be found.

The major tech companies involved are already arguing their AI outputs are unique and should be considered new works. And industry experts who spoke to Business Insider agreed that direct outputs from generative AI tools developed, owned and managed by tech companies will not enjoy the same shield provided by Section 230.

“Plainly on the face of the statute, generative AI falls outside of it,” said Aziz Huq, a professor at the University of Chicago’s Law School who studies AI regulation.

Already, Meta is “discussing at a high level” the legal implications of generative AI, the person familiar with the company said. This person asked not to be identified discussing private matters. Representatives of Meta and Google did not respond to requests for comment. A representative of Microsoft declined to comment.

“The big players love and want to keep 230 for as long as they possibly can,” said Anupam Chander, a professor of law and technology at Georgetown Law and a visiting scholar at the Institute for Rebooting Social Media at Harvard University. Even though a company as large and profitable as Meta or Google could withstand years of legal attacks, generative AI could prove to be a liability minefield, because Section 230 “will not be available as a defense in most cases,” Chander explained.

“It could dramatically undermine their business, or even perhaps make some parts too risky,” Chander added.

Meta’s open-source solution

The person familiar with Meta noted that one possible way the company is hoping to avoid or at least delay future liability for generative AI content is by leaning on its open-sourcing of Llama, a large language model that is available for free for most developers.

Chander noted that this line of thinking has been broached in his legal circles very recently. Although he is “not persuaded that is an effective way to definitely avoid liability… it depends very much on the exact situation.”

If a company created an LLM and then handed it off completely and for free to developers, with no role in how it was used or for what purpose, that could avoid liability. Otherwise, by creating generative AI tools and providing the outputs, Chander sees few ways a tech company will be able to refute its role in generative AI content.

The new thing = new legal consequences

Huq thinks Meta and other tech companies won’t be able to squeeze generative AI content under the umbrella of Section 230, as currently written.

“Everything they’ve been doing so far has been as an intermediary,” Huq said. “But everyone recognizes this is a new thing, there has been an emphasis on the phase shift from deep learning to an LLM, the companies say it’s new. So, okay, then you can’t expect the same liability protection. There will be new legal consequences because it’s new.”

Blurring the line

Even before generative AI became the new tech craze a year ago, the line distinguishing an internet platform as a host rather than a participant had already “almost evaporated,” according Jason Schultz, a professor at NYU Law who leads its Technology Law and Policy Clinic and is a lead in the AI Now Institute.

Again, this blurring of the line has to do with AI. The technology has been used for several years now to run recommendation algorithms that are designed to learn and show what users want to see. It’s also used in content moderation, which is largely automated using AI-driven programs that are trained on what is good and bad content.

“The amount of intention and design behind something like ChatGPT is far more intense than something like early Twitter,” Schultz said. “Someone tweets something on Twitter, which shows it, and people will see it. That’s the classic 230 case.”

Generative AI is not that.

“These tools are not just passing content through,” Schultz said. “They are in fact generating content, or the LLM or image diffusion generator are producing the actual content. And they want to make you feel like you’re being spoken to by a human.”

Already, Schultz noted, courts all the way up to the US Supreme Court have looked at cases bringing up issues of recommendation algorithms, weighing whether the effort to grab users and boost engagement on a platform constitutes enough participation for tech companies to have some liability for alleged harms. The Supreme Court has yet to dilute Section 230 for tech platforms, yet recent cases showed “the more complex the tech has become, the more unsure the justices were” about what Section 230 shields, Schultz added.

“One way to look at an LLM is, yes, as a giant pass-through. The other way is to think of it as the actual content creator, because it comes out of the program feeling fresh and original,” Shultz said. “I’d guess most judges will see them as content creators.”

Are you a tech employee or someone with a tip or insight to share? Contact Kali Hays at [email protected], on secure messaging app Signal at 949-280-0267, or through Twitter DM at @hayskali. Reach out using a non-work device.