Can JARVIS hold a patent?

With help from Derek Robertson

This week, Digital Future Daily is dedicating 5 days to the fast-moving landscape of generative AI and the growing conversation about how and whether to regulate it — from pop culture to China to the U.S. Congress. (Read yesterday’s edition on the collision course between AI and the music industry’s biggest stars here.)

Alexandria, VA — Today the National Inventors Hall of Fame Museum — headquarters of the U.S. Patent and Trademark Office — was packed with lawyers, academics and corporate executives arguing about whether AI systems should patent their creations.

Well, maybe not argue, exactly. This was the first of two “listening sessions” held by the USPTO to decide whether its rules governing who can patent an invention need to change to accommodate innovations made with the help of artificial intelligence.

The USPTO is now firmly at the center of this emerging debate. On Monday, the U.S. Supreme Court declined to review a case that challenged existing laws about human inventorship — leaving in place an earlier ruling by the U.S. Court of Appeals for the Federal Circuit that an inventor listed on a patent must be a natural person. The case involved two patents filed by Dr. Stephen Thaler, who listed an AI system called “DABUS” as the sole inventor of a food container and an emergency light beacon. The USPTO did not grant DABUS the patents, saying only humans can receive a patent.

But the open questions around AI-driven innovation have grown bigger than Thaler. And if the U.S. inventorship rules are going to change — the Patent Office will be the one to do it. And any changes to inventorship rules would trigger a chain reaction in the American innovation landscape — ideas hatched by a generative AI model, for instance, could belong in part to the AI model itself or to its creators, setting up a more complex web of intellectual property “ownership” in the future.

Back at the USPTO headquarters, surrounded by examples of groundbreaking human innovation, Dr. Bijan Tadayon, CEO of Z Advanced Computing argued that even the powerful generative AI models dominating headlines this year are essentially blackbox statistical machines that don’t really understand the innovations they produce. They require human guidance to come up with something novel enough to be worth filing a patent for. Multiple speakers at today’s listening session could agree that AI systems should remain, at best, co-inventors on a patent.

But Tadayon and others also noted that AI models — and the humans guiding them — are getting better, teeing up the ultimate question at the heart of the debate: When a machine conceives a new idea independently of its human creators, should it get the credit? And how would changing USPTO policies around human/AI inventorship impact the socio-economic incentives people need to keep innovating?

And to be clear, machines inventing things semi-independently from humans is not just a hypothetical — it’s reality. Novartis’ JAEGAR is one example of a generative AI model coming up with novel active compounds for new antimalarial drugs. Corey Salsberg, the head of IP affairs for Novartis laid out the biopharma company’s approach to drug discovery with the help of powerful AI models. “It’s a multi-stage process,” Salsberg told me. The base AI model often comes from outside universities or companies, he said, which pharmaceutical companies like Novartis then bring in-house for customization, which means training the AI model on proprietary compound libraries. And scientists still need to screen and test whatever novel active compounds the custom AI model spits out.

That’s a pretty long invention chain. And tracing human inventorship all the way up that chain under current law can be a flawed approach when trying to find out who should ultimately hold a patent. Salsberg said current law grants equal credit to all listed co-inventors — meaning that if AI model developers were listed as co-inventors, the patent process would reward the software developers who ultimately had no hand in the downstream process of creating a new drug or technology deserving a patent.

Lolita Darden, one of the speakers at today’s session, called on the USPTO to mandate disclosure whenever AI systems are used in the invention process. And lobbing the ball back to the patent office, Morgan Reed, president of the App Association, recommended that USPTO patent examiners be equipped with AI tools to screen an incoming “gold rush” of AI-generated patent filings.

As for how revenue from a lucrative, patented innovation should be distributed to its various inventors — whether they be human, AI, or simply humans who make powerful AI models — that would fall into the legal arena of patent licensing agreements, a USPTO spokesperson said.

For now, the USPTO is listening — and if today is any indication, IP stakeholders have a lot of feelings about how the issue of AI-enabled inventorship should be handled. And some USPTO officials at the listening session said Director Kathi Vidal has the appetite to resolve the human vs. AI inventorship question this year.

meta's new privacy tool

Meta has exclusively shared with DFD today that it’s adding a new tool to its Meta Quest app store: An “App Privacy” tab that will specifically identify what information and data an app’s developer can collect from users.

The tab will feature three categories of information: Basic info like one’s username and profile picture; sensor and device data like eye movement and facial expression tracking; and a list of which Meta policies the developer is subject to. I spoke yesterday with Rob Sherman, Meta’s VP & deputy chief privacy officer for policy, about how the new functions will work and how they play into the company’s efforts to keep younger users safe. That last issue dovetails with some senators blasting Meta’s recent decision to let those users into the nascent metaverse.

“We asked developers to programmatically request access to data, so as a part of running their app it has to call a specific API in order to get access to hand tracking, for example,” Sherman said, by way of explaining how the company will ensure data collection is transparent. “They have to specifically ask the operating system for permission to access [data]... so we’re able to ensure a much tighter loop between what the App Store listing is saying the developer wants to get and what they’re actually getting.”

In other words: The developers can’t access the data without triggering a disclosure in the new App Privacy tab. As Meta proposes a new version of the internet that will require massive amounts of highly specific data to operate, Sherman made the case that moves like this one are part of an overall strategy to make it safe for teens along with everybody else.

“When it comes to teens it’s a separate effort — there are teens using virtual reality today, and I think that’s on balance a good thing,” Sherman said. “One of the efforts that we’re trying to undertake as a part of that, beyond this specific app privacy effort, is to build additional protections including the parental controls that we’ve already announced, to make sure that teens have a safe and protected experience as they’re using our products.” — Derek Robertson

rnc x ai

The future is here — well, the future of political campaigns, anyway.

Timed to the announcement of President Joe Biden’s re-election campaign this morning, the Republican National Committee released its first 100 percent AI-generated ad attacking the president, as Axios’ Alex Thompson reported.

It’s somewhat unclear why this particular ad had to be AI-generated, it mostly featuring stock-like images of Biden and Vice President Kamala Harris interposed with scenes of generic urban and international unrest. (Although it does inure one against the classic campaign problem of using footage out of context that you really shouldn’t have.)

Its content, then, is less interesting than what it augurs: The first presidential election cycle that will feature the powerful generative AI tools that have dominated the past year of tech news. Deepfaked political images and voices: They’re not just for Overwatch jokes anymore. — Derek Robertson

Tweet of the Day

the future in 5 links

Stay in touch with the whole team: Ben Schreckinger ([email protected]); Derek Robertson ([email protected]); Mohar Chatterjee ([email protected]); Steve Heuser ([email protected]); and Benton Ives ([email protected]). Follow us @DigitalFuture on Twitter.

If you’ve had this newsletter forwarded to you, you can sign up and read our mission statement at the links provided.