← POLITICO Tech
POLITICO Tech
Google’s query: How many inventors use AI?
00:17:42
Patent lawyers at Google had a question: How many of the company’s big, new ideas were developed using artificial intelligence? The patent world has been grappling for years with whether or not AI can be considered an inventor, and Google needed to know how the technology is being used today. On POLITICO Tech, Google’s head of patent policy, Laura Sheridan, joins host Steven Overly to explain the findings and why the company doesn’t think AI counts as an inventor.
Loading summary
Transcript26 lines
- [00:00]
A
Okay, I have to tell you, I was just looking on ebay, where I go for all kinds of things I love. And there it was, that hologram trading card. One of the rarest. The last one I needed for my set. Shiny like the designer handbag of my dreams. One of a kind. Ebay had it.
- [00:15]
B
And now everyone's asking, ooh, where'd you get your windshield wipers? Ebay has all the parts that fit my car. No more annoying, just beautiful.
- [00:26]
A
Whatever you love, find it on eBay. EBay. Things people love.
- [00:32]
B
Hey, welcome to Politico tech. Today's Thursday, December 12th. I'm Stephen Overlea. Patent lawyers at Google had a question they could not answer. How many of the company's big new ideas were developed using artificial intelligence? See, the patent world has been grappling for years with whether or not AI can be considered an inventor. And earlier this year, the Biden administration released new rules that state inventions made with AI can qualify for patents as long as humans made a significant contribution. But Google wanted to measure AI's contribution. Laura Sheridan leads patent policy at Google, and she tells me that a few months ago, the company began collecting data each time an employee submits an idea to be considered for a patent. One thousand submissions later, nearly one in five of them used AI in some capacity. And that number is likely to grow. But Google doesn't think AI should be considered an inventor. And on the show today, Laura tells me why. Here's our conversation. Laura, welcome to Politico Tech.
- [01:55]
A
Thank you.
- [01:56]
B
So before we dive into kind of these big picture questions about AI as an inventor, I'm curious to hear from you how this is playing out within Google. How many of your inventions today are just developed using some form or some degree of generative AI?
- [02:13]
A
Yeah, it's such a good question. And it goes back about a year and a half ago, I testified in Congress in the IP subcommittee of the Senate Judiciary Committee to talk about this very issue, looking at AI and inventorship and exploring how AI was being used as a tool in the innovation process. And that was kind of the early days of us looking at this issue and figuring out how our engineers were indeed using AI as part of the innovation process. So we updated our invention form. Basically, how it works at Google is we have a form that our inventors fill out to submit their invention to the legal team for review. So we added a question to that, specifically saying, was generative AI used to develop this invention? We added that this fall, and it turned out that as of today, 17% of the time generative AI is used to, to support the development of the invention. So we thought that was a really exciting stat. We were excited to watch that continue to grow and change over the coming months and years.
- [03:23]
B
And do you dig into kind of how generative AI is being used? I mean, if I put a query into Gemini, for instance. Right. Is that considered a use of generative AI as part of the inventing process at Google?
- [03:35]
A
That's a question that we're going to dig into now that we have this data. So basically, we've asked the question of the inventors and we've learned 17% of the time it's part of the process. But now we're going to figure out how it's being used as part of that process. But what I expect is it's a combination of things, a variation depending on what the goal of the engineer is. One way it can be helpful is you want to understand all the literature in a particular space. Inventing is all about building on what came before. So you might use generative AI, you might use Gemini to understand. Tell me about the literature in this space, help me process it so I can build from that and innovate with that in mind. Other ways, you know, you could say, I have this really cool invention. There's a specific feature. What are other ways I could carry that feature out and, you know, use it to kind of brainstorm? So we are going to explore that deeper as we look at each of these inventions and say, you know, go to the engineer and say, how did you actually do this? What tool did you use and how was it helpful? I think it's an open question and we'll have more on that as we, as we dig into these. These various inventions.
- [04:52]
B
Got it. I'm curious, have you seen an example so far, an instance you can sort of describe for us to give us a sense of how AI has been used in one of these inventions?
- [05:02]
A
Well, you know, I can go back to the hearing, you know, a year and a half ago, one of the examples I shared then, which is still continued to develop and be an example today, is the use of AI in the chip design process. This is a very specific model. It's something that's specific to chip design. And you would have the inputs you have based on what you want the chip to be like and the parameters of it, and you would use the model to help you with that development process. That was something I talked about then, and today we're still using generative AI in that way. It's called alpha Chip. So I think that's one of the more specific use cases. And then there's of course the broader are use cases where with, you know, Gemini, you can be using it to sort of iterate on more free form, brainstorming kind of activity.
- [05:54]
B
Got it. So I mean, this question about sort of AI as an inventor is something that the patent world has sort of been debating for a while now. Right. And the government, the Biden administration has really weighed in this year with some guidance on how it views this issue. Google, I know, has commented in the past that AI should not be credited as an inventor on patents. Where are your views on that today?
- [06:18]
A
The same. And I think the issue has been percolating for a while and it was an understandable one. I think the concerns stem from inventors were starting to use AI more in the innovation process. And the question came up, will this impact the ability to get a patent? And everyone wanted to understand the answer to that. So that's where the patent Office was coming from. That's where Congress was coming from, trying to understand, I think, get an insight into, from industry, how AI was being used and what that meant in terms of the ability to get a patent. And so our view a year and a half ago, our view today is that it is still very much being used as a tool. And we, in looking at our patents and the inventions that come into us from inventors throughout Google, we're not seeing a, an inventive contribution from AI being used as a tool. We're seeing, of course, that it's extremely helpful in the innovation process. But we are still not seeing anything that gives us concern with respect to inventorship questions.
- [07:27]
B
Because that's, I think, the question I have, which is I've heard this argument that AI is a tool, just like any technology is kind of a tool, right. To aid in some human process, invention or otherwise. And I think with generative AI, to me that seems true up to a point. Because, you know, if you're for instance, asking it to summarize the literature, you know, that's sort of A1 function. But then when you start to ask it to come up with ideas for X feature or help create a solution to some problem that you're encountering, to me that feels different. To me, that feels like you're asking it to come up with ideas of its own. And I wonder how you navigate that difference.
- [08:07]
A
Yeah, then that is, I think, sort of the origin of all these questions. So that is a really good way to kind of frame it. And you're Right. I mean, it is doing things that are sophisticated compared to tools that came before. And it's giving you information that you would have taken maybe years to get and now you're getting it in minutes. But in our experience in using it, it is not outputting an invention. It is outputting information that the inventor then uses to further the invention itself. So, you know, if it's giving ideas for materials that could be used, it's not saying, you know, this is the one and here is the invention based on it, it's giving ideas for it. You know, and same with, you know, constraints within a chip design. You still need to validate, you need to further tweak things and you're not getting sort of the fully baked output. You're not getting, in our view, anything close to that. That's been our experience so far. So it is definitely helpful and very helpful compared to some of the software tools we have been using before. But it is not outputting something that is the invention.
- [09:17]
B
Got it. One of the ways in which the government has weighed in on this question is they said if something is entirely generated or invented by AI, it cannot be patented. But you know, there's sort of this gray area where if AI contributes to some degree, but a human significantly contributes, then it can still qualify for a patent. And so I guess my question is, do we know yet where that line is and what more do we need to do to kind of define, you know, the human from the machine when it comes to inventing things?
- [09:52]
A
Yeah, we have something called a significant contribution test under the law where you're kind of looking at what the human's role was and making sure there was significance to it when it comes to the final invention that is resulting in the patent. And I think frankly, like any legal issue, there are always shades of gray and inventorship. I think our position when we were engaging with the patent office on this as well is inventorship is one of the murkiest areas of patent law. It is a challenge. It has been a challenge forever and it is always highly fact specific. And this is outside of any AI related questions. This is just inventorship in general. It has been hard. So I think actually we were pleased that the AI questions were actually bringing inventorship to the surface because we actually thought we are long overdue for any, any thinking on inventorship. These are where the fights happen that no one knows which way it's going to go one way or the other. If someone says, this is my invention, I should have been on this patent, those are the kinds of fights that no one can predict the answers to with two humans involved? I actually think it's a lot clearer when it comes to using AI as a tool. I think it is a far clearer question than it is or has historically been when we've had these inventorship fights before, especially because that's interesting. Yeah. And I'm sorry, I can geek out on this forever, Stephen.
- [11:24]
B
No, but it's. No, no, it is interesting because, like, I mean, everyone thinks AI is this big, you know, complexifying factor in so many things, including patents. So it is interesting to hear, like, actually these questions may be easier to resolve than the, like, human dispute questions.
- [11:39]
A
Yes. Yeah. And AI, I think from my patent policy standpoint, AI has done a really good job of bringing a lot of policy issues in the space to have conversations we just wouldn't have had about them. But for coming up in this context, inventorship is one. Prior art is another. That's another concept that we've been talking a lot about. Obviousness. There's a whole slew of just fundamental patent issues that AI has kind of created conversations around in this, frankly, much cooler context than patent policy issues are used to coming up. We're kind of the geek of the policy world, but it's made a lot of things that we needed to be talking about finally getting attention. My dad works in B2B marketing. He came by my school for career day and said he was a big roas man. Then he told everyone how much he loved calculating his return on ad spend. My friends still laugh at me to this day. Not everyone gets B2B, but with LinkedIn, you'll be able to reach people who do. Get $100 credit on your next ad campaign. Go to LinkedIn.com campaign to claim your credit. That's LinkedIn.com campaign. Terms and conditions apply. LinkedIn, the place to be, to be.
- [12:59]
B
You know, one question that sort of comes up again and again in different contexts is, does existing law kind of COVID artificial intelligence sufficiently or are new laws need it? Right. And I guess when it comes to patents, in your view, to sort of existing patent law kind of COVID everything and contemplate everything it needs to when it comes to AI, or do we need Congress to really weigh in here with some new regulations or new laws in this space?
- [13:29]
A
So, in my view, existing laws working well for AI innovation, where I think it's really important to be thinking about the question of existing law or whether changes are needed. There is a lot of importance in making sure the system is just Fundamentally working well. And so this is going to be a little bit weedy for a second, but bear with me. The Patent Office right now runs a deficit in terms of the money it takes to examine every patent that comes through its doors until eight years after a patent gets granted. And so this is not an AI question, this is just a general patent system. But it's making sure the resources are available upfront for the agency to be doing a good job up front so that it's able to do really efficient work and devote time to the really challenging aspects of deciding whether or not to grant a patent. And from an AI standpoint, there's time for training of patent examiners who that's the 9,000 person workforce at the Patent Office, making sure they have time to learn the state of the art when it comes to AI. Because the latest stat I have is that more than 50% of patent examiners are actually looking at AI related patent applications. And that number is huge and that's going to grow. So if you have more than half your workforce needing to understand at least the basics on when I look at this patent application, is this new, is this non obvious, what am I looking at? That takes time and that takes training. So I think all this takes more resources for the Patent Office to be able to do this. So this is not a law change so much as just looking at the workings of the agency and make sure it's operating at peak.
- [15:19]
B
Can you sort of explain for folks who are not as sort of steeped in this, why these questions about inventorship and patentability matter so much to Google? You know, why are these questions like, so pivotal for your business?
- [15:34]
A
Yeah, of course. I mean, I think first and foremost we have patents on AI. We have patents that have been developed with the support of AI as a tool in the process. And we don't want uncertainty around whether or not, you know, those are valid patent rights. So I think it is an issue of certainty for the system to understand where those boundaries are. For the system to work well, there needs to be certainty about what the rules of the road are. And so that has always just been fundamental to the patent system. The more we can sort of understand the dynamic and how the law is going to apply, the better off we are. And so for this, I mentioned it earlier, making sure there is confidence that when you are using these tools in the innovation process, which we're doing 17% of the time so far, that you're comfortable with your ability to still get a patent on that resulting invention, that's important to us. I think that's really important to a whole slew of industry as more and more businesses are making use of this technology for innovation.
- [16:42]
B
Well, Laura, thanks for being here on Politico Tech.
- [16:45]
A
Thanks so much.
- [16:48]
B
That's all for today's Politico Tech. If you enjoy Politico Tech, be sure to subscribe. And for more tech news, check out our newsletters, Digital Future Daily and Morning Tech. Our managing producer is Annie Reese. Our producer is Afra Abdullah. I'm Stephen Overlea. See you back here tomorrow.