Welcome to Skilled in the Art. I'm Law.com IP reporter Scott Graham. On Monday, the USPTO posed some intriguing questions about artificial intelligence and the patent system and made a formal request for answers. I've convened a Skilled in the Art panel of experts to take a quick stab at four of them (with light edits for length and clarity). I've also got a couple of amplifications from last week's briefing. As always, you can email me your thoughts and follow me on Twitter.

From left, John Dragseth, Ryan Abbott and Kathi Vidal

Survey Says: AI Edition

Academics have been debating for awhile whether machines can be inventors for the purposes of patent law. Earlier this month, University of Surrey IP professor Ryan Abbott and others upped the ante, forming the "Artificial Inventor Project" and filing patents around the world that list an AI machine as the inventor.

The USPTO, which convened a conference earlier this year on AI and IP, is now formally requesting comments from the public on patenting artificial intelligence inventions. PTO Deputy Director Laura Peter publicized the request in a blog post Monday, highlighting four sample questions the agency intends to address. I've asked four experts, including Abbott, for some off-the-cuff answers. As you'll see, consensus could prove tricky to achieve here.

PTO QUESTION 1: Do current patent laws and regulations regarding inventorship need to be revised to take into account inventions where an entity or entities other than a natural person contributed to the conception of an AI invention or any other invention?

John Dragseth | Fish & Richardson | Senior Principal: I think the current law is flexible enough to handle "machine-heavy" inventions. The law simply asks what humans made a meaningful contribution to the idea of the invention—that's plenty flexible, as long as we assume there has to be a human inventor somewhere. We are going to end up with situations where the "invention" occurs long after the inventor completes her work—maybe even after she dies. Out-of-context, an already-dead inventor sounds silly. But if you understand that she's getting credit for the computer system she set up—and as long as we make sure there's proximate cause between her contribution and the invention—all should be good.

The example I always use is an AI system for mixing and selecting plant or drug candidates in massive numbers, where someone or someones smart build the system (probably a biotech person and a computer science person), and then lab techs watch the system to select the candidates. If the selections are "dumb," e.g., select any plant that is 3 inches tall or taller after a week, then the first people are the co-inventors—even if they are long-gone. If the selections are "smart" and require judgment, then someone needs to figure out if the lab tech should also be an inventor. It's a very flexible legal framework, and rightly so, because there are a million tricky issues on inventorship that go beyond AI.

Side note: The PTO's question refers to what happens when "an entity or entities other than a natural person contributed to the conception of an AI invention." That can't happen, at least not for a while. Machines cannot have ideas, so they don't contribute to an invention. Even if you want to stretch the concept of what an "idea" is, we already have doctrines holding that one person's work can "inure to the benefit" of another person—so it wouldn't be too hard to make the work of a machine inures to the benefit of the person whose ideas put the machine in motion.

Ryan Abbott | University of Surrey:  Yes, it isn't clear when an AI meets traditional inventorship criteria, rather than a person, whether a patent can subsist. Also, who or what would qualify as an inventor? Whether an AI could be considered an inventor, or an AI's user, developers, or owners? Who would own any resulting patent rights? Whether an AI could be a "co-inventor," and so forth.

Yet, there have been claims of this occurring, and AI-generated inventions even getting protection for decades. I wrote an article on this.

Kathi Vidal | Winston & Strawn | Managing Partner for Silicon Valley:  Though any proposals or changes to the law will have to withstand, at some point, Constitutional-level scrutiny including on what our Founding Fathers meant by empowering Congress to secure for limited times exclusive rights to "inventors," the bigger issue in my view is whether expanding the definition of "inventor" to include AI would advance the Constitutional objective of "promot[ing] the progress of . . . [the] useful arts."

Or, on the other hand, would allowing for the patenting of inventions conceived by AI (which AI could do with relative speed and ease and little expense) threaten to preempt fields and lock-up technology in such a way that stunts progress? There is enough of a divide between the incentives needed to encourage innovation in the bio and pharma space (just read the Federal Circuit's recent Athena v. Mayodecision) compared to the electrical, computer and mechanical arts. Adding AI to the mix may stretch our "one-size-fits-all" patent system to its breaking point.

Rich Martinelli | Orrick, Herrington & Sutcliffe | Partner: I think accepting inventions made by AI systems would actually require a fundamental re-think of our entire patent system.

There are two general justifications. One, it is a system that provides a limited time monopoly to incentivize investment in solving hard problems for the long term benefit of society as a whole. Two, the less popular idea that there is some sort of moral right in owning the outputs of your intellectual efforts.

Since we are a long way from granting general personhood to AI systems, the second justification does not work at all for AI created inventions. And, because the whole point of AI is a tool to solve hard problems more easily, we cannot guarantee that the economic balance struck by the patent system is fair for AI-created inventions. If the goal of AI systems is to make problem-solving easier and easier, we may reach a point where inventions can be made simply by stating the problem to the AI system and requesting a solution. Historically, merely stating a problem has not been sufficient effort to secure a patent. It probably also should not be sufficient for a 20-year monopoly when a machine can do the rest of the work for you.

If the work of using AI to create a solution is hard, then we should probably just call the person that did the hard work the inventor. If this were the case, AI would be somewhat analogous to the problem currently facing the patenting of diagnostics in the life sciences space. Namely, it is time consuming and expensive work to isolate biological markers for diseases, but the process is routine and uses industry standard techniques, so it does not appear particularly inventive. But, no one is trying to name a gene sequencing machine as an inventor.

PTO QUESTION 2: Are there any patent eligibility considerations unique to AI inventions? 

Dragseth: Generally, under Section 101, you are looking for particular physical inputs or particular physical outputs for the mental or computing work performed by a person (e.g., in Mayo) or a computer (e.g., in Alice). AI inventions are often carried out entirely inside the computer, so that can create some issues. But it's not impossible, as the Federal Circuit has found "inside the computer" inventions eligible when they do cool enough stuff.

Abbott: Well, [as I described] above, it is not clear that something without a traditional human inventor could receive patent protection. Perhaps some of the time if someone recognizes the inventive subject matter of AI output, but that isn't clear if its inventive nature is obvious (drug X treats condition Y) and the user didn't otherwise contribute to the invention's conception. Especially not if the AI can identify its own output as salient.

Vidal: The judicially-created exceptions to patentability are premised in some respects on a fear of locking up technology or preempting a field. We don't allow for the patenting of mathematical formulas, because doing so would preempt every use of that formula. The issue with AI inventions is not that any given invention would preempt a field, but that the speed with which AI could invent raises similar concerns. For this reason, patent eligibility, like other aspects of our patent system, needs to be carefully rethought.

Martinelli: I do not know that there are eligibility considerations arising from AI created inventions per se. If an AI creates an optimized piece of hardware, the invention could be described in a utility patent with little indication that the source of the idea was an AI.

That being said, AI technology could easily create the same sort of patents that led to many of the software eligibility problems. Many eligibility cases arise in the software/business method space because the enablement part of the invention is trivial. In other words, many software patents generally describe a problem and the functions the system needs to perform, but they do not provide a detailed description of the technological specifics because doing the coding is trivial. The Courts have found these sorts of inventions ineligible on the grounds that you cannot merely describe a system and say "do it on a computer." By the same token, stating a problem and saying solve it with an AI algorithm feels similar.

PTO QUESTION 3: Does AI impact the level of a person of ordinary skill in the art?

Dragseth: I don't think AI has a direct effect on the level of skill in the art because, as I noted above, I think the inventors are the people who set the computer system in motion, and perhaps the people who interpret the computer system's actions (if that interpretation is meaningful in character).

But a lot of AI inventions involve applying computer smarts to some other problem, like drug selection, or signal processing, or other discrete technologies. So you are more likely to have an inventive "team" whose skill sets don't overlap much. That can make it really hard to provide a simple definition of the skilled artisan like we are accustomed to—e.g., "a person with an electrical engineering degree and four years' experience in the field designing X." But we already have cases with invention teams (shoot, there was a Canon ink-jet case once with a couple dozen inventors, and the Federal Circuit was cool with that). AI just means we'll have more.

Abbott: I wrote an article on this too. Yes, I would say that AI is already widely augmenting the skilled person in a variety of areas, and today, we might think about the "skilled person using AI" much the way we might think about the "skilled persons" in areas where team-based innovation is the norm. I suggest explicitly including this in the Federal Circuit's factor test for the level of ordinary skill. In the future, as AI moves from augmenting to automating R&D, the skilled person may need to become the inventive machine. At which point, we may need a new test for obviousness that focuses more on secondary factors or on the ability of AI to reproduce the content of a patent application.

Vidal: If the level of ordinary skill in the art is revised to the level of AI, then the exercise of determining obviousness could become circular. If operating on a data set, AI combines two references, then one could say that under an AI standard those concepts were obvious to combine. If AI invents based on a given data set, then one could argue that it would have been obvious from the perspective of AI to come up with that invention. Because AI can readily analyze the results of each combination and adjust its next combination, the whole hindsight analysis collapses. In a way, AI has hindsight from the get go. (My partner Chuck Klein adds: "Using AI might also affect reasonable expectation of success, and redefine "finite" number of potential solutions in the obvious-to-try context.")

Martinelli: I'm not entirely sure, but to the extent it defines the level of skill in the art downwards, you run smack into the problems [I've described] above.

PTO QUESTION 4: Do the disclosure rules (enablement, specification, etc.) need to be altered for AI-related patent applications? 

Dragseth: Perhaps I haven't thought enough about this, but I think the disclosure rules are plenty flexible as they already stand too. First, they are directed to what the person reading the patent can achieve—not to who, what, or how the invention was made. Now, I've been presented with some crazy-smart inventors in the AI space, so that certainly makes it harder to write a good, fulsome patent application. But that's true in all complex technologies.

I would like to see the PTO open "disclosure" rules to permit the filing of videos and such, which can be critically useful in understanding complex subjects. I assume that if a picture is worth a thousand words, then a 1,000-frame video is worth a million. That's not a function of AI though, but more a function of other types of "interactive" inventions we see and the ability to have an easily-accessible host for video that would meet disclosure requirements (e.g., putting a YouTube link in a specification, as long as YouTube keeps the video there for 20 years).

Abbott: This will have to be my next article, but nothing yet …

Vidal: Every rule needs to be rethought when it comes to AI and all the data the AI analyzed to come up with the invention, including not only enablement and the specification, but also which pieces of prior art must be disclosed to the PTO and whether AI must not only disclose that art (which could be impossibly voluminous), but also must rate the art based on how it weighed the references during its design process. If we are going to reward AI inventions, we need to make sure the public receives the appropriate quid pro quo.

Martinelli: Yes, there are big issues here with enablement in particular. The most common form of AI today is deep learning where AI systems are trained on huge data sets. If you are seeking to patent a solution created using deep learning, I think it may be difficult to enable your invention without also providing the data used to train the model. No one can redo what the inventor did without the data.

Again, it comes back to the incentives that the patent system is intended to create. Your AI "solution" might not be of any benefit to the public after your patent expires if the public does not have the data needed to re-create the solution. But submitting the data used to train the AI creates a whole slew of other problems (from privacy issues to the huge size of the data sets). Maybe this could be solved by requiring a patentee to submit a final trained model with its application. But there are issues with this too. For example, you would have to worry whether the model submitted is actually the best one the inventor could make or maybe the inventor is giving you a working but not commercially viable solution. (Maybe we would need to put the teeth back into the best mode defense.)

Cleanup on Aisle 1

The Ninth Circuit's decision Friday to stay pending appeal U.S. District Judge Lucy Koh's order enjoining Qualcomm from using its "no license, no chips" approach and from refusing to license chip competitors was a seismic event for the wireless industry.

But it came just hours before my deadline for Skilled in the Art, and in borrowing from my ALM colleague Ross Todd's fine article about the stay, I neglected to do what Ross did and that's credit the attorneys who scored this monumental (though interim) win.

Tom Goldstein of Goldstein & Russell signed Qualcomm's motion. Also on the brief were Cravath Swaine & Moore partners Gary Bornstein and Yonatan Even; Keker, Van Nest & Peters partners Bob Van Nest, Eugene Paige, Cody Harrisand Justina Sessions; and Morgan Lewis & Bockius partners Richard Taffet, Willard Tom and Geoffrey Holtz. Not to mention plenty of amici curiae.

Also interesting to note a few lawyers who joined Qualcomm's team while the case has been pending before the Ninth Circuit. One is Former Tenth Circuit Judge Michael McConnell, who overlapped at that court with Neil Gorsuch. McConnell is now senior of counsel at Wilson Sonsini Goodrich & Rosati, which also represents wireless patent goliath InterDigital.

Qualcomm also has added Kellogg, Hansen, Todd, Figel & Frederick partners John Thorne and Aaron Panner to its formidable appellate roster. Thorne has served as general counsel for the High Tech Innovators Alliance, a consortium of tech giants otherwise dedicated to fighting "baseless patent assertions," and Panner—stop me if you've heard this before—overlapped with Neil Gorsuch when Gorsuch worked at Kellogg Hansen in the early 2000s.

Cleanup on Aisle 2

I need to clarify an item from Friday's newsletter. I wrote that Techtronic Industries and Morgan Lewis & Bockius persuaded the Federal Circuit to rule a garage door opener patent ineligible, eliminating most of a $22 million patent infringement judgment—including enhanced damages, attorneys fees and an injunction. But I noted that a second patent that supports the core jury award was upheld.

That's only partly right. A second patent was indeed upheld, and it did support part of the jury verdict. But it's a very, very tiny part. The damages portion of the verdictfor No. 7,635,966 on a rechargeable battery backup system was only $57,300.

So The Chamberlain Group v. Techtronic Industries is a bigger appellate win than I thought for Techtronic, Morgan Lewis and co-counsel DLA Piper.

That's all from Skilled in the Art today. I'll see you all again on Friday.