Yesterday in my patent prosecution course, students turned to AI tools to help them draft patent claims. None of the AI-proposed claims were ready for prime-time, but they served as a useful starting point as the students organized their thoughts. More and more attorneys are turning to these same AI tools to help them be more productive and efficient while delivering a higher quality work product. It is tough, for instance, to read all the prior art. AI tools can help mine the references for potential obviousness problems — and provide a pin cite to the key language in the art.
USPTO Director Vidal recently released a new memorandum concerning the use of artificial intelligence (AI) in patent office proceedings. directorguidance-aiuse-legalproceedings. The memo recognizes that AI tools can be powerful both for applicants and for USPTO examiners. But, AI tools cannot be used to avoid ethical duties. The memo thus provides firm guidance that existing ethics rules on candor and misconduct apply even when AI tools are used to generate legal filings and evidence. This comes on the heels of several high-profile cases of “AI hallucination” outside of the PTO context, where language models like ChatGPT produced false information that lawyers presented as fact or law.
For example, submissions to the USPTO generally require a signature, and by affixing a signature, the signatory-who has to be a person-certifies, among other things, that “All statements made therein of the party’s own knowledge are true,” that “all statements made therein on information and belief are believed to be true,” that “after an inquiry reasonable under the circumstances” any “legal contentions are warranted by existing law” or “by a nonfrivolous argument for the extension … or reversal of existing law,” and that “factual contentions have evidentiary support” or likely will have evidentiary support after a reasonable opportunity for discovery.
Quoting from USPTO Rule 11.18. This rule is based directly on the Federal Rule of Civil Procedure , Rule 11 that has been applied in the AI context. Dir. Vidal goes on to highlight a few particular circumstances:
- Simply assuming the accuracy of an AI tool is not a reasonable inquiry.
- A submission (including an AI-generated or Al-assisted submission) that misstates facts or law could also be construed as a paper presented for an improper purpose because it could “cause unnecessary delay or needless increase in the cost of any proceeding before the Office.”
- Etc.
Be careful out there everyone!
Riffing on this post along with the subsequent, it’s easy to imagine that “fake prior art” has the potential to turn into a serious stumbling block for both the PTO and the courts, just as “fake data” has been an issue all along. Note that I’m not referring here to “fake prior art” created willfully and specifically for a particular case but rather just reams of computer-generated b.s., including b.s. about publication dates. And of course there’s nothing really “new” about fake prior art or using computers to create it. What’s “new” is the sheer volume of it that may become necessary to comb through. Imagine giant virtual online libraries of searchable PDF versions of “scientific journals” that never existed, or massive catalogs of “operating manuals” for machines that were never built. Oh, it’s going to be thrilling! But hey it’s a small price to pay for all that inexpensive super trippy artwork that humanity has been salivating for.
People may in the future be less concerned with outright “fake” prior art, but rather instead be concerned about prior art that was just made by an AI and disclosed (never patented, never put in production etc.). And there may well become gigantic reams of that. It may not be harder to produce than AI pictures. Tho nobody is doing it yet, I could with little trouble (many millions of $$$) automate an AI to augment many already existing disclosures with many an ez add on features, reams of which would not be obvious (and thus patentable), but also if disclosed this would provide gigantic amounts of prior art in short order. And that’s just making it do simple things.
6,
Whether or not YOU do that, there ARE efforts already underway TO do that.
In the land of music, for example, there has been published – and copyrighted – one ENTIRE set of every possible melody for a given set of conditions (IIRC, a select octave).
See now that might in fact be a good thing. Eventually that copyright will end if the congress gets off its ar se. Then it’s free from copyright forever. And probably none of it gets enforced anyway.
Music… not getting enforced…
You funny.
“But hey it’s a small price to pay for all that inexpensive super trippy artwork that humanity has been salivating for.”
Is it also a “small price” for current action in the not-so-grown-up arts that has Generative AI making inventions – inventions to which NO human could legally claim inventorship to; placing those inventions into a black box; closing that black box; taking that black box into another room; having a human open that black box; and having that human process the invention through the patent system?
And by the way, this was publicly admitted to by a major Pharma company in a US government sponsored, public session on AI.
Remember back when some jerk in The PTO OMB was going around giving unreality talks saying that every patent attorney was expected to read and understand the full contents of every patent or publication they cited in a Form 1449 [to avoid IC because it was found in a prior search or related subject matter application], even on obscure technologies that needed a specialized Ph.D. to understand? [Not even to mention the absurd presumption that PTO examiners are actually examining everything on a long list of cited technical publications and patents.]
Yes, an AI search dump could make that whole Rule 56 et al art citation in applications conundrum far worse.
As I recall, Examiner duty uses different key words – and no such “0h N0es, AI” arises because every bit as fast as an AI might “populate,” an AI could “de-populate.”
An even more general lesson for attorneys [not just judges] is the decision supporting Judge Newman’s one year suspension noted in IPWatchdog today. The lesson is that just the refusal to cooperate in a formal professional investigation of an attorney is, in itself, sufficient grounds for a temporary practice suspension or other professional sanction, irrespective of, and without having to reach, the Merits of the complaint.
That’s just it – it was not an abject refusal, but a refusal instead to be railroaded.
There is a world of difference, and this latest decision should have reflected that.
Instead, we get a case of “Black Flu.”
Newman should have refused to sign her employment contract if she didn’t like the rules.
More than certain that there were NO rules concerning acquiescing to being railroaded in any employment contract or oath of office that she took.
But you be you and ‘celebrate’ this – until the day when you face being railroaded.
Much how you ‘celebrated’ the destruction of a cogent jurisprudence on eligibility, and then turned around and B 1 tched and M0aned about the Current Court’s moving back to center with the likes of Dobbs.
A question (for being careful out there):
How, and to what, were citations from the students provided that reflected the AI work as opposed to their own work?
Has the BlueBook been updated for this new modern age?
Using AI for claim drafting, even for an initial set of claims, raises another important question – confidentiality. Obviously the AI program will need access to the client disclosure regarding the invention. Can you assume that the information in that disclosure will stay secure and private? Will the AI save that confidential information as part of its training and learning process and then use it elsewhere?
This contrasts with ‘un-controlled’ cloud storage (which is not totally out of the woods itself).
No doubt. I would never allow for cloud storage of confidential information related to my practice (or any information for that matter). If a lawyer does not have complete control of all information in their practice, then there is the possibility that whomever has control could grant access to someone else. The malpractice implications there are obvious.
Certainly DVan – and yet, I am also certain that most all of the Amlaw 200 do NOT exercise the control that you indicate that you do.
I, for one, have not and will not use AI in any part of my patent prosecution practice. I think Dennis is living in an Ivory tower and has no clue what real patent prosecutors do on a day-to-day basis.
I’m of a certain age so, to me, the thought of AI drafting an initial set of claims for new attorneys, is abhorrent. Learning how to draft claims from scratch helps develop the skills a good patent practitioner needs. When AI does the initial thinking for you, you’re playing along with the AI own inherent bias and shortcomings.
I don’t like it. I don’t like it at all.
When AI takes all your jobs. There will be plenty for those that work in the menial positions. And lately we have plenty of people that be happy to fill those positions. What is this Country thinking.
I don’t see what will follow the age of AI. Used to be, each New Age complimented the people and their way of life.It is no longer about the people.
I’m also pretty skeptical about claim drafting. Ditto for the point-of-novelty parts of the disclosure, as that stuff is definitionally new.
OTOH, I’d consider it for the “enablement one level of abstraction below your claims” junk we need to add to every disclosure. OTOOH, I already have well-reviewed boilerplate for most of that, so it’s probably not a huge savings.
Vidal should concentrate on getting her own house in order. In the past several months she has [expletive past tense verb] up just about every aspect of the PTO’s computer systems:
-she shut down EFS and PAIR and forced people to use Patent Center, which doesn’t provide all the same functionalities (yet she has repeatedly lied that it does);
-she shut down the old trademark searching system and replaced it with something far inferior that seems to require a PhD to make use of;
-she imposed a $400 fee for those who insist on filing their patent applications in pdf rather than “docx”, even though the former is the only method an applicant has to be sure that what he submits is what the PTO has, whereas the latter isn’t a standard format and it’s known that the PTO introduces errors when handling “docx”;
-and in the past week she shut down the systems for recording patent and trademark assignments and replaced them with systems that weren’t tested by users beforehand and that thus far have proven to be disasters;
-and in all these cases, the PTO steadfastly refuses to listen to user input or even acknowledge it.
And where is Dennis on this? Nowhere. Because he doesn’t get that these things are no less more important to applicants than the latest case law, even though it should be plain as day that if the PTO takes an equation or chemical structure in your patent application and converts it to something else, that can have disastrous consequences down the road; or that wasting everyone’s time by creating and implementing new, crappy systems is not ok.
Yeah – this misses some of the more important ethical issues that arise with using a third party AI with client confidential data.
Thank you, Atari Man. Why is there no publicity about the USPTO IT debacle? Where is PatentlyO? Where is IPWatchdog? Where is mainstream press? DOCX erodes the integrity of the U.S. patent system. Vidal should focus on getting her house in order.
“In the past several months she has [expletive past tense verb] up just about every aspect of the PTO’s computer systems”
You’re more right than you know. The PTO migrated its internal patent search to the cloud, and the results have been disastrous. There have been many outages and when it does function it runs much slower than before. Examination time has effectively been cut.
Don’t you guys have a union for such grievances?
I wish AI would help attorneys draft claims that aren’t ar se at the outset. I ain’t holdin my breath, they’ll find a way to f it up. And then find a way to miss whatever on third look over.
As if the comparative level of arse-ness does not weight heavily against examiners…
As suggested, the logical effect of attorney use of AI that gives erroneous or misleading case citations submitted to the PTO is a violation of the cited section of USPTO Rule 11.18 and FRCP Rule 11 by the paper signatory. In reality, as one example among many, the Sup. Ct. decision in Diamond v. Chakrabarty is frequently misquoted without any sanction or punishment by truncating off the part actually relevant to Alice and Mayo Chakrabarty express patentable subject matter exceptions.
Oh please — giving full and proper quotation treatment to the Supreme Court jurisprudence on eligibility each and every time would show — each and every time — that the Supreme Court has created a Gordian Knot of conflicting mess.
It’s a good thing that blogs don’t “require” this as your own posts would be routinely sanctioned.
Comments are closed.