Will generative AI replace patent lawyers?
Zachary Silbersher
By now, we have all read about the recent debacle of an attorney who filed a brief drafted by AI that turned out to have fabricated case citations. The case is a cautionary tale. But it will hardly diminish the threat that AI poses to the legal profession. That threat applies equally to patent lawyers. Will generative AI eventually replace patent lawyers?
The Economist recently wrote that 44% of legal tasks can be handled by AI. The article points out that one of the major consequences for lawyers is that AI could reduce their numbers. (Many probably view that as a good thing.)
The concern among lawyers is legitimate. The existing AI capabilities suggest that there may be a day in the near future where writing a brief may be as simple as inputting the right prompts into a generative AI platform. ChatGPT may be hallucinating case citations right now, but that wrinkle is surely to be ironed out eventually. Accordingly, when it comes to patent lawyers, it’s worth asking how big of threat AI poses to their future.
The question is muddied by a closer examination of what patent lawyers actually do. Lawyers do a lot of things, but for the purpose of this post, we’ll focus on written work product. Very generally, patent lawyers draft three things: license agreements, patent applications, and persuasive arguments, i.e, responses to Office Actions or briefs in IPRs and litigation.
License agreements are contracts, and AI’s threat to lawyers writing contracts is likely overstated for now. AI can easily provide draft contracts that include all material terms. Yet, even before AI, few licenses were drafted from scratch. Lawyers typically begin with an existing precedent and revise the terms to fit their particular client’s matter. AI can easily replace the drafting and coordination of contracts, but lawyers will still be needed to understand the implications of terms and reach compromises or agreement with the other side. There may obviously be a time in the future where your lawyer is essentially HAL doing all of that for you, but that’s presumably still aways off for now. (Although, given the pace of AI right now, who knows for sure.)
Patent applications are a bit more tricky. It is not uncommon to start a patent application with disclosure from a related technology or a disclosure provided by the client, but large portions of patent applications (at least the parent applications) are still drafted at least partially from scratch. Undoubtedly, AI could seriously cut down the legwork of that effort by harvesting background disclosures from applications and sources covering related technology. AI also could significantly alter the calculus of prior art that must be disclosure during prosecution. Yet, at some point, AI’s capability to finish your patent application is likely to run up against the cold fact that the disclosure is presumably novel and never previously disclosed on the internet.
So, what about briefs, i.e., the primary tool and battleground for litigators? To keep the scope of this blog post manageable, we’ll limit the discussion to briefs addressing invalidity arguments. Whether a lawyer is drafting a response to an Office Action, preparing a petition for inter partes review, drafting invalidity contentions, or drafting a summary judgment brief, the fundamental exercise is the same—comparing limitations of the purported invention to features of the prior art.
As we’ve already seen, AI can already easily give you the skeleton of draft brief that includes legal citations. Ask for a draft summary judgment brief arguing for the invalidity of a certain patent claim, and generative AI can presumably serve up something useful. Go a step further and ask it to cite cases in your jurisdiction that are both recent and share your procedural posture and do not presumably include any otherwise negative holdings, and AI can likely give you a tremendous head start. That’s what armies of junior and mid-level associates spend much of their time doing. Maybe their days are numbered.
But, at some point, someone has to write the facts. The facts are typically unique to particular cases. Oftentimes they include confidential or proprietary information (although, less often during invalidity proceedings)—which would theoretically diminish AI’s utility.
In addition, someone need not only write the facts, but shape the facts. AI may soon be capable of easily serving up the law, (despite recent hallucinations to suggest otherwise.). Yet, can AI shape the facts with an eye towards persuasion? Importantly, this is advocacy, not just description. During an IPR, the petitioner is likely to draft a description of a particular piece of prior art very differently from the patent owner. On the other hand, if AI can generate a song in the voice of an existing artist, then perhaps it can shape facts with an eye towards making a persuasive argument.
At another point, someone has to compare the facts to something else. Can AI compare the facts to the law or compare aspects of the invention to the prior art? Again, this is not description, but advocacy. Both sides are going to emphasize different features of the invention or the prior art, and correspondingly, play down the importance of others.
Admittedly, I don’t know the answer to all of these questions. My presumption is that AI’s capability to do these tasks is not only far off, but potentially never off. That may sound unconvincingly cavalier. Indeed, if generative AI is already wowing the world with its feats of mastery within both engineering and arts, then surely knowledge workers such as lawyers should be threatened by AI’s enormous prowess.
Yet, within the context of arguing invalidity before some tribunal, is that really true? For engineers, the hallmark of a particular challenge is finding a solution that works in the world. The rocket either launches or it doesn’t launch. For artists, the hallmark of the creative process is making connections that are at once surprising, but at the same time still accessible. Yet, for litigators, the hallmark of successful advocacy requires persuading another human’s mind.
Having the goal-posts of your profession set up between an adjudicator’s ears is a unique one. The blessing and curse of being a litigator is that you may have the technically correct argument, but still lose. And vica-versa. Indeed, litigators spend considerably more time assessing the psychology of their arguments than most people realize. In other words, how is the jury going to receive this? How will the judge? Is it too technical? Too much in the weeds? Too dumbed down? The technically correct argument isn’t intuitive, but the intuitive argument isn’t technically correct?? This is especially so for patent litigators, where the disputes often lack a human story or a “gut” feel for who is in the right.
Perhaps it is naïve to think that computers cannot do the mental work of assessing whether a given invention is obvious in view of available prior art or determining whether there was a motivation among persons of ordinary skill to combine specific references. Perhaps this post is nothing more than a knee-jerk, defensive reaction. Versions of this post are likely being written by members of every knowledge profession facing a threat to their viability by generative AI.
The only thing that will tell us for sure is time. And possibly a lot sooner than we think. Yet, looming behind the answer to these questions is another one, perhaps even more daunting than whether generative AI will replace patent lawyers: if AI can eventually do all the work of lawyers, then what’s stopping it from doing all the work of judges?