We Can Remember It for You Wholesale: Artificial Intelligence in the Practice of Intellectual Property

In this edition of Legal Currents and Futures, The Colleges of Law continues sharing a series of thought pieces about artificial intelligence and the law.

New Artificial Intelligence (AI) tools have set headlines for their abilities, the big business deals surrounding them, and their pitfalls. OpenAI’s ChatGPT is the fastest-growing app of all time, leading to an investment by Microsoft to embed the technology into web search and Microsoft Office. ChatGPT is the “fastest growing app in the history of web applications.” Google is racing to catch up with their competing product, Bard; meanwhile, OpenAI’s Dall-E, Stable Diffusion, Midjourney, and a plethora of other AI image generation products have run rampant over our computers and phones.

However, these technologies open up many new legal questions. Who owns the works generated by AI? Should users or their creators be liable for their output? Is it fair that AI digests the works of uncredited, uncompensated individual creators across the internet? These questions are unsettled, and many more legal questions about AI technologies will be posed in the future.

In addition to new questions, AI also opens new business opportunities and ways to improve efficiency. Properly used and supervised, AI can speed up creative work, automate repetitive tasks, tackle rote tasks, and provide simple scripts for standard tasks. These opportunities also come with risks. The current generation of AI provides biased, potentially-infringing, and out-of-date answers, that is, when it is not hallucinating—providing dead wrong information with absolute certainty.

Unfortunately, general AI products are not trained on specific legal datasets. Lawyers should be particularly careful when using them. Consider Peter LoDuca and Steven A. Schwartz, two attorneys practicing law in New York. They researched case law using ChatGPT and cited the discovered cases in a brief to the court. The only problem was that those cases did not exist. When the court eventually learned of counsel’s bogus citations, it issued sanctions and required the attorneys to send their client a copy of the sanctions order, a transcript of the sanctions hearing, and a copy of the affidavit wherein they submitted the fake cases. They were also required to submit those same materials to each federal judge improperly identified as authoring the false opinions. This is all in addition to a fine of $5,000 each. Mata v. Avianca, Inc., 22-cv-1461 (PKC) (S.D.N.Y. Jun. 22, 2023) (Dkt. 54). In reaction, at least one other court now requires attorneys to certify that AI drafted no portion of filings before the court, while other courts ask lawyers to indicate if AI was used for any portion of a brief. Judge Starr in the Northern District of Texas requires human attorneys to verify each filing because AI tools are prone to hallucination and bias and have not sworn an oath to uphold the law. Other local rules will doubtlessly proliferate as lawyers become more comfortable with (and make mistakes because of) the technology.

Consequently, lawyers should use AI tools cautiously, particularly general market AI tools. On the other hand, AI tools trained on specific legal datasets have already hit the market. Casetext’s CoCounsel and Disco’s Cecilia aim to bring AI to existing electronic files and ediscovery, respectively. And, in our practice, new specialized tools like PatentBots and Huski.ai already promise to make patent drafting and trademark practice more efficient.

In addition to specialist tools, generative AI can accelerate the non-legal work of a law firm, if supervised properly. AI can be used to generate first drafts of promotional materials or advertising. Or it can be used to quickly try a change of tone, such as “make this paragraph more exciting.” (See our bios for an example.) The more adventurous and diligent lawyer may feel confident enough to use AI to write first drafts of boilerplate provisions or summarize documents. But of course, none of these tasks can be completely entrusted to AI. They must each receive a thorough fact-checking by a trained human.

AI’s Current Use Within and Effect on IP Practice

Currently, AI use remains limited in intellectual property practice. That is likely to change rapidly. Major research providers Thomson Reuters and LexisNexis have integrated AI partly in their search engines, with more integration to come. LexisNexis’ AI-powered freeform text searching already shows dramatic promise at improving an aspect of legal research that was stagnant—and not particularly good—for years. An increasing number of tools look to shorten the cycle of research, writing, and proofing legal documents and briefs. There are AI contract writers, readers, and summarizers that operate on user prompts. Some of those products operate upon closed datasets such as one’s own contracts or briefs. As with all AI tools, these tools will be best trained by more extensive, higher-quality datasets; datasets more likely to be available to larger national law firms and clients.

Likewise, tools for patent drafting, review, synthesis, and drawing generation have increased in number. Large portions of the process most patent attorneys undertake to prepare patent applications are becoming automated or partially automated. However, many of these tools do not take into account how patent lawyers actually operate. For example, most patent attorneys begin applications by drafting claims, then prepare drawings that correspond to those claims, and then prepare the specification after that. Despite this standard process, there exist several patent AI tools that prepare drawings from wholly complete specifications. See, e.g., PatentDraw, available at https://patentdrawai.com/ (accessed September 11, 2023). This simply does not match typical workflows for patent attorneys, but the tools will evolve and become more relevant and useful quickly.

The number of tools available for finding prior art for patent invalidation continues to increase as well. Some of the best tools in the past several years have relied upon crowdsourced prior art discovery. See, e.g., Patexia. These tools have proven quite effective, so long as the rewards offered are adequate or one can obtain the attention of motivated individuals seeking relevant prior art. AI, particularly with its focus on large language models, seems extremely well-suited to the process of obtaining relevant prior art, so long as that art is available on the Internet or in a suitable data set. See, e.g., IPRally. Similar tools exist and will increasingly exist for trademark and copyright practice.

Individuals and companies also use AI to test the frontiers of intellectual property law. One individual, Stephen Thaler, seeks to enshrine AI-created works and inventions in the law as protectible with the AI as the “author” or “inventor.” So far, Mr. Thaler has failed as the USPTO, Copyright Office, and Courts have roundly held that authorship and inventorship require humans as the primary movers. Thaler v. Vidal, 43 F.4th 1207, 1209 (Fed. Cir. 2022) (refusing patent for non-inventorship), cert. denied Thaler v. Vidal, 143 S. Ct. 1783 (2023); Thaler v. Perlmutter, Civil Action No. 22-1564 (BAH), 2023 U.S. Dist. LEXIS 145823  (D.D.C. Aug. 18, 2023) (refusing copyright registration). Others, typically content creators or authors, attack AI software that have relied upon their works to create “new” works. See, e.g. Doe v. Github, Inc., No. 22-cv-06823-JST, 2023 U.S. Dist. LEXIS 86983, at *3 (N.D. Cal. May 11, 2023) (litigation re GitHub’s use of users’ code to power its Copilot software); Andersen et al v. Stability AI Ltd. et al., Case. No. 3:23-cv-00201-WHO, (N.D. Cal. Jan. 13, 2023) (litigation regarding images used to train AI).  In copyright, these works are all arguably derivative works and, thus, compensable under the present Copyright Statute. 17 U.S.C. § 101 et seq. AI is also being used to allegedly defame individuals by making up lies about them while hallucinating.

What does the Future Hold?

The list of potential uses of AI in intellectual property is long. It almost certainly will speed drafting briefs and patents, will significantly aid in research, and will likely be the first step in every document written at some point in the near future. AI will speed and streamline eDiscovery processes in litigation, enable rapid standardization of contract terms, and likely enable parties to reach an agreement more quickly and enable rapid and efficient research and review of prior work. Web-based service providers like online retail, auction sites, web hosts, domain registrars, and similarly situated entities already use AI to rapidly detect users who infringe upon site policies or commit fraud and remove them before consumer complaints. AI-powered tools also automate takedown requests seeking to remove counterfeit goods from the internet. These and other functions will elevate law practice, removing much tedium but forcing attorneys to “up their game” to more high-value work. The initial stages of this change will challenge some lawyers. Still, with time, AI promises to unlock greater creativity, allow lawyers time for more engaging work, and perhaps create better overall lawyers than those operating before these tools were available. Just as word processing sped up the law practice and texting sped up the expected responsiveness, AI adds more breadth and power to a single lawyer or group of lawyers, allowing them to work more efficiently and effectively to support clients.

According to ChatGPT, Jonathan Pearce boasts an electrifying track record, providing invaluable guidance to the titans of interactive entertainment and virtual reality. Meanwhile, Christopher Dugger is a legal powerhouse specializing in intellectual property and privacy law, dedicated to arming businesses with the tools they need to protect their invaluable creations. Both are intellectual property attorneys at SoCal IP Law Group LLP, whose practice increasingly involves artificial intelligence and its application to technology writ large.