PatLitig.ai (forthcoming AI-related & AI-based work on patent litigation)
- There is every reason to try incorporating AI into patent litigation workflow, BUT: First, do no harm: any discussion of using AI in litigation must begin with a parade of horribles relating to attorney and expert overly-credulous of AI chat output in court filings, without sufficiently checking for so-called “hallucinations” (which are really instances of standard LLM BS-ing not always working)
- Damien Charlotin, AI Hallucination Cases — Database currently with over 380 cases internationally; you can select jurisdiction, “nature of hallucination” (fabricated case law, false quotes, misrepresented case, ), outcome/sanction, monetary penalty, party (attorney vs. expert vs. judge vs. pro se litigant [which are largest group])
- Possible responsibility to check the other side’s work?: Lawyers Dinged for Failing to Detect Opponent’s Fake Citations
- Experts too: Judge Strikes Part of Anthropic (Claude.AI) Expert’s Declaration, Because of Uncaught AI Hallucination in Part of Citation (the AI chatbot front-end to this web site reminds me that there is discussion elsewhere on this site of ways experts might responsibly use AI-generated material in expert reports — TODO: umm, where?)
- Possible use of AI (given uncharacteristically flowery language) in attorney’s apology for using AI — hilarious AboveTheLaw article: Lawyer Cites AI Hallucinations, Responds With Pretentious Meditation On Nature Of Being (but falsely characterizing it as “going all Camus” — a horrible slur against the author of The Stranger and The Plague — given that brief says e.g. “Your Honor, in the ancient libraries of Ashurbanipal, scribes carried their stylus as both tool and sacred trust—understanding that every mark upon clay would endure long beyond their mortal span”)
- AI hallucinated caselaw makes its way into trial court decision, caught by appellate court: Shahid v. Esaam ; Above the Law article
- Excellent IPKat article: Hughes, “Use of AI in the patent industry: The spectre of hallucination” (mostly on AI indicia appearing in scientific papers, patents, and published patent applications, and giving pre-LLM examples of failure to proofread)
- News re: attorneys & AI hallucinations
- Claim chart drafting using AI tools such as ChatGPT for infringement and invalidity contentions, and drafts of expert reports.
- Perplexity.ai Patents (beta) looks very promising: see chat re: Google Transformer patents ‘786 and ‘978 compared to OpenAI ChatGPT and GPT-1 paper; notes on Perplexity.ai in outline of forthcoming Patent Litigation book
- Mining PTO dataset on patent litigation, as well as PTAB data, and linking to patent claims (to help assess “what is special about claims in litigation?”).
- Comparing litigated vs. non-litigated patents across CPC classifications, and across patent-claim characteristics.
- Finding potential infringement or non-infringement using semantic similarity of claims to, or distance from, product-related text (this will leverage earlier work on “CodeClaim”).
- Writing Python code with ChatGPT to analyze huge datasets (as often seen in litigation, e.g. one case of thousands of spreadsheets, including some CSV files with >million rows, exceeding Excel current capacity)
- Using AI in discovery, including developing local models to avoid confidentiality problems with cloud-based AI models.
- Discoverability of AI training data?:
- Building local offline (non-cloud; for use under court protective order) source-code examination/comparison tools, using e.g. CodeBERT and CodeLlama .
- AI ability to help generate (or even self-generate) inventions: [TODO: This not specifically re: patent litigation; probably move]
- What are long-term implications for patent system’s goal of incentivizing invention, if (per IJ Good’s 1965 projection) AI is the “last invention that man need ever make” because it could design even better machines? Probably still need the incentive to disclose, rather than keep inventions as trade secrets (actually an oxymoron, since an invention by definition under patent law is sufficiently disclosed).
- OpenAI & Argonne Labs [TODO: find link specifically re: inventions]
- Inventive AI contradicts the idea of LLM as the perfect PHOSITA (person having ordinary [non-inventive, though also all-knowing] skill in the art)
- On AI & PHOSITA, see IPKat: Hughes, “An LLM is not (yet) a person skilled in the art (T 1193/23)”; and Putting the Person in PHOSITA: The Human’s Obvious Role in the Artificial Intelligence Era.
- See Thaler DABUS cases [I probably won’t have much here on issues of AI as a listed invention; I find it fairly uninteresting, because I assume patent attorneys will find a way around inventorship restrictions]
- See Plotkin, The Genie in the Machine: How Computer-automated Inventing is Revolutionizing Law and Business (2009)
- See John Koza patents based on technology generated with genetic programming, e.g. US 6,211,726 as described at “Synthesis of a High-Current Load Circuit (A Human-Competitive Result Produced by Genetic Programming)”
- Case study: Google not only came up with revolutionary Transformer architecture (“Attention Is All Your Need” paper, 2017), but obtained Transformer patents (including US 10,452,978, granted Oct. 2019); yet it was OpenAI that first productized (ChatGPT, Nov. 2022). See brief chat with Google AI on why Google did not litigate Transformer patents against OpenAI [TODO: extract some details here]
- Use as case study in how important or not patents and/or trade secrets are to current AI progress
- How much is current AI “open source” vs. trade secrets?
- Do current AI patents include useful disclosures, or no more (perhaps less) than arXiv.org papers, Hugging Face, GitHub?
- TODO: compare “Attention Is All Your Need” with disclosures in US 10,452,978 (16 page PDF)
- [TODO: probably create separate page on Patents & AI, distinct from this page on Patent Litigation & AI?]
