Just Getting Started

The AI revolution is not navigating around the legal system...it's headed straight for it.

November 2022 is going to be marked as the beginning of something. It is not the beginning of AI nor of the wide use of AI or even large language models (LLMs). Those have been in use for years in a wide variety of tech companies. But, that AI focused natural language processing (NLP) has been largely confined to tools that were in the background for most users. The widespread attention to the November 2022 release to the public of ChatGPT by so many non-techies, non-AI enthusiasts has caused an explosion in interest in LLMs and AI in general.

The legal system is dependent on language and documents (pleadings, opinions, articles). There is nothing special to a LLM about the particular words that are used throughout the legal system. In fact, Chat GPT is trained as a language generalist. It can reproduce medical diagnosis, recipes and some rudimentary legal documents. The growth of the data these models are consuming to produce their results is not linear, it is geometric.

The number of different parameters (think levers to move as part of fine tuning the results of the tool) has also grown geometrically.

The anticipated GPT-4 (Yes, the building of it is already underway) will again geometrically increase these key underlying features. So what? As one twitter lawyer recently said, “what lawyers do no one is every going to trust to a machine, AI or no.” Well, consider what lawyers have become accustomed to just in my time since graduating law school in 1995. I was still, yes, still running around the library and figuring out how to locate cases in Reporters. Lexis was a recent toy that our school had adopted for limited use in the “computer lab.” In 2023, the “computer lab” is on everyone’s phone and all those reporters are searchable free on Google Scholar and have been for more than a decade.

There is no part of the legal system that has not been touched by technology, for better or worse since I was graduated law school. AI and the use of LLMs and the continued application of NLP tools is no exception to that rule continuing.

What do we do, us lawyers, that these kinds of tools cannot replicate? We get a set of facts, apply the law (either that which we know from experience in a narrow legal area or by researching some novel question) and then draft documents to combine those two features into a coherent argument. I have seen the sentencing data myself from my local county (scraped and analyzed using Python one of the main languages used to apply models like GPT-3) and it reveals that criminal cases are not all that different. There are not, it turns out, 70,000 different sentencing schemes over those cases from 2009-2019. There are patterns and many of them. The same with drafting pleadings, appeals, etc. (How many of us use templates, ones that we have created ourselves or cobbled together from other sources to make our own?). Templates are just rudimentary versions of what Chat GPT can do right now.

Accessing chatGPT just now and entering this prompt:

---

“Create ohio pleading argument for exclusion of evidence obtained via a bad search warrant”

---

Resulted in this argument in approximately 4 seconds:

Your Honor,

I respectfully request the exclusion of evidence obtained through a bad search warrant, as it was obtained in violation of my client's Fourth Amendment rights.

The search warrant in question contained numerous material falsehoods and omitted important information necessary for the magistrate to make a proper determination of probable cause. Specifically, the warrant failed to provide sufficient information to justify the search of my client's residence and the seizure of the evidence in question.

As a result, the evidence obtained through the search warrant is inadmissible under the Fourth Amendment's exclusionary rule. The exclusionary rule serves as a safeguard against unconstitutional searches and seizures, and it is essential for protecting individual rights and deterring police misconduct.

Furthermore, allowing the evidence obtained through a bad search warrant would reward the government's misconduct and undermine the integrity of the justice system. It is the responsibility of the court to ensure that law enforcement officials follow the rules and obtain warrants based on accurate and complete information.

In light of the bad search warrant, I request that the court exclude the evidence obtained through the search and suppress any derivative evidence obtained as a result of the illegal search. I respectfully ask that the court protect my client's constitutional rights and uphold the rule of law.

Thank you for your consideration.

Well, there is no case law at all you might say. Fair enough. However, GPT-3 (like the previous models) can also be fine tuned using a specific data set. It is not if, but when various companies begin training models on state and federal law, legal opinions, to the point where entire pleadings can be drafted in response to natural language prompts.

Fixing my garage door and my dryer

In the past five years, I have had an electric garage door opener and fairly new clothes dryer break. I was able to diagnose the failure, order the parts and repair both of them after less than an hour total watching YouTube videos. I not only saved money, but I became what would have been even five years ago, a sophisticated repair specialist on two systems I knew nothing about until they malfunctioned. That is not AI, that is the ease of access and application of knowledge unheard of without YouTube which is now ubiquitous in our lives.

The Everyday Lawyer

We have all come across clients whose depth of knowledge about their business and frequent enough conflicts have made them very aware of the legal issues that typically erupt in their business lives. Why do such clients need lawyers? Well, for one thing, they are not licensed and that is an absolute barrier to them, wait, it isn’t!

There are many non-lawyers who are going to increasingly rely on these tools and many not even yet built to simply enter well constructed prompts, rely on the pleading output and file their cases, make their arguments, etc. without ever calling a lawyer to begin with. It would be foolish to think every current business and personal client is going to ignore a tool, the $129.00 lawyer for example, (not yet built, but just go with me on this), that reliably provides them the proper format and arguments for a complaint, answer, motion to dismiss, motion for discovery, along with natural language explanations of how to use those arguments. All of that material will not be written by a lawyer, but it will be a tool that produces its content by consuming millions of documents written by lawyers. It will have the benefit of the work of more content than any one lawyer can possibly consume and recall. In short, it will be better at remembering, drafting and applying the law eventually. That “eventually” is not 50 years away. It is perhaps not 2024, but it is not 2050. Remember the geometric rise in capability noted above? Combine that with the use of GPT like tools to consume just the narrow world of legal documents and the future is clear.

Unresolved Issues

Recently, a well known emerging AI lawyer platform, www.donotpay.com, which bills itself as “The World’s First Robot Lawyer” advertised its intent to use its AI based tool in court. The plan was to have a non-lawyer enter into a hearing, with a communicate device in his ear so that the court proceedings could be heard by the AI tool offsite and instructions could be relayed to the litigant, arguments, questions, responses, etc. However, when actual lawyers became aware of this plan, they shut it down. The story is here. The threat was punishing human(s) at www.donotpay.com with practicing law without a license. Will that threat last to wall off the law from AI encroachment? I am doubtful, but it will all play out in a very public way as it has already.

One issue is what if the company above had just persisted? Or better yet, what about the eventual website where non-lawyers can pay a fee and enter their facts and other information and obtain pleadings via email that they then submit pro se? Who is the lawyer in that situation? Legal Zoom and other technologies are already drafting these documents and have been for years expressly for use by non-lawyers.

I can envision lawyers opening a practice specifically to cater to such clients. Receiving AI produced pleadings, reviewing them and offering edits to the pro se litigant and moving on. Is that really practicing law? Is that something a lawyer could be liable for in malpractice?

Gaming the judges. Judges have patterns and habits that AI can identify by consuming all their legal opinions, transcripts of their hearings and other public statements, speeches of those judges. Another service that will certainly arise is one in which a judge’s legal temperament will be reproducible. What better pleading to deliver to a court than one which actually is written in the judge’s voice?

Using ChatGPT just now with the following prompt:

“Make an argument in favor of vegan food in the style of Donald Trump”

The following response, arguably pretty well voiced to Donald Trump’s style, was produced:

Look folks, let me tell you, vegan food is fantastic, it's tremendous. Some of the best food out there is vegan, believe me. And let me tell you, it's healthy, it's good for you, and it's good for the environment. We need to start thinking about our planet and our health, and that's why we need to be embracing vegan food.

And you know what else? Vegan food is great for business. There's a growing demand for vegan options, and restaurants and companies that offer vegan food are seeing huge success. So not only is it good for our health and the environment, it's also good for our economy.

Now, I know some people might think that vegan food is boring or tasteless, but that's just not true. There are so many delicious and flavorful vegan options out there, and chefs all over the world are creating amazing vegan dishes that rival any meat-based dish out there.

So let's embrace vegan food, folks. It's the right thing to do for our health, our planet, and our economy. And let me tell you, it's gonna be huge.

It will not be a challenge for a LLM to produce pleadings written in the voice of the judge reading the pleading. The capacity to generate these nuanced manipulations at the sub-conscious psychological level will not only improve, it will never get worse.

Could this turn judges into merely arbiters of the best AI pleading?

Speculative markets for case outcomes. In Ohio, where I am, January 1, 2023 marked the explosion of sports betting platforms across all advertising mediums. The number of odd events connected to a game, match, etc that one can bet on is impressive. Undoubtedly, a tool will emerge that corporate legal departments will rely on to decide whether settling or pursing litigation is worth it. They will do so using GPT like tools that consume pleadings and then evaluate a pleading proposed by outside counsel, score it and determine that going forward is mathematically untenable. AI will be causing the settlement of cases in such a scenario.

I am sure I have missed some of the novel issues many of you have come across. Please comment with your questions or concerns and I will continue to do my best to respond to as many as possible. Thanks for reading!