Can AI Be Used to Resolve Textual Ambiguity in Court Cases?

“The imperfect is our paradise.
Note that, in this bitterness, delight,
Since the imperfect is so hot in us,
Lies in flawed words and stubborn sounds”

— Wallace Stevens, “The Poems of Our Climate”

Legal disputes often turn on simple disputes over the meaning of words. Latin named canons of construction have been a bedrock of lawyer training for centuries; but, these rules are imperfect, situational, and rely on human judgment regarding language and (sometimes) extrinsic evidence.  So, leave it to Wallace Stevens—the great American poet and insurance lawyer—to point out that verbal imperfection is a fundamental human trait.

Can computers interpret language more clearly, or will they be able to in the fullness of time? While this was long the subject of science fiction, recent rapid advances in AI technology make what was once a fantastical flight of fancy potentially within reach. A recent opinion in a U.S. federal court insurance coverage case provides a fascinating gloss on the subject.

Snell v. United Specialty Ins. Co. (“Snell”) turns on resolving a potential verbal ambiguity, as many insurance cases do. 102 F.4th 1208 (11th Cir. 2024). James Snell was sued for the alleged negligent installation of an in-ground trampoline.[i] United Specialty Insurance Company denied coverage to Snell arguing that, while the policy at issue covered claims arising from Snell’s performance of “landscaping,” it did not cover claims arising from the assembly and installation of a trampoline.[ii] Snell’s lawsuit followed.

The primary issue was the “ordinary meaning” of the word “landscaping” and whether Snell’s “installation of an in-ground trampoline, an accompanying retaining wall, and a decorative wood ‘cap’ fit within the common understanding of the term ‘landscaping’ as used in the insurance policy.” See Id. at 1221. The District Court found for the insurance company.[iii] The Court of Appeals affirmed.[iv] Among other things, a dictionary definition helped the courts reach their conclusion. Pretty standard stuff.

The case in and of itself isn’t noteworthy—a workaday insurance case like thousands that are litigated across the United States every year. Judge Newsom’s concurrence takes a surprising turn, however, with an extended discussion of the potential uses of AI for such cases. He begins by “pull[ing] back the curtain on the process by which [he] thought through one of the issues in the case” to make a single proposal:

Those, like me, who believe that ‘ordinary meaning’ is the foundational rule for the evaluation of legal texts should consider—consider—whether and how AI-powered large language models like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Clause might— might—inform the interpretive analysis.[v]

In other words, Judge Newsom suggests that AI-enabled large language models (“LLMs”) could be useful to judicial interpretation or, more generally, the “interpretive enterprise.” See e.g., Id. at 1226-1234. He explains that while “laboring over the question [of] whether the trampoline installation qualified as ‘landscaping,'" he reviewed a buffet of definitions ranging from (a) Wikipedia’s definition—“any activity that modifies the visible features of an area of land,” to (b) Meriam Webster’s definition—“to modify or ornament (a natural landscape) by altering the plant cover.”[vi] After finding that the dictionaries “left something to be desired,” he wondered: “Is it absurd to think that ChatGPT might be able to shed some light on what the term ‘landscaping’ means?” See Id. at 1222. 

Wondering this, Judge Newsom queried the following on ChatGPT and Google’s Bard (now known as Gemini):

• What is the ordinary meaning of “landscaping?”[vii] 
• Is installing an in-ground trampoline “landscaping?”[viii]

See Id. at 1224-225. Ultimately, he found that both ChatGPT and Google Bard’s responses to each question were sensible and could provide insight into how ordinary people may use the word “landscaping.” See Id. More importantly, his process caused him to consider whether LLMs can generally be a useful tool, alongside “dictionaries, canons, and syntactical context,” in judicial interpretation when determining the ordinary meaning of word or phrases. See Id. at 1232.

Why might LLMs be useful tools?  Judge Newson offers the following:

1. LLMs are trained using ordinary language inputs, including hundreds of billions of words, related data, and their usage, which allows the LLMs to capture how individuals use language in their everyday lives. 
    
2. LLMs are increasingly able to “understand” context. Given (x) as mentioned above, LLMs are trained using large data sets, and (y) researchers have developed (and continue developing) methods of converting language into math for computers to process and analyze, LLMs are enabled “to absorb and assess the use of terminology in context and empower[ed] … to detect language patterns at a granular level” and even predict how such language is used.
    
3. LLMs are accessible to judges and lawyers as well as to ordinary citizens. As a result, LLMs can “democratiz[e] the interpretive enterprise … by leveraging inputs from ordinary people and being available for use by ordinary people.” Further, LLMs can serve as an inexpensive research tool.
    
4. LLMs research can enhance the transparency and reliability of “the interpretive enterprise” or, at a minimum, do not involve more discretion or opacity than what is inherent in our current practices—i.e., the use of definitions, whose construction is not always self-evident, by judges who seldom describe the process by which they selected one definition over another.
    
5. LLMs hold advantages over empirical interpretive methods, including large surveys of ordinary citizens regarding the meanings of certain words and phrases because, among other things, LLMs are widely available and accessible.

See Id. at 1226 -1231. 

Conversely, he notes these potential downsides:

1. LLMs can “hallucinate,” which means they can fabricate information. Over time, however, the hallucination risk is likely to decrease.
    
2. LLMs might not fully account for underrepresented population’s usages. Specifically, “[p]eople living in poorer communities (perhaps disproportionately minorities and those in rural areas) are less likely to have ready internet access and thus may be less likely to contribute to the sources from which LLMs draw in crafting their responses to queries.”
    
3. Lawyers, judges, and would-be litigants might try to manipulate LLMs, either by reverse engineering a preferred answer, or seeking to corrupt the data upon which LLMs train and use to form their responses.
    
4. There may be concerns that reliance of LLMs will lead us into dystopia, where “robo judges” are algorithmically resolving human disputes.

Another issue to consider is the means and methods by which an LLM is used. As Professor Harry Surden noted at a recent presentation at Cambridge University, the way that a LLM is queried can have a material impact on the result that you receive. For example, asking for the “ordinary meaning” can lead to a different result than asking for a “reasonable interpretation.” Additionally, these concepts can be terms of art with court glossed meanings that differ by jurisdiction, a distinction that a general LLM may not pick up. Thus, training would be essential for a judge to rely reasonably on this technological innovation for reliable interpretive assistance.

*          *          *

Good or bad, flawed or not, the sheer breadth of knowledge to which modern AI has access and the rapid advances in this technology make its use seem somewhat inevitable. At the very least, the technology will be added to a litigator’s and judge’s interpretive tool-set. One would expect learning to responsibly rely on the technology is likely to become as much a part of training as the canons of constructions mentioned at the outset. Judge Newson’s concurrence may be one of the first extended judicial discussions of the use of the technology by lawyers but it’s a certainty that it won’t be the last.

^{[i] See Id. at 1212.}

^{[ii] See Id.}

^{[iii] See Id. at 1212.}

^{[iv] See Id. at 1221.}

^{[v] See Id.} 

^{[vi] See Id. at 1223.} 

^{[vii] ChatGPT and Google’s Bard (now known as Gemini) responses are as follows:}

^{Query: What is the ordinary meaning of “landscaping?”}

^{ChatGPT’s Response:} 

^{“Landscaping” refers to the process of altering the visible features of an area of land, typically a yard, garden or outdoor space, for aesthetic or practical purposes. This can include activities such as planting trees, shrubs, flowers, or grass, as well as installing paths, fences, water features, and other elements to enhance the appearance and functionality of the outdoor space.}

^{Google’s Bard Response:}

^{The ordinary meaning of landscaping” is the process of modifying the visible features of an area of land, including:}

• _{Adding or removing vegetation, such as trees, shrubs, and flowers}
• _{Installing hardscape elements, such as walkways, patios, and retaining walls}
• _{Creating or enhancing natural features, such as ponds, waterfalls, and rock gardens}

^{Landscaping can be used to improve the aesthetics of an area, increase its functionality, and/or protect it from erosion. It can also be used to create a sense of privacy, define spaces, and add value to a property…}

^{Landscaping can be a DIY project, or it can be hired out to a professional landscaping company. The cost of landscaping will vary depending on the size and scope of the project.}

^{[viii] ChatGPT and Google’s Bard (now known as Gemini) responses to this query are as follows:}

^{Query: Is installing an in-ground trampoline “landscaping?"}

^{ChatGPT’s Response:} 

^{Yes, installing an in-ground trampoline can be considered a part of landscaping. Landscaping involves altering the visible features of an outdoor area for aesthetic or practical purposes, and adding an in-ground trampoline would modify the appearance and function of the space. It's a deliberate change to the outdoor environment, often aimed at enhancing the overall landscape and usability of the area.}

^{Google’s Bard Response:}

^{Whether or not installing an in-ground trampoline is considered landscaping depends on how you define the term…}

^{Here is a table that summarizes the different ways to define landscaping and whether or not installing an in-ground trampoline would be considered landscaping under each definition:}

^{Ultimately, whether or not installing an in-ground trampoline is considered landscaping is a matter of opinion. There is no right or wrong answer.}

^{See Id. at 1234-35.}