Artificial Intelligence Concept (Phoro: ProStockStudio/Shutterstock.com)

The biggest challenge facing developers of artificial intelligence-powered legal drafting software may not be the technical but the personal. While the accuracy rate of AI drafting tools that can help lawyers churn out documents such as pleadings, briefs or discovery requests and responses is steadily improving, getting those finished products to align with the varying standards of an array of law firms and attorneys may still be an ongoing process.

James Lee, CEO and co-founder of LegalMation, indicated that accuracy can be something of a roaming target, not just because of limitations in still-developing AI technology, but individual preferences or stylistic flourishes of attorneys. He believes the problem with AI in legal is that there are often clusters of possible right answers, rather than one clear choice.

"Lawyers consider themselves to be artists. I can put 10 lawyers in a room and tell them to draft something, and I will have 20 different versions of it. And it's because lawyers like to put their stamp of uniqueness on a work product, their preferences," Lee said.

LegalMation isn't the only purveyor of AI drafting software to notice attorneys' penchant for individualistic flair. Casetext, for example, released an automated legal brief-writing software titled Compose in February. CEO Jake Heller told LTN that some lawyers are still feel some hesitation when it comes to automation.

"Some people hear automation and their initial reaction is, 'Oh, that's impossible or will that take away from my voice,'" Heller said.

But the reality is that text generation has continued to make strides over the past five years, even if it is still building to successfully replicating a person's style or voice in writing. Dan Roth, a professor in the Department of Computer and Information Science at the University of Pennsylvania, indicated that AI-generated text can now be counted on for grammatical correctness, while overall coherency can still begin to wane, depending on the length of the document.

Overall accuracy also remains something of an issue, for example, with AI-generated contracts, running the risk of omitting certain key conditions. "So you really have to be careful with how this is going to be used and what is the context," Roth said.

However, the accuracy issue that may still prove the most difficult for legal tech providers to overcome is conforming AI drafting tools to an individual client's stylistic expectations. Instilling that kind of purposeful bias or direction would most likely require the ability for a lawyer or law firm to be able to feed a product their own training samples rather than relying on "factory settings."

LegalMation has already performed similar work for clients in a beta testing form and tracked the feedback. "The accuracy scores go way up from the general out-of-the-box model, because [the product] is literally trying to mimic what those lawyers are doing at that firm," Lee said.

So will those kinds of tailor-made AI products become commonplace? There could be both technical and marketplace limitations that slow the development of such features.

Lee thinks that a "one-size-fits-all" product with limited customization likely covers more users and thus would be friendlier to the mass market. But for legal tech companies attempting to reach larger, more sophisticated entities, a higher degree of customization and integration may be required.

At least the AI should be up to the challenge. Per Roth at the University of Pennsylvania, AI can capability identify an individual person's writing style based on factors such as the use of prepositions or frequency of punctuation. That ability has now extended to being able to generate text mimicking that person's style.

"That's something that we've made a lot of progress in," Roth said.