Skynet, Attorney-at-Law? AI shows Utility in Legal Work.

Matlock: America’s favorite TV show! Or maybe not. But what if Matlock had… robots?

Stay with me for a second here – the idea may not be that far from reality. From, we read:

The robot apocalypse isn’t all bad news: A new study suggests artificial intelligence makes better lawyers than humans do.

LawGeex pitted 20 experienced attorneys against a three-year-old algorithm trained to evaluate contracts. Spoiler alert: the computer won.

“Few would be surprised that artificial intelligence works faster than lawyers on certain non-core legal tasks,” according to the report. “However, lawyers and the public generally believe that machines cannot match human intellect for accuracy in daily fundamental legal work.”

I am certainly willing to believe that machines can match human accuracy in daily legal work. In fact, more than just willing to believe; I have no doubt that machines can outstrip humans in producing highly accurate legal work. Accuracy is a matter of zeros and ones – is something right, or is it wrong? Is it on, or off? Zeros and ones are what a computer does best. Where programmers can figure out how to apply zeros and ones, computers can excel.

Participants were given four hours to identify and highlight 30 proposed legal issues in five standard non-disclosure agreements (NDAs).

Lawyers and the AI, for instance, were penalized for missing an exemption relevant to the contract, or mistakenly identifying an exemption where it was irrelevant.

In the end, LawGeex’s neural network achieved an average 94 percent accuracy rate, compared to the lawyers’ average of 85 percent. And while it took humans anywhere from 51 minutes to more than 2.5 hours to complete all five NDAs, the AI engine finished in 26 seconds.

Score one for the technology.

Assessing accuracy is a cinch for a computer. But where does the software encounter trouble?

The process is not without its complications, though. No existing computational language models could read legalese coherently, making it difficult for the machine to understand technical language. Add to that the high accuracy required in statutory AI training.

Legalese is highly contextual. Understanding to whom legalese applies, and how, requires an understanding of the situation at large. This is why legal disputes are ultimately decided by a judge and/or jury per Western legal tradition, as opposed to administrative law decided on an ad hoc basis by bureaucrats. It is the recognizance that the letter of the law cannot account for every possible legal situation which may arise in daily life, and that an interpretation of the law as decided upon by a judge or jury is necessary in order to apply it in a just manner that also makes sense.

In this area, AI has trouble. Because judgement and context are (as yet) analog concepts that exist outside of zeros and ones, I suspect AI will continue to have trouble.

But where AI can be put to work in a legal capacity churning through documents for accuracy, lawyers will find themselves relieved of an enormous amount of busywork. This is bad news for paralegals who rely on relieving lawyers of their busywork, but great news for a public who traditionally views lawyers as expensive and difficult to afford. When lawyers are able to utilize teams of low-cost AI to work through paperwork, this will lead to a lowering of cost in legal services. What was once not easily attainable to the common man outside of pro bono work – a high-quality lawyer – may become more attainable as computerization of legal work lowers costs and increases quality.

Now go stream a few episodes of Matlock and imagine how much cooler it’d be with robots.

Categories: The IT Philosopher

Tags: , , , ,

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: