Application of artificial intelligence (AI) to business law is the subject of much hope and some hype among legal tech promoters, a handful of forward-thinking law professors, alternative legal services providers — and their avid followers in the legal media.
Which brings me to the other (much larger) group — law firm lawyers who base their livelihoods on billable hours and pursuit of associate leverage — and in-house lawyers who pretty much take their professional cues (they’d contend vociferously that this isn’t the case) from what their counterparts in outside law firms do.
Among this group the application of AI to business law meets with one of two responses:
- Harrumphing skepticism, or
- Damn-with-faint-praise — “we’ll look at this — later”.
AI is quite real.
It will be a significant factor in the legal industry.
Some applications won’t work. Others will. And failure vs. success will largely be a function of access — or lack of access — to available data.
Stephen Embry — a member of Frost Brown Todd LLC who litigates around the country — writes:
” … The true limitation of AI is not its technical capabilities. It’s not the availability of the data. It’s the access to the data that exists.”
And for purposes of applying AI to your company’s civil litigation, this “data that exists” is of three types:
- Proprietary access only — such as data about the settlement of cases held by a law firm or company).
- Publicly available — and easy to get (see my next post’s coverage of AI tools for legal research and for document review).
- Publicly available — and hard to get (see my next post’s coverage of AI tools in “legal analytics”).
Use of AI to predict outcomes in civil litigation isn’t happening any time soon — except within very law firms or very large companies whose volume of proprietary data will be deemed statistically sufficient to anticipate settlement results.
For AI’s use in predicting civil litigation outcomes “the data that exists” is held mostly in fragmented law firm and company silos — this is the proprietary access only category.
Let’s say your company faces an employment lawsuit. Then let’s narrow that down to — say — Title VII sex discrimination.
First off, the vast majority of those cases settle (the vast majority of almost all categories of cases settle).
Almost all settled cases in Title VII sex discrimination contain confidentiality provisions — “gag” clauses — as a condition of settlement. A litigant who discloses the dollar amount or other terms risks losing what they got in the settlement.
The Title VII sex discrimination litigant wishing to use AI to predict outcomes therefore has a problem getting access to the data that exists.
Aaron Crews, Chief Analytics Officer at Littler Mendelson — one of a handful of practices that has a large volume of such Title VII sex discrimination cases — contends that lack of access to relevant data about past settlements effectively prevents use of AI to predict future settlements:
” … I don’t need you to tell me whether I’m going to win. I need you to tell me if I should settle the case and for what. That takes data about previous settlements and outcomes not publicly available.”
How might you get a sufficient universe of settlement data to which to apply AI?
Unlike the data universe consisting within published judicial opinions — for instance — there is no place to look this up.
You might think of combining the internal Title VII sex discrimination case records of Littler Mendelson, Seyfarth Shaw LLP, Ogletree Deakins, and Jackson Lewis. Each is a legal powerhouse with lots of experience defending companies in Title VII sex discrimination lawsuits. Or you might think of companies doing so directly.
But law firms just don’t collaborate that way. And I’ve never seen a company do anything like that.
Hence my view that AI won’t be used to predict outcomes in civil litigation any time soon — except where single large law firms or large companies have sufficient data to work with — among their own records.
Yet use of such single-source data will raise problems of its own: Application of AI to empirical evidence is challenging enough without having to rely on an incomplete universe of data.
See Part II and Part III of this series.