By Jason Wilson
Thanks to Mike Shatzkin, I started thinking about January 2011. It’s like reading those fools over at Publishr, who have this annoying habit of making you think about what will be, rather than what is now. I hate it, but you know, secretly love it. Seriously, do you know anyone in the legal publishing business blogging about anything interesting?
Ah, no.
So let me say right off that I don’t like trying to make predictions. But I’m listening to the new Black Keys‘ album—which is fantastic—so I’m feeling emboldened to make one (yes, I bought it directly off their site; I am a loyal fan):
By January 2011, Thomson Reuters Legal will announce a new category of legal content (READ: book) that is composed entirely by algorithms.
I wouldn’t have gotten on this topic if it hadn’t been for Episode 58 of the Thomson West’s Podcasts. I remember seeing this video last year and thinking, “it took him two and a half hours to write 21 headnotes for a U.S. Supreme Court case.” And for me (a bespoke publisher), that was fine, but I kept thinking, for Thomson Reuters, there was no way that was okay. After all, the Heller case (the subject of Episode 58) was released nearly two years ago, and I’m wondering (as Toby Brown suggested), whether West editors have figured out how they might be obsolete, and right quick?
Fast forward to April 2009, this piece shows up in City Pages referencing Thomson Reuters’ Categorization and Recommendation Engine (affectionately known as CaRE):
The computer uses a system of statistics, including Bayesian probability, to predict where documents should be categorized. CaRE suggests key numbers for new cases, identifies cases affected by a new decision, and performs a host of other tasks. Before CaRE, West had hired freelance attorneys to do this work. Now, a computer can do it more quickly and more accurately.
“A computer can do it more quickly.” Where have I heard that before?
Seriously, I had no idea what CaRE was and I have toured the entire facility twice! I mean, from top to bottom. [fn *]
So, I asked the Internet for more information about CaRe, and this pops up, from 2007. Read it. Seriously, it’s about ResultsPlus, and it’s extremely interesting. I did, and I thought, sheeiit, Peter Jackson is, in fact, really, really smart. And that was three years ago. Before WestlawNext and WestSearch.
I have every suspicion that my prediction is either close to coming true or has already become true with Thomson West’s KeyRules products. I don’t know that for a fact, but these are some clever chaps, so I wouldn’t put it past them.
The question is, how do you feel about relying on content completely assembled by an algorithm? And better yet, if you are an author, how long do you think you can continue to write before AI takes over, completely?
[Image (CC) by K!T]
_________________
[fn*] I have had the pleasure of touring the entire facility, from the editorial floors to the farm. I have always enjoyed visiting the Eagan facilities, and am convinced that the folks there honestly want to make your (i.e., legal researcher) job easier, faster, more efficient. I just disagree on how that should be accomplished. Oh, and how much it should cost you to find the answer.
{ 4 comments… read them below or add one }
Unfortunately, as is so often the case with technical journalism, only the facts have been changed!
CaRE technology does suggest key numbers for case headnotes, but these suggestions are always reviewed by editors. The case history task is still performed manually. I described an old research project called History Assistant to the journalist, and this was lumped together with CaRE, even though they are totally unrelated and the former was never fielded.
What CaRE does do every day is perform a host of other categorization tasks, such as routing case headnotes to sections of analytical materials where they should be cited. (This is indeed work that used to be done by freelance attorneys.) CaRE also plays a major role in ResultsPlus and WestSearch.
Peter,
Thank you for taking the time to clarify the role of CaRE within the Westlaw system. The History Assistant project sounds interesting. Would you mind expanding on it for the readers (and me too)?
Actually, the real question is whether you've conquered the "lack of an inferential capability" problem yet in text mining?
History Assistant is a moribund project, but you can read the whole story here:
Peter Jackson, Khalid Al-Kofahi, Alex Tyrrell and Arun Vachher (2003). Information extraction from case law and retrieval of prior cases. Artificial Intelligence, 150, 239-290.
I think the inference problem remains unsolved, despite all the talk about 'semantic technologies'. Most of the inferences supported by current systems are trivial ones.
You must log in to post a comment.