Predictive Coding

[jcoyne]

http://www.out-law.com/en/articles/2016/february/uk-court-approves-use-of-predictive-coding-as-basis-for-e-disclosure-for-the-first-time/

 

In light of the UK approving the user of Predictive Coding as part of the document review, I would like to see Vound either implementing a Predictive coding engine into Intella or providing an interfacing mechanism (perhaps similar to the external OCR process?) into the product. 

 

I know that the Intella has grown up in the forensic search space but a significant part of its future use could/should be in edisclosure review.

 

Any views would be welcomed, even is it's 'no way!' 

[admin]

Hi Jason,

 

I think Fletcher from the article was right when he said

 

"I don't therefore see this ruling as opening the floodgates for predictive coding, as cases will need to be of a certain value and have sufficient quantities of electronic documents to make it viable." 

 

Some years ago when the US courts did something similar we were inundated with the question ‘do you have predictive coding in Intella?’  Walking around Legaltech that year it was all the buzz - predictive coding brochures and signs on so many booths. 

 

It surprised us that a number of callers wanted to know if we had this feature called predictive coding. When we tried to explain, most did not  want to know what was involved in terms of sampling/training with subsets, the cost of having skilled and knowledgeable users on hand, the systems it may require and so on. In short it is a process that has its own requirements and not a button.  More importantly, did their firm attract cases that could benefit from it on a regular basis, regular enough to justify the costs and return a profit on their investment?   They just had to have it because everyone was talking about it.

 

This keeping up with the Jones attitude was IMHO driven not by need but by a number of vendors’ blatant marketization of the term predictive coding. This was coupled with a number of the usual suspects in the form of legal bloggers pushing this as an agenda to try and drum up sales for specific vendors. Those vendors will continue paying to advertise in return for more favourable reviews of their predictive coding offering and product. (IMHO paid bloggers or #cash_for_comment bloggers are blight on our industries. I have no problem with bloggers getting paid, but they need to make it clear when they have a tie of some kind to any vendor they write about.  Some do this openly, some do it in a way that is well hidden to the first time reader, so they are appear to be an independent reviewer when they are not). We don't hear about predictive coding nearly as much anymore. (We see a similar frenzy of marketing around the current buzz term "Information Governance".  Again a very real area but some vendors have seen dollars and let the marketing hounds off the leash to go get the money anyway they can. Again, Information Governance is  not a feature but a process much broader than just one software package).

 

We looked at what the vendors in this space were doing for predictive coding (PC, TAR, CAR, etc) and we found that most were more smoke and mirrors than real features. There were only a few doing "machine based learning" that has been around for a long time and really qualified as predictive coding.  Not all predictive coding is created equal!

 

So to answer your question - we have coders  that have extensive experience with AI / machine based learning subjects, so we are well placed to understand and make a valuable contribution to this field.  I think it is inevitable that we will add some form of "machine based learning" to assist in the identification of data. However we want to offer it in the same way as all our features - our mantra is being the easiest product on the market to use.  When we add it we want every case and user to benefit, not just the multimillion dollar cases that require huge lit support resources.  

 

For now, if you need to have predictive coding for a specific case, we are happy to recommend, off list, the few vendors we see as real options. 

 

Lastly, do have a read of the EDJ adoption articles for predictive coding. Greg has a good, independent and research based view on this.  He is not a #cash_for_comment blogger.