http://ow.ly/cnlvv
An article by Monica Bay posted on law.com on the LTN webpage. The article examines comments made during sessions of the Virtual Corporate Counsel Forum, from the panel discussion entitled "Debunking the Myths about Predictive Coding". The comments discussed by Ms. Bay are attributed to David Kessler, Esq. of Fulbright & Jaworski, and Howard Sklar, Esq. of Recommind.
The article states, "The two lawyers explained some of the basic concepts of technology-assisted review, such as how sampling is used to determine baseline responsiveness, and how prioritized review results in responsive review batches. "Sampling is used to determine baseline responsiveness," said Sklar." At the beginning of the process, you are trying to get a key set of relevant documents. Then, you can 'interrogate' those documents with sophisticated technology" to establish which documents are most appropriate.
Once you have found those documents, then "you can train the system on relevant documents, so you can find 'more like this,'" Sklar said. The process takes "interaction between the computer and human feedback to define the relevant terms."
The article also provides a link to a PDF version of the powperpoint presentation covered by the panelists.
The article also discusses the four steps of predictive coding:
Step 1: Using predictive analytics to create review sets, with human review.
Step 2: System training on relevant documents (computer suggestions).
Step 3: Human review of computer suggestions: "adaptive identification cycles" (train, suggest, review).
Step 4: Statistical quality control validation.
In addition, seven myths about predictive coding were also discussed, including:
- Predictive coding is automated coding
- Defensibility depends primarily on the technology
- Predictive coding is inherently risky
- Culling documents using predictive coding is risky
- Technology is replacing judgment
- 99% is reasonable and 95% is not
- Transparency is the answer
No comments:
Post a Comment