Many eDiscovery software providers claim that they have incorporated TAR and artificial intelligence into their platform. These buzzwords are great if the technology is being implemented in a meaningful way.
What is TAR? It isn’t a computer term but one that was created in the legal space. It began with simple technology assistance such as keyword searching and metadata filters. TAR is a process that isn’t predictive but rather systematic. It is not a process that works independently of human input; thus, the word “assisted” within the acronym. TAR is an evolving term containing many different aspects of technology for legal.
It ranges from the utilization of keyword searches, clustering, natural language processing and seed sets which provide various methods of gathering similar records together. So how does artificial intelligence and machine learning fit into the world of TAR eDiscovery? First, we should loosely define what AI is and how it works within the TAR framework.
Intuition or Statistics?
Machine learning is closely related to (and often overlaps with) computational statistics, which also focuses on prediction-making through the use of computers. Translation: computers are great at repetitive mathematical tasks but aren’t actually that intelligent. The way we make decisions based on past experiences and knowledge.
For computers to discern and make decisions also requires a pool of experiences or data sets. These “experiences” or raw data can be coupled with a set of rules or a “model” which machines use to predict outcomes.
Natural Language Processing
One of the machine learning models we use at Cicayda incorporates Stanford’s Natural Language Processing (NLP) algorithms, which analyze all data within an eDiscovery matter before a review even begins. This analysis sorts and categorizes all the key players and subjects within the matter by date, time, name, subject, geo-location, and a swath of other criteria. Intuitive and interactive visual analytics show these relationships to inform your review strategy. Reviewers can also toggle different criteria and see how the data change instantly, regardless of data size.
Cicayda’s way is one method and definition of TAR within the legal space but is certainly not the only one.
How to Train your AI
It is imperative that you know what you want or what problems you are trying to solve when considering the use of AI or TAR in your eDiscovery workflow. Machine learning comes in many different flavors, and some may not provide any advantage to your workflow. What we see at Cicayda as the most relevant application of the various machine learning technologies are as follows.
- Pre Analysis – Data analysis across an entire matter, giving reviewers a high-level overview or starting point for review.
- Review Model Training – A machine learning model that observes how each reviewer moves through a matter in real time and can begin to provide relevant suggestions or documents to a user.
- Fully Trained AI – Once your AI has observed your review habits for a long enough time, the model should be able to complete a review autonomously with the click of a button.
- Review Comparison – Users should also be able to compare how their completed review and the review of autonomous AI model trained from their habits align and differ.
If you are considering an eDiscovery platform that boasts TAR, Artificial Intelligence, or machine learning ask the questions below.
Does Technology Assisted Review incur additional fees?
How does your version of Technology Assisted Review alleviate my workflow?
Are you implementing Natural Language Processing?
How do you train your AI model?
Do you need experts to help you evaluate your options and utilize the best technology for the matter?