Since OpenAI made ChatGPT publicly available in late 2022, artificial Intelligence has taken the world by storm, from academia to industry to everyday answers for the masses.

Yet AI and machine learning are no strangers to many industries, including legal and, notably, document review, the costliest part of the eDiscovery process. The harsh reality of modern eDiscovery is that as document collections continue to grow without a commensurate increase in the time available for review, mounting pressure is placed on review teams to develop techniques to maximize their efficiency while simultaneously ensuring high quality.

In the past, lawyers reviewed hard copy documents one-byone to find relevant information. Today, with matters containing document volumes often in the hundreds of thousands if not millions of documents, setting eyes on every document is simply not realistic. Approaches to eDiscovery document review range from purely human driven processes such as keyword search and followed by linear review to predominantly AIdriven approaches using various forms of supervised machine learning. This includes technologyassisted review, also known as TAR or predictive coding.

The objective of AI or machine learning approaches is to minimize human review (and wasted effort and dollars) while maximizing effectiveness by prioritizing the most likely relevant documents for review. The benefits of applying AI approaches for document review are widely understood:

– Increased speed: TAR can help to automate many of the tasks in document review, which lead to time savings.
– Improved accuracy: TAR can help to identify patterns and trends in documents that humans may miss, resulting in more accurate results and fewer errors.
– Reduced costs: TAR reduces the cost of document review - automating tasks and improving efficiency by enabling reviewers to set eyes on the document to be most likely relevant first.
– Increased productivity: TAR can free up human reviewers to focus on more strategic tasks

Approaches to TAR

Just as TAR has many acronyms, there are also many approaches to TAR – fundamentally all of which leverage human judgments about documents to assist the computer in finding more relevant ones.

Earlier protocols
Preliminary protocols, simple passive learning (SPL) and simple active learning (SAL), rely on one-time training. In essence, a senior lawyer or subject matter experts tags a reference, or control, set and then reviews a few thousand documents to train the algorithm against that reference set before running it over the entire document collection. These approaches were shown to have limitations, as discussed later in this paper when looking at some of the limitations of using TAR and opportunities posited by the newer continuous active learning (CAL) protocol.

Continuous active learning protocol
In recent years, one of the most efficient approaches to TAR involves a combined human-machine approach based on the continuous active learning (CAL) protocol. CAL, known as TAR 2.0, is an extension of the active learning methodology where the learning process is continuous and integrated into the review process. As reviewers code documents, the system continually learns and updates its understanding of what is considered relevant, thereby improving its suggestions over time.

To read the full whitepaper discussing conduct concurrent first pass and secondary review workflows for greater cost savings and faster time to results click here .