CCBJ: What is OpenText’s Aviator program and overall vision of AI?
Bruce Kiefer: OpenText sees the current versions of AI centered around LLMs as complements to our information management DNA. Content can be organized, enriched, and retrieved through LLM interactions. Aviator is the umbrella term used by OpenText to bring AI into our business applications. Aviators can take many forms, but you will generally recognize them as chat interfaces woven into product lines. There are Aviator initiatives for our content management tools, business networks, BI, developer, security, and of course legal tech.
What is Legal Tech Aviator and what new features have been rolled out or will be rolling out for e-discovery over the last few months and the remainder of 2024?
OpenText has been bringing machine learning innovations to litigation products for over a decade. There have been some difficult problems over the years that didn't have an obvious answer until the rise of useful LLMs in the last four to five years. With the launch of Axcelerate with Aviator in 2024 we released two new features with an additional feature releasing at the end of July:
- Concept Cluster Labels: OpenText has used various forms of clustering in legal tech for a while. Most of these tools have been using complex algorithms to create groups of documents based on the strength of n-grams (more than one word phrases). We all know how to produce the most frequent n-grams in a list, but it was always hard to create a narrative summary. Looking at this problem with an LLM opened up new ways to communicate what is inside a concept cluster. Aviator provides a generically repeatable pipeline to use case level content, prompts, and orchestration to get to an outcome. From this generic workflow, we have and will continue to release improvements that save our customers time. We can now provide those readable summaries of the concept group labels in OpenText Axcelerate.
- Key Document Summary: With Axcelerate with Aviator customers can also take a key document (complaint, meet and confer, brief) and search across many documents and use the LLM to summarize the main points across many documents. We have features to batch summarize individual documents which can be used for large document or transcripts or chats.
- Aviator Review: Releasing soon, Aviator Review, will provide customers with the ability to carry out responsive review across some or all of their documents. Because we understand how critical cost control and certainty are in document review projects we have built in the ability to refine your review criteria against a subset of the data. Once you are satisfied with the results that are being returned on the subset of data Aviator then provides an estimate of the cost to run the review criteria against the larger data set.
What benefits or advantages do you foresee the new Legal Tech Aviator features delivering for enterprise legal departments?
OpenText is lucky to have a group of product managers and developers that are very familiar with eDiscovery and have been through many rounds of refining and enhancing predictive coding and TAR to deliver defensible, faster and less expensive review. We know that novelty is simply that, it doesn't deserve to take time from our customers. We're being thoughtful in the problems we are trying to solve and the benefit we can bring to customers - mostly through time savings. The current LLMs and our quickly maturing way to interact with them are producing interesting results that we are working to bring to the platform as new components of Aviator.
With the state of AI review at present what do you see as the interplay between AI review and human review? What part do you see ALSP and law firm review teams playing?
Over time our industry will begin the transition from TAR (technology assisted review) to HAR (human assisted review). The decade of TAR can briefly be summarized as propagating human decisions. A host of tools ushered in TAR - support vector machines, reinforcement learning, graphs, and more were put to work on finding similar content where a previous decision could be used again. If you found this email responsive, there's over an 85% chance you will find this email responsive too. HAR will be different. Humans will describe what they are hoping to do, and the technology will carry out that work. The heavy lifting will shift from the human to the technology. The humans will be used to validate results, provide additional confirmation, or tease apart tougher decisions where the technology can't make a decision on its own. Customers will of course judge their own risk appetite. There could be cases that demand eyeballs on every document. But other cases may benefit from tilting the work from TAR to HAR. We want to provide the best options for your case.
What do you see as the future for AI review in legal matters?
The rise of managed review provided excellent cost options for carrying out document review. We've reached the end of a long, sustained pricing battle to review documents with people for less money. The remaining frontier is having the technology carry out more and more of the work. It's not hard to see a time where first level review for responsiveness, privilege, and hot documents could be carried out through the HAR model and then manage the next stages. The long history of technology tells us it gets better, faster, and cheaper over time. I expect the same for LLMs and that means more and more work loads could be done via technology.
Published July 25, 2024.
