Unless you have been living under a rock, you probably have heard the phrase artificial intelligence (AI) in the context of law more times than you care to count. From AI-powered contract analytics to advanced search visualization and beyond, it seems as if any software company worth a damn in the legal space is touting machine learning and AI like there is no tomorrow. However, when you look under the hood, many of these technologies are marketing razzle-dazzle rather than true AI.
Thankfully, times, they are a changin’.
With the recent release of true cross-matter AI, DISCO has taken legal technology squarely into the realm of AI-amplified law. Unlike the traditional TAR (technology-assisted review) models used by many competitors, DISCO’s cross-matter AI models are driven by abstract semantic representations rather than keywords or phrases. When organizations leverage model sharing, they are going beyond the “bag of words” approach historically employed and sharing insights about the underlying mathematical architecture of the data and relevant concepts. Leveraging a vector of 300 numbers via a technology called fastText, a DISCO AI model learns to respond to these representations and can quickly offer insights across a wide data ecosystem.
What does this mean? Real AI-powered insights in a fraction of the time, regardless of your matter size.
How is DISCO cross-matter AI different?
In my past life, running a global ediscovery program and serving as a discovery SME at the big four, I had the opportunity to work with every emerging technology out there. One of the largest hurdles I faced in everything from complex multinational litigation to regulator-driven investigations was gaining a quick understanding of the data universe and identifying the key facts necessary for case development, settlement posture, or approach for a regulatory investigation. Many “bleeding-edge” tech companies claimed they had portable models or the ability to take insights from past machine learning-powered reviews and use them to reduce the time necessary to gain similar insights from a new data set.
True model sharing would have been amazing for my clients with massive litigation portfolios and/or many similar matter types. Unfortunately, the quality of what I got was less “holy grail” and more “red Solo cup.” Rather than a nuanced and multi-vector model, I was faced with exporting a spreadsheet of words and phrases that may or may not have helped my new matter. This manual approach was suboptimal for a multitude of reasons including confidentiality concerns, efficacy of the weighted insights provided, and improvement of speed to evidence. Instead of an advanced algorithm, all I got was a vocabulary list.
Beyond words and phrases
In decoupling the underlying model from discrete words and phrases, DISCO AI is able to dig deeper into the linguistic nuance of coding decisions and better inform the future model, DISCO AI examines the document, taking into account the order, meaning of words, and sentence structure to arrive at intelligent insight as to whether the document is of substance to your review. This allows the underlying algorithm to understand correlations like man is to king as woman is to queen as well as differentiate between “Man eating chicken” vs. “man-eating chicken.” This context yields a more refined model that can provide more accurate insights when shared to a new data set than a simple vocabulary list.
Legacy model sharing generally required physically exporting the model in a static form and uploading it to start a seed set for the new data set. New insights gained from interacting with the model in a new data set were stuck in the new data set and could not be incorporated to improve and refine the original model and the scale of insights that could cumulatively grow was fairly limited. DISCO created an interactive two-way model sharing approach that allows you to easily create a feedback loop between some or all of the cases interacting with a given model. This approach greatly amplifies the learning potential for each model while keeping control of the evolution of the model with the case teams. This two-way street allows for amplified learning to help accelerate insights in future data sets with the flip of a toggle.
Better models mean faster insights and your decisions amplified.
Battle testing models
As a continual guardrail to model degradation, the system was designed to continually weigh the efficacy of each new model. What does that mean? As the model evolves, the accuracy and results are continually compared to the past model to ensure that only the most accurate model is deployed on a go-forward basis. This dynamic approach keeps the model on track while ensuring the case team gets the most accurate recommendations possible. This approach effectively nullifies the risk of leveraging AI.
Compared to the training rounds, statistical validation, and manual Excel export of legacy tech claiming to offer model sharing, DISCO’s integrated and user-friendly approach to AI is a game-changer. Users simply have to toggle on AI or AI model-sharing by model topic to begin getting the results. No alternate workflow, no complex data ingestions and staging in a middleware tool — just click and get better insights. This ease of use combined with a price tag of free (!) for DISCO AI takes away many of the barriers to use I faced when dealing with case teams.
Netflix is so easy my five-year-old niece and 80-year-old grandfather can easily interact with the queue to get the type of movies they like. DISCO has made AI as intuitive and easy to use. Attorneys leveraging AI simply get better data sooner when deploying DISCO AI and cross-matter AI without having to learn a complex workflow.
Did I mention it's free?
Unlike legacy tech which can cost between $50-150 per GB or possibly require acquiring new servers and licenses, DISCO includes AI and cross-matter AI on every case for every customer at no charge. The belief is that cases using AI get better insights for less time and money, we want to remove any hurdles from using it in every case.
What does this mean for legaltech?
Cross-matter AI (and frankly DISCO AI) moves legaltech out of the metaphorical Stone Age and squarely into the 21st century — no more spreadsheets with vocabulary lists masquerading as AI. Using the AI to help sift through documents during early case assessment can really take ECA to a new level. Normally you’d just be looking at your documents' metadata during ECA (file types, email domains, etc.). But, with cross-matter AI now you're getting insights into the actual contents of your documents.
Additionally, because cross-matter AI benefits from the deeper insights provided by fastText’s multi-vector approach, two-way feedback between cases, model sharing, and continual A/B testing of the models, users get dramatically better insights in a fraction of the time and without the risk of the legacy manual approach. This negates many of the existing fears around leveraging AI and combined with the simplified user interaction with DISCO AI will help dramatically increase adoption of this fundamentally better approach to uncovering key insights in ediscovery.