With any new technology, one must balance buying into the hype too soon with failure to adapt, and when it came to the application of cloud solutions to ediscovery, I was confident in placing them in the hype category. After a comprehensive nine-month process of vetting virtually every platform on the market, I came to the conclusion that cloud platforms were mostly marketing fluff, providing negligible advantages over on-premise tools.
This belief persisted over several years. That is, until I looked under the hood of DISCO, and found a platform purpose-built to take advantage of the powerfully scalable and elastic capabilities that are unique to cloud-based neural networks.
As my recent move to become the Chief Innovation Officer at DISCO should show, I am convinced that not only is now the time to adopt cloud optimized technology, but that those that lag in adoption will find themselves at a major disadvantage compared to those that act decisively. Simply put, I love to win, and DISCO is clearly the winning horse in this race to the next evolution in ediscovery.
I did not come to this revelation lightly, so I wanted to provide the factors and insights that convinced me — and provide background on the neural networks are the catalyst for this shift.
So what changed my mind?
Despite promising marketing, the earliest cloud platforms offered the same structural model as on-premise tools — with the same limitations.
Rather than optimizing the virtual, infinitely scalable and decentralized capacity offered in the cloud, or seeking to capitalize on the rapid scalability and computational power within the decentralized cloud networks, the providers sought to exactly mirror current solutions, warts and all. Traditional computing, bolstered by human-trained algorithms, was relabeled “AI,” but the reality was nothing of the sort, as anyone who has dealt with imprecise and labor-intensive technology assisted review (TAR) and analytic tools can attest to.
DISCO took a fundamentally different approach, building a platform to capitalize on the strengths of the cloud and leveraging a neural network structure.
The resulting platform capitalizes on the natural scalability and elasticity of the cloud, super-charging attorney decision-making and drastically reducing both time to insight and cost. The sheer computational power of this tool far outpaces the traditional on-premise solutions and is the clear leader in the cloud race.
Beyond all the hype, what the heck is a neural network and why should I care?
Neural networks are computer systems that leverage a structure similar to the human brain to dramatically increase computation power at a fraction of the cost and time. While rather new to the ediscovery sphere, neural networks are a mature subcategory of deep machine learning first conceived nearly 70 years ago (predating the computer, in fact) that have become more accessible as computing power has increased and more data is available.
The neural networking approach is fundamentally different from the traditional computing model that requires human input and direction (either explicitly or via algorithms) to derive answers to questions we already understand.
Rather than remaining dependent on the limited human capacity to help it extract insight from data, a neural network has thousands or millions of individual processing nodes (“neurons”) working in parallel to analyze examples and extrapolate connections and solve a discrete problem. As this is all done in parallel, a well-constructed neural network optimizes computational power and reduces time to insight.
The neurons take in inputs and provide data outputs that are refined across successive layers. They weigh incoming data and determine if this adds value to the next layer of neurons. Data that is not relevant to the problem is thus weeded out, and an increasingly refined subset of data is produced. The model, and the complex machines powering it, yields precise results from even disparate and imprecise data sets like human communication.
We see deep learning powered by neural networks is around us every day (i.e., facial recognition and language translation). And the potential to the practice of law is staggering.
Gordon Calhoun, Partner and Chair of the Electronic Discovery, Information Management & Compliance Practice at Lewis, Brisbois, Bisgaard & Smith LLP, summed up the transformational power of this approach:
“Deep learning neural nets have the potential to take us beyond text data, allow us to better understand the evidentiary value of the cocoon of data that surrounds us and use the IoT as a virtual witness in the courtroom. I expect these advances will transform many other areas of the practice of law, as well.”
But the inter-webs told me deep learning only works on massive data volumes?
Historically, due to the limitations of both computational power and data set size, deep learning powered by neural networks was limited to petabytes of data. As a result, providers without purpose-built cloud platforms claim that data sets in ediscovery are too small to use deep learning and that it is all hype.
The reality is that their systems were not built to execute deep learning — and creating an approximation will always yield inferior results.
By contrast, DISCO, as a native cloud technology, has the advantage of massive GPU (graphical processing units) compute-on-demand to power the latest machine learning technologies and algorithms (such as Google’s Word2Vec and a series of convolutional neural networks) to deliver higher levels of classification accuracy, faster than ever previously seen in the legal space.
As a result, the platform’s ability to correctly predict the likelihood that a tag should or should not be applied to a document is consistently in the 85% to 95% range, even with as few as 50 examples and datasets as small as 2,000 documents. This means that although traditional TAR has consistently floundered with smaller datasets, even the smallest case can see benefits in cost savings and speed to insight when powered by DISCO.
So what does this all mean for ediscovery and the practice of law?
As a former (and not terribly good) gamer, it makes me think of a recurring line in the game Portal, “The cake is a lie” — only in this case the AI is a lie. Not in terms of the application or the effect it will have on the practice of law and most aspects of life. Rather, the lie is that it has been applied as broadly as marketing gurus would have you believe. The truth is that TAR, CAR, and all of the many “AI” tools dependent upon human guidance and algorithms are a far cry from actual artificial intelligence, deep learning or otherwise.
By contrast, DISCO sits at the bleeding edge of the ediscovery universe in that it has a purpose-built cloud platform and infrastructure designed to go beyond what sentient beings can do, drawing insights and connections beyond human perception. With the rapidly increasing volume, variety, and velocity of data, the practice of law needs companies like DISCO that are innovating and planning not just for how to improve today but how to continually evolve as our risks, obligations, and obstacles evolve.
So when people ask me why DISCO, I answer, how could I choose anyone other than DISCO? We are at the precipice of monumental change and I fully intend to stay not only relevant, but to shepherd in the next chapter of discovery and the practice of law.