The business world has been hearing about the impactful changes being brought by artificial intelligence (AI), blockchain and related innovations, from whispers around the watercooler to a key topic of discussion at this year’s World Economic Forum in Davos, Switzerland.
In a new book, AI Concepts for Business Applications, author and litigation attorney Nelson E. (Nick) Brestoff, offers a “no-math explanation” of the breakthrough story of one form of AI known as deep learning. Brestoff contends deep learning will change everything and challenges his readers to come up with business-relevant innovations of their own.
We spoke to Brestoff, who was a California litigator for 38 years and an inventor, about this innovative technology, his eight patents in this area, and how lawyers and law firms can use deep-learning to identify and provide an early warning of either risks to avoid or financial advantages to obtain.
Why did you write this book at this time? What issues were you seeing that needed to be addressed?
Nick Brestoff: I wrote the book for a number of reasons. First, because AI is a label that’s overbroad. I wanted to share what I’d been able to learn by realizing, in late-2015 and early-2016, that a leading academic and practitioner, Andrew Ng, was likely correct to assert that, just as electricity changed everything during the Industrial Revolution, AI in the form of deep learning was going to change everything now.
Second, I learned that deep learning is just the nickname for software that’s actually a multi-layer neural network — similar to the networks of neurons in our brains — that takes advantage of today’s superfast computers that also have massive memories. I also stumbled onto the notion that what deep learning was doing was using examples to build a model for a topic of interest and then assessing the images or text (actually, the data) that the model had not seen before. We had learned to get computers to do pattern-matching, simple as that.
Third, the problem was that business leaders, including in-house counsel, can’t scan or assess what’s in that ocean of images and text that every enterprise generates every day, and so they don’t even look.
Fourth, I had a quest. Having been a litigation specialist for 38 years, I wanted to see the litigation “smoking gun” emails in time to conduct an internal investigation and, hopefully, nip the risk in the bud.
Last but not least, in late June 2016, I read the opinion of the U.S. Court of Appeals for the Federal Circuit in Bascom Global v. AT&T Mobility and realized that a software system using deep-learning models was potentially patent eligible, notwithstanding the U.S. Supreme Court decision in Alice v. CLS Bank. Now, three years later, and after all of my eight applications for patents were granted by the U.S. Patent and Trademark Office, I wanted to explain how the patent applications were written.
Bottom line — I’m close to passing the torch to a younger generation, and I needed a book to explain my journey.
As you know, AI and its ilk are viewed in the marketplace with a mix of hype, awe, and fear — especially in the legal industry. How can legal practitioners get past that mentality and view AI and deep learning in way that could benefit their practice?
Nick Brestoff: The main theme of AI Concepts is to explain why deep learning is already changing entire sectors of the economy, from agriculture to social media; and that anyone — even without having a science education or learning how to write code — can climb on top of what’s going on. Then they’ll be able to see the business problems that the tech personnel don’t appreciate and will be able to interact with the data scientists and programmers who can come up with solutions.
And, just as deep-learning models are created from examples, in the book I use my patent applications as examples for getting across the point that understanding deep learning doesn’t require math or programming skill. There’s no math in any of my patent applications.
More specifically, what can a legal practitioner do to innovate his workflow or develop his practice using these concepts?
Nick Brestoff: In the legal tech space, my focus was on in-house attorneys and how they deal with potential litigation. I was not seeking more efficiency; I was inventing an entirely new workflow. In my first patent, I described an “early warning” system that would enable prevention. Working with only one software developer and one data scientist, we successfully built a model for employment discrimination which we tested on the Enron dataset and found a pre-litigation example of it. And that’s just one category. In federal court, the judiciary requires each plaintiff to name only one of its many categories of litigation, such as, antitrust, breach of contract, or fraud, etc., when the complaint is filed. Needless to say, there are many examples of each category, and by examples, I mean only the factual allegations.
Down the road, there will be databases for specific types of lawsuits in the state courts as well.
Then, with models for each type of risk, all of yesterday’s emails can be assessed by the models (using parallel processing) to provide in-house teams with an insight to each risk in near real-time. Once the prevention ball starts rolling, I predict that adoption will be widespread because the internal ability to avoid litigation will be seen as a competitive advantage.
What would you most like to see a reader come away with from your book?
Nick Brestoff: As long as a reader can identify a business problem, and examples of it, a deep-learning model can be ginned up, tested, and used as a filter to present reviewers with only the small amount of data that pattern-matches to the model of the problem.
That’s where human reviewers come in. Of the data that’s flagged as related to each problem, they’ll have questions: Which is it? Is it a False Positive or a True Positive? (Either way, let’s at least save the True Positives to re-train the model from generic to company-specific.) Should a further investigation take place? If confirmed as a True Positive, who else is involved? Does it violate one of the enterprise’s policies, for example, those against discrimination or hate speech? Should the matter be escalated to a decision-maker?
My readers should come away with the appreciation that while no one may ask these questions now, with deep learning, the questions can be asked and answered.