E-Discovery
Have Our Robot Overlords Finally Arrived?
March 16, 2018
Have Our Robot Overlords Finally Arrived?
If you look at some article titles from the past year dealing with artificial intelligence (AI) and the legal industry, you could be forgiven for thinking so.
The robot lawyers are here—and they’re winning
‘World’s first robot lawyer’ now available in all 50 states
These articles aren’t just appearing in niche publications—they’re on the BBC’s website, Atlantic Monthly, and tech culture media outlet the Verge. AI is once again an ascendant technology, and as the popular thinking goes, it’s different this time.
Recent artificial intelligence developments in the legal field include:
- Chat bots like DoNotPay that automate basic legal tasks (such as fighting parking tickets)
- Case outcome prediction tools
- Software that helps set bail and determine whether to grant parole
But if you look at law firms and in-house legal departments, AI has not taken hold to the extent these articles proclaim. According to the Blickstein Group's Law Department Operation Survey in 2017, 66% of legal departments are not using AI.
So which is the truth? Are law degrees soon to become worth less than the paper they’re printed on? Are lawyers going to join journalists, cabbies, and factory workers on the list of dying occupations? To answer that question, we need to understand how AI is used in the legal field today, and how it differs from previous generations of artificial intelligence applications.
The first generation of artificial intelligence, pioneered in the 1950s and in ascendance through the 1980s, was rule-based. Rule-based AIs were programmed to analyze situations using rules hard-coded by human programmers. They were, at a fundamental level, limited by their creators (presumably reducing the risk of Terminator-style scenarios where they rise up to take over the world.)
But that doesn’t mean rules-based AIs were primitive or inferior. Deep Blue, which beat world chess champion Gary Kasparov, was a rules-based AI.
The next major generation of artificial intelligences were machine learning AIs. In traditional (or supervised) machine learning, “A human gives lots of examples to the computer and gives it the correct answers to a particular problem. If the training goes well, then the system can predict answers down the line for the rest of your data with a high level of accuracy,” explains Kelly Atherton, Director of Analytics and Managed Review at Night Owl.
In e-discovery, supervised machine learning AIs (often referred to as predictive coding or technology assisted review) are accepted as “black letter law,” but few would argue they’re a threat to the legal profession. They rely on human-generated and coded seed sets, so if not as rigid as rules-based AIs, they are still limited by the competence of their human masters.
However, the rising crop of legal AIs are another quantum step forward from these supervised AIs. They are “deep learning” or “unsupervised” machine learning AIs. Not dependent on seed sets and human inputs, deep learning AIs discern hidden patterns in the data and use them to draw conclusions and make recommendations. Atherton explains, “Unsupervised machine learning AIs are able to identify hidden patterns in the data, and they make them more readable and organized. Generally speaking, with deep learning it's going to learn from observational data and examples and it figures out on its own what rules to apply or how to solve a problem at hand.”
If that sounds a little creepy, though, there’s honestly no reason to be afraid—at least in the realm of e-discovery. Atherton continues, ”In e-discovery how we would see this, if we're looking at some documents, a data set, is that instead of just looking at the text as words or phrases, and looking how many times the documents appear in the document, the deeper learning algorithm is going to first interpret the text as a series of characters, it's going to take what it learns there, move it to the next layer, which maybe is looking at syllables, and then to the next layer looking at how that data forms inputs and so on.
“Basically, I think what it does is just going to help us organize the data faster and better. Examples of unsupervised machine learning within e-discovery would be something like clustering where without an input from the human the machine is just going to take all the data, look for patterns into it, group it into similar groups documents, then it's up to the human to decide what's important in there.”
Even with the fairly substantial leap forward from machine learning (and predictive coding) to true deep learning AIs in e-discovery, lawyers don’t yet have anything to fear. As Arthur C. Clarke said about technology, “Any sufficiently advanced technology is indistinguishable from magic.” So for the legal profession, an admittedly technology-resistant culture, deep learning AIs may appear to be coming for their jobs. But in reality, they are still simply a highly effective technology that stands not to replace lawyers, but to improve their ability to perform their duties efficiently and effectively.
If you’re interested in finding out more about practical use cases for artificial intelligence, check out Exterro’s Essential New Technology webcast on demand today.