Artificial Intelligence in the Legal Field
I don’t know about you, but when someone tells me that Artificial Intelligence will take the job of a lawyer, I usually assume they don’t understand the legal industry. Sure, computers can replace some of the clerical tasks we do, but AI can’t competently replace a lawyer. It can’t write a brief, negotiate a settlement, or argue a motion in limine. Strategic planning is well outside of its capabilities, right? It can’t adapt, or learn, or make intuitive decisions. So, what’s all the fuss about? Why do we need to pay attention?
You see, my idea of Artificial Intelligence is Watson from Jeopardy, or the computers designed to play chess against the Grand Masters1. These are robust and impressive machines, but they aren’t very smart – not like a human. They’re not reasoning, they’re just working through a problem with an algorithm written for that very purpose.
These aren’t the computers that will take the job of a lawyer. They’re just super-computers using brute-force tactics to solve their problems. For example, IBM’s Deep Blue didn’t make any intuitive decisions when it played Grand Master Garry Kasperov in 1996 and ’972. It merely evaluated thousands (and now millions) of chess moves per second – which is impressive considering the average human can evaluate two3 – and chose the optimal move, time and time again. That kind of computer can’t write a brief.
Additionally, Deep Blue doesn’t learn from its opponent. Adjustments are made between games by IBM engineers, but it has no way of adapting its style to react to what has happened4. It will make the same decision based on its evaluations every time, no matter what the result previously. That kind of self-assessment is key to successful negotiation.
But that’s a limited view of AI. Although technically artificial intelligence – certainly the progeny of AI – those concepts are not the real source of potential disruption. Again, they cannot intuit, or learn from their mistakes. The real potential lies in a developing area of AI called Deep Learning5.
Deep Learning is a subset of Machine Learning (which is itself a subset of Artificial Intelligence)6, where a computer places values on each decision that it makes, reevaluates these decisions based on what occurred, and then weighs each of them against each other (giving them a score). The more data that is fed into the computer, the more information it has to re-weigh each of these decisions. As it gathers more information, its scores become more precise. Essentially, it learns which of its decisions were more correct by comparing them to decisions that it knows are right (obviously it is much more complicated than this, but for our purposes, this will work). This is how a computer can identify cats in YouTube videos without being told what a cat is in the first place7.
AI in the Legal Field
When you think of AI in the legal field, try to get the idea of an anthropomorphic robot moving freely about the well out of your head. We’re not talking about technology that will supplant legal jobs one-for-one. AI will change the way we think of law, and the role of lawyers. Instead, focus on the predictive text in your Google search bar. Yeah, it’s wrong sometimes, but it’s a little bit creepy how right it can be. And it keeps getting better.
Every time you enter a search into your Google search bar, the system gets better at predicting your searches in the future (remember – feeding data into the system). More specifically, it gets better at delivering the result more relevant to you based on your current and previous searches. It’s learning what things you’re more likely to be looking for and weighing them more heavily when it gives you results. That’s not just brute-force. There is a little finesse there too.
Companies like Lexis Nexis, and ROSS Intelligence are beginning to use this predictive technology and deep learning to enhance legal research techniques with products like Lexis Answers8, and the artificial lawyer ROSS9. They are attempting to return answers to legal questions more quickly and more effectively. The users of these technologies should be faster than their unaided counterparts. But we’re just talking about serving up information to the lawyer who must eventually parse through it. This still feels like a very long way away from making strategic decisions.
Additionally, this specific use of deep learning doesn’t seem like it’ll have any effect on how we define a lawyer. Yes, it’ll make us faster, but it won’t encourage us to fundamentally change our practices. There are only so many controlling cases out there, and lawyers don’t spend their days re-researching law that hasn’t changed. How much will the user really gain?
Although legal research is important, it’s not what makes a lawyer. The tough part of law, is applying the facts, determining what your best legal theory is, reading your adversaries, and crafting a compelling argument – in a word, strategy. Until a computer can strategize, it doesn’t have a chance at replacing us.
Man Versus Machine Part II
Once IBM and other engineers proved that super-computers could consistently beat Grand Masters at evenly matched chess, the AI community changed the focus of computer-human board games. The community began focusing on a game called Go10. A game where humans were vastly better than machines.
For the uninitiated, Go is an ancient Eastern game with two players who alternately place black or white polished stones on a 19×19 square gameboard in an attempt to control the majority of the territory. Because of the surface size and the number of potential moves at each turn, it is not a game that lends itself to brute force – there are simply too many options to consider. It’s a game of intuition, and strategy, not of ones and zeros. Many players make moves based on how the board looks and feels, not working out all the potential scenarios in their heads11.
Google, among other companies, has been developing a machine that is intended to compete with the best Go players in the world – AlphaGo12. The idea is that the machine, on one hand, limits the amount of options that it will examine, and on the other, examines the remaining options for their potential for success. As it makes these decision, it uses deep learning to rate them, and ultimately make better decisions in the future. AlphaGO was fed hundreds of thousands of Go matches to learn from, and ultimately, it began to improve its game by playing itself (don’t worry, it was told to do this)13.
Through this process, it has become so proficient, that in early 2016 it beat one of the top human players in the world – five times14. And it did so by teaching itself to play. It did so through deep learning. More recently, in early 2017, it beat Ke Jie, the number one player in the world at the time – three out of three games. And it’s been given an honorary grand master rank of 9 dan (which is the highest rank available)15. This is not a computer that’s just working out all the possible scenarios. It’s making judgment decisions based on past performance.
Right now, this just means that engineers have successfully created a machine that shows signs of intuition and an ability to learn from its experience. They’re not arguing cases yet. But that’s not all we do. I can easily envision a scenario where this sort of machine is used to apply facts to caselaw and determine the optimal use of client resources. Although, that won’t replace us outright, it will redefine what it means to be a lawyer.
- Human–computer chess matches
- Deep Blue versus Garry Kasparov
- One-on-one with Murry Campbell
- Deep Learning
- What’s the Difference Between Artificial Intelligence, Machine Learning, and Deep Learning?
- Google’s Artificial Brain Learns to Find Cat Videos
- Lexis Answers
- ROSS intelligence
- Deep Blue versus Garry Kasparov
- Go (Game)
- How the Computer Beat the Go Master
- AlphaGo versus Lee Sedol