Peer to Peer Magazine

Summer 2019: Part 1

The quarterly publication of the International Legal Technology Association

Issue link: https://epubs.iltanet.org/i/1136335

Contents of this Issue

Navigation

Page 28 of 67

30 used by the Legal AI products. If it gets the right results, it's less important to understand "how" it creates the results. You don't need to know the academics of AI to understand how various capabilities can help your business. There are many types of complicated technolo that users don't understand how they work at a detailed level, but still come up with new and innovative uses for that technolo. "Automation" or "Automatically Making Decisions" We think of AI as performing cognitive work. It's not really "automation", it's the ability to analyze lots of data (and the key is having lots of data) automatically making decisions. "Automation" and even Robotic Process Automation (RPA) are defined processes where automated decisions become part of a process. They don't necessarily require AI at all. This is another area where there is a lot of technical information available and in the end, we only need to understand enough in order to make decisions on what Legal AI to choose. Training and Trusting Legal AI When a human goes to their first job out of college, we don't expect their intelligence will make them top in their field immediately. We coach them and train them. We let them shadow experts and we supervise them. Yet we expect that when we hire "artificial intelligence" to come in and immediately tell us things about our documents or databases, that they will be perfect immediately. We don't really coach and train AI. Granted, that training and coaching might look different than with a human, but we need to somehow let them grow into the responsibility with decreasing oversight. It should be not trustworthy right away, otherwise it's not artificial intelligence, it's just a script. An analo is automated driving. These cars get into fewer accidents than humans, but people are afraid because "didn't one hit a pedestrian in Phoenix last year?" Pedestrians get hit by human drivers all the time. Even though AI isn't perfect, perfect shouldn't be the standard. It's not for humans so it shouldn't be for AI either. The key issue here for Legal AI is: who is going to do all the training? Who is going to bring the legal domain expertise to bear and make sure the output is accurate and appropriate and reliable? LAWGEEX Beats Human Lawyer We probably shouldn't buy into the problems we hear about in the press. Look at this blog entry about Lawgeex on Vice TV last week. https://blog.lawgeex.com/hbos-vice-news- features-lawgeex-in-the-future-of-work- and-lawgeex-beats-human-lawyer-again/ It performed much more accurately than the human - 95% vs. 85%. Is that good enough? We hear about is the errors Lawgeex made - which is easy because it's new, so it's easy to criticize. We won't move forward if we don't consider a new construct for Testing and Proving Legal AI. AI-based solutions have the potential for error or bias. Understanding and dealing with that error or bias is one of the challenges of proving / testing Legal AI. Courts, litigators, corporate legal operations and compliance professionals, and regulators want evidence that AI operators know what they are doing. There are also concerns with accuracy, data cleanliness and where data comes from, even more so for Legal AI. Developing a quality assurance framework and a good testing methodolo for Legal AI, and then executing the tests against the tools and grading them, is the next step in proving the promise of Legal AI. Given that the vendors don't want to give away the secret sauce, it becomes about the data and multiple populations extending permutations. A challenge for QA is that some of the problems to which we apply true AI don't have an "answer" that you can check against (e.g. asking AI to read all case law from a judge and predict or recommend the right judge for a case – how can you validate correctness?) Like children learning, it is T H E R E ' S M O R E O N L I N E ! Find more great resources on artificial intelligence in legal on ILTA's website using the search feature. www.iltanet.org

Articles in this issue

Links on this page

Archives of this issue

view archives of Peer to Peer Magazine - Summer 2019: Part 1