We trust our computers to handle our to-do lists and calendars because they never forget, right? While computers are good at remembering what we tell them (and a big thank you to Google for remembering my kids’ birthdays), one of the current weaknesses of artificial intelligence (AI) is that it cannot apply what it learns in a different scenario. For example, an AI that learns to play chess does not have a leg up when learning to play checkers. Essentially, computers have a “catastrophic forgetting” problem that forces them to relearn what they already knew just because they are presented with a new project.
Researchers are now making breakthroughs to overcome this ‘forgetfulness’ problem. Working in connection with neuroscientists, researchers are attempting to have AI learn more like humans so they can apply what they have learned in one context to another related context without starting over. In other words, teach computers to learn more like humans do so they stop forgetting what they already learned.
So if we succeed in teaching computers not to forget, how does that help us in court? One of the hottest eDiscovery topics in recent years has been the use of technology-assisted review (TAR) to significantly reduce the number of attorney hours spent reviewing documents. TAR basically has a computer learn through trial and error. A human reviewer corrects mistakes made by the computer until the AI is ready to review the remaining documents on its own. The same technique has been used for decades to teach a computer to play 20 questions (and if you have not tried 20q.net’s online version – I highly recommend it).
TAR is a great tool for very large document sets because even after the time and expense of teaching the computer (who needs some trial and error to get it right) there are many documents left for the computer to handle on its own. In such cases TAR ends up saving lots of money because those remaining documents did not need to be reviewed by an attorney. But when you start with a smaller document set, by the time the computer learns to handle the review there may not be enough documents left to make the investment worthwhile.
Now imagine that the computer taught to review documents did not forget what it was doing when the next document review came along. For example, what if after handling one employment discrimination case, the computer knew what to look for the next time around. There would certainly still be case-by-case adjustments, but it could significantly decrease the number of trial and error runs it takes to get the computer ready. That could make TAR cost-effective even in smaller document sets.
We are not going back to the days of always having small document sets, not even in small dollar value disputes. Our society continues to create more and more data, so we need to find cost-effective ways to handle this fire hose of data. My prediction? Within a decade lawyers will regularly use TAR to help them handle the analysis and review of data in cases worth less than $25,000. Seem too fast? Do not underestimate the speed technology changes our lives. After all, the iPhone has existed for less than a decade (launched June 29, 2007).