Pop culture has heavily influenced our general idea of what artificial intelligence (AI) is, from engineered villains like Agent Smith to wholesome robots like Wall-E. Hollywood loves to exaggerate for dramatic effect, but there is still some truth tethering these futuristic characters to our present reality. Our forays into AI are still elementary at best, but many industries are using the little that we know to benefit by adopting AI to eliminate redundant and manual tasks. For the legal industry, there is promising potential to significantly boost efficiency by automating routine, high-volume tasks, such as legal research, diligence, document drafting and electronic discovery.
However, one study shows that more than 90% of respondents said “no” or “do not know” when asked if their law firms use AI. Surprising at first, but not so much if you ask why. Another survey did ask why, and the responses show that more than 80% of law firms fear that either new technology will permanently replace human resources, or their clients’ new technology tools will reduce the need for lawyers and paralegals.
The legal industry is generally known as one of the slower-moving industries in adopting innovative technologies, and most law firms by nature prefer to follow, not lead. But, as time has shown us again and again, the more we dig in our heels, the deeper we sink. So, let’s see what kind of legal jobs will be replaced by AI and how corporate legal departments and law firms can prepare for the disruption.
A Glimpse At AI History
The official field of AI study established in 1956 at Dartmouth College, just six years after Alan Turing, created the Turing Test. But it took 60 years for a chatbot named Eugene Goostman to successfully trick human judges into believing that he, too, was an actual human by passing the Turing Test in 2014.
Progress in this field clearly does not happen overnight. Ask any developer who’s still working on solving what we might consider “simple” problems using AI. Plus, implementing AI in an organization is turning out to be much harder than we thought.
Top-Down Problem Solving
Not long ago, it was thought that only routine tasks could be automated; tasks that can be specified by reference to a set of rules. In this top-down method, the starting point is the business’ problem/question. Let’s imagine that there is an initial belief in a corporate legal department that it has been overcharged by some vendors or law firms on some unapproved expenses or fees.
In this approach, it’ll need to select those suspected vendors and carefully review their invoice line-item against the contract and/or billing guidelines to check whether the invoicing is correct or if there are unapproved or excessive charges.
Certainly, knowing all the rules, you can write programs to validate them in the data. For decades, computer programmers wrote millions of lines of code to verify business rules and automate workflows for organizations; however, as the number of rules grew, programs became more and more complex and costly to maintain.
Bottom-Up Problem Solving
The other idea is to start with the data and go up to build valuable insights and rules. In this method, machine-learning algorithms are used on massive amounts of data and computer power to learn and build up to guiding business rules. This eliminated any considerations of how complex problems were or how contextual solutions needed to be in order to automate them.
In this approach, you could start with all your invoice data and use algorithms to see if there are any interesting trends or anomalies in the line items. After sifting through the different insights, the system can notice spikes like travel charges for a case type that normally doesn’t require travel. Invoice reviewers would find it valuable for auditing purposes.
People who have wrestled with AI algorithms know that theory is far from real life. Implementing AI in organizations with real data is extremely difficult. Here are three of the challenges:
• Labeling data is difficult and costly. In situations where data is limited or expensive to collect, existing rule-based and workflow engine applications are more effective than an AI solution.
• It needs massive amounts of data to train. The more data, the more accurate the machine’s responses. In situations in which data is limited or expensive to collect, existing rule-based engine applications are more effective than an AI solution.
• There’s an issue with explainability. AI products are inherently a “black box,” and one of the important factors in most decision-making processes in business is to have clarity of the steps involved in solving a problem.
Solving problems that require the full spectrum of intelligence remain beyond the reach of current AI systems. The human brain’s ability to use both the top-down and the bottom-up approaches in tandem — connecting dots and performing tasks such as presenting a cohesive and persuasive case to a diverse jury — is a unique intelligence that machines might not be able to fully emulate in the near future. Traditionally, legal services contain a mix of some tasks that can be automated and some that cannot, such as those that require unique experiential knowledge.
It is exciting to see how AI will continue to progress over the next decade. This past year taught us how easy it is to fall behind, and legal departments and law firms should remember this and can embrace legal technology to stay ahead. Smart lawyers who adopt technology and transition into hybrid legal tech culture sooner can reap great rewards. Regardless of circumstance, AI’s scalability can ensure that they will be able to better adapt to the future, continuing to provide independent professional judgment, focusing on meaningful, complex and mission-critical work for their clients.
Sooner or later, attorneys will say goodbye to long hours and menial tasks, and instead, dedicate that time to their practice and their clients, doing strategic planning and formulating complex arguments.