AI Isn't New to Law: How the Practice of Law Should Embrace AI

Remember when AI was the stuff of science fiction? With the release of OpenAI’s GPT-3 and GPT-4 and other AI generative systems into the consumer market in late 2022, artificial intelligence became an accessible tool that can be applied in almost every computer environment. UBS, the global financial services company, reported that GPT-3 and GPT-4 reached 100 million active monthly users only two months after its release, making it the fastest growing internet app in history. 

AI systems began to be integrated into commonly used information storage systems years ago. But when corporate service providers applied AI technology to solve industrial problems, its presence seemed purely benign. Now, as AI tools become available to the general public, the prospect of people “being replaced” by an AI application seems more threatening.

We know that AI will not replace lawyers, but how much legal work can it be trusted with?

Proponents of AI’s virtues foresee the elimination of tedious, routine administrative work, billing, and scheduling. Sceptics doubt AI’s potential application in the practice of law since its intelligence is, well, artificial. 

But the doubters might be surprised to learn that AI has played an important role in law for years. We  use AI every time we access legal research programs, and most firms probably use AI assisted document management systems. AI technology has become essential for the operation of e-discovery managed document reviews and data searches over the past two decades.

Understanding what AI is—and what it is not—helps to identify where it can be of value and what limitations it currently has. Not only will AI certainly impact your practice in the future, it already has.

Artificial intelligence (AI) is the technology by which computers perform tasks usually accomplished by people using human intelligence. Functions within AI’s current capacity include recognizing speech, solving problems based on data analysis, translating languages, identifying patterns, and playing games.

AI processes massive amounts of data according to programmed algorithms designed to accomplish specific tasks. An algorithm is a precise set of instructions written in plain English that direct the computer to execute particular actions. As the sophistication of the AI systems advanced, the computers algorithms that could themselves write more algorithms were uploaded, relieving human programmers of the burden of designing ever more complex strings of flowcharts and directions.

As AI developed, it increasingly appeared to imitate human thinking. This illusion troubled those who feared the machine was simulating a living being. Of course, AI is not actually thinking. What AI is doing is processing data patterns and organizing them according to the preexisting relationships inherent in the data when it was input. 

The enormous amount of data that is uploaded to AI systems enables it to learn new information and then adapt to quickly retrieve relevant information based on the recognizable patterns in its memory. 

Microsoft’s ChatGPT, Google’s Bard, and OpenAI’s  GPT-3 and GPT-4 are each just one form of AI currently getting intense media attention. The GPT is an acronym meaning “generative pretrained transformer.” ChatGPT and similar software are computer systems that are pretrained by consuming massive amounts of information from internet databases it used to learn the patterns in which words appear in a selected language.  Included in its education were books, Wikipedia articles, webtexts, and other written internet articles. This allows it to predict the next word in a sentence based on the frequency with which that sequence of words appeared in the data it trained on. 

Remarkably, ChatGPT is said to have studied more than 400 billion words. 

The problem with using ChatGPT or any other interactive text generating AI program for meaningful work is that it learned everything it knows from the internet. The system is asked to provide information in response to requests from users. Its answers are often inaccurate and necessarily repeat biases embedded in the internet data.

Simply put, AI and ChatGPT systems are what they eat. They are limited to draw from only that which they are trained to reference.

When it is employed in a legal practice to draft a contract, it will produce a grammatically perfect draft that will repeat the most frequently appearing expressions on the internet for each desired contract term. What it won’t do is discriminate between reliable and unreliable data, incorporate changes required by recent court rulings, comply with recent statutory mandates, or conform to context-specific or unique jurisdictional rules. 

At least it won’t do so yet.

The development of specialized information databases from which AI GPT systems can learn should empower the software to improve its reliability and utility enormously.

Like all tools, AI and Chat-GPT systems can increase efficiency and reduce the time and attention humans spend completing tedious tasks. Specialized AI software can perform complex legal research surveying thousands of judicial decisions in a fraction of the time required for a team of lawyers or paralegals to complete the job.

AI’s document review and error detection capabilities can improve the accuracy and thoroughness of any written document, detecting and correcting oversights, and filling gaps in necessary text. Costly errors and administrative or clerical mistakes can be flagged and corrected before they cause damage to the firm or the client. Humans err when distracted or fatigued; AI technology is relentlessly alert and energetic. 

Every case being litigated in the world today is necessarily an e-discovery case. AI technology can already search and screen tens of thousands of documents and bits of electronically stored information (ESI) to select relevant pieces of discoverable data, sequestering privileged or confidential items from the output. 

Civil litigation firms and insurers are using AI systems to predict the likelihood of prevailing in particular lawsuits by processing the outcomes of previous cases that share the same fact patterns or legal elements. The predictions AI software provides can influence the stakeholder’ decision whether to proceed with or abandon the litigation.

One article published in December 2022 asked, “Will AI Make Lawyers Obsolete? (Hint: Be Afraid)”. Before we get too panicky, let’s remember that AI, GPT-3 AND GPT-4, ChatGpt, and Bard are computer software programs that reflect what they ingest. They do not now and never will possess the features essential to perform the sensitive, nuanced, interpretive functions lawyers fulfill every day of their professional lives.

AI can be a valuable tool if used to free busy lawyers from ministerial tasks, improve the quality of their drafting, and increase their opportunities to better serve their clients, their colleagues, and the profession. The bar’s rules of ethics will be amended to incorporate the appropriate use of artificial intelligence, just as they were when teleconferences, email, and e-discovery changed the way law is practiced.

The only risk posed by AI and GPT systems to the legal system is overestimating technology’s ability to produce original ideas or become a substitute for a human’s voice and pride of authorship.