What Lawyers Need to Know About Ai in the Law Amid the Latest in Legal Language Mimicry

• Publication year: 2022
• Author: Written by Dr. Lance Eliot

WHAT LAWYERS NEED TO KNOW ABOUT AI IN THE LAW AMID THE LATEST IN LEGAL LANGUAGE MIMICRY

Written by Dr. Lance Eliot

ABSTRACT

This article explores the emergence of Artificial Intelligence (AI) technology known as Large Language Models (LLMs) that are able to generate or transform text into computer-produced narratives appearing to be human-written. Besides examining the rise of LLM technology in contemporary societal uses such as chatbots, the latest in legal research is embarking on trying to use such AI to perform legal tasks. A recent research effort described here has applied an LLM to the California Bar Exam and illustrates the strengths and weaknesses of this latest AI capability. Lawyers, judges, and others in the legal profession will likely encounter LLMs in the latest LegalTech used for law offices and within the courts, thus, legal professionals need to be sufficiently apprised of the benefits and drawbacks associated with this newest form of AI.

INTRODUCTION

Artificial Intelligence (AI) is increasingly being applied to the performance of legal tasks. The ABA has identified six primary areas in that AI is transforming the enactment of legal services, consisting of litigation review, expertise automation, legal research, contract analytics, contract and litigation document generation, and predictive analytics.⁰¹ LegalTech apps have emerged that infuse AI capabilities such as Natural Language Processing for eDiscovery when exploring a large corpus of discovery documents, whilst in the realm of Contract Life Cycle Management there are advanced AI features being used to scrutinize contracts and compose legally compliant contractual passages.⁰²

Among the various types of AI techniques and technologies, one of the newest AI approaches consists of being able to generate natural language narratives. These AI-based programs are used to produce compositions that appear to be human written. The underlying algorithms are trained on everyday written compositions such as found on the Internet and seek to mimic the text by an intricate form of computational pattern matching. Within the parlance of the AI field, these algorithms are known variously as generators, transformers, or more popularly Large Language Models (LLMs).

Recent news headlines managed to bring LLMs to the forefront of public attention when a Google engineer announced that he ardently believed that he had encountered sentient AI while using a chatbot system known as LaMDA (Language Model for Dialogue Applications).⁰³ LaMDA is a state-of-the-art language generator that utilizes a complex artificial neural network architecture.⁰⁴ Though this LLM is able to generate impressive-looking conversational text, it is decidedly not sentient and the Google engineer was mistaken in his unsubstantiated presumption, for which he was subsequently let go by Google.⁰⁵

[Page 22]

These AI algorithms have conventionally been used in generic conversational contexts. Researchers are now opting to explore how LLMs perform in specialized domains such as medicine, finance, insurance, and other particular professions. In this article, an initial exploration of LLMs in the legal profession is showcased. Lawyers, judges, and other legal practitioners are likely to encounter LLMs in both their personal and professional lives. Knowing how the LLMs function and being cognizant of the AI limitations will be vitally important, especially as LLM technology is embedded into LegalTech used by law practices.

LLM AS A FORM OF LANGUAGE MIMICRY

Large Language Models can be somewhat straightforwardly described as being a form of language mimicry. You have likely daily encountered a limited version of textual mimicry when you are using any modern-day word processing app and made use of the autocomplete features.

Having word processing or email apps attempt to predict the next word in a sentence is quite routine these days. The more advanced versions try to predict the remainder of the sentence consisting of a multitude of words. Even more advanced apps will predict an entire paragraph. To glean what LLMs are devised to do, imagine that you keep scaling up this textual mimicry until an AI-based LLM is aiming to predict or generate whole stories based on slimly stated starter sentences.

One of the cornerstone computer programs within the AI field for having foundationally crafted this narrative mimicry capability is a program known as GPT-3, which stands for Generative Pre-Trained Transformer version 3. GPT-3 was developed and continues to be maintained and advanced under the auspices of a company and research lab known as OpenAI.⁰⁶ The third version is the latest in a series of iterative changes and additions that seek to improve the computational pattern matching capabilities. Pre-training refers to the aspect that...