Artificial Intelligence, ChatGPT, Language and Legal Professionals

By Cesar Pereira (C.Arb) & André Guskow Cardoso, Partners at Justen, Pereira, Oliveira & Talamini

The popularization of the use of artificial intelligence (AI) tools, such as ChatGPT, raises several issues.

ChatGPT consists of an AI system based on a large language model (LLM). It is an application of the GPT-3 system. Since its use was made available to the public at the end of 2022, the system has achieved great popularity (with more than 100 million users). It is a “generative” AI system capable of generating text in natural language in response to queries made by the user.

This feature is the reason for the great repercussions around the ChatGPT. There is greater awareness of the effects of AI tools when it comes to natural language, as the effects are direct and concrete, and identifiable by large numbers of people.

Artificial intelligence and the law

For the same reason, its repercussions in the field of law is also evident. Law works with language. The preparation of briefs, submissions, and opinions, as well as the rendering of judgments, arbitral awards, and orders, or the drafting of contracts or agreements, all involve the use of natural (legal) language. Therefore, a tool that manipulates language and creates logically linked texts in response to user queries potentially has a very large impact on the legal profession.

The issue is the subject of growing debate. One must consider the consequences of using artificial intelligence tools that deal with language and enable the generation of texts and legal documents. The great information-processing capacity and the possibility of combining such information in natural language make generative AI tools (such as ChatGPT) highly relevant for law professionals.

A revolution in the way of working within the legal professions can be foreseen. At the current stage of technology evolution, the possibility of effective competition between humans and AI is not considered. However, such systems should become indispensable tools for law practitioners. In the same way that word processors are the norm and typewriters belong in museums, generative AI systems will become quickly essential instruments in the legal field. Mastery of such tools – the ability to prompt AI to perform the tasks needed for greater efficiency and quality in the legal work at hand – will be soon a basic requirement for law professionals.

In addition, the capacity to process texts in natural language and comply with specific instructions also allows the development of so-called ‘computational law’, in which laws and regulations are automatically processed and interpreted by computational systems and combined with information and documents that are provided to the system.

This new reality is already perceived in several domains and will certainly also impact the arbitration process, in which structured disputes usually involve a large volume of information and documents.

On 8 March 2023, International Women’s Day, the current stage of development of ChatGPT was put to a practical test.

Back in 1993, a group of scholars and practitioners created in Vienna, the Willem C. Vis International Arbitration Moot competitions. The initiative spread to Hong Kong in 2003. Today, hundreds of universities participate every year in moot hearings with student advocates arguing the sides of claimant or respondent in a fictitious case before panels of professional arbitrators.

Two Brazilian organizations – the Chartered Institute of Arbitrators Brazil Branch and Arbitration Channel – had the idea of a moot hearing between a highly qualified university team and another team of non-law students instructed only by ChatGPT. The experiment was carried out on YouTube and its results are fascinating.

The excellent university team was the winner of both the in-person and virtual editions of the 2023 Rio Pre-Moot (Federal University of Rio de Janeiro – UFRJ).

ChatGPT was represented by a student actor and a young political scientist, neither with any training in law. They faithfully read everything ChatGPT had prepared for them beforehand or drafted for them on the spot, in response to the Arbitral Tribunal’s questions or the opposing counsel’s arguments.

The internationally renowned Tribunal members were Amb.(r.) David Huebner C.Arb, Sophie Nappert and Louise Barrington C.Arb.

The ChatGPT preparation required a couple of weeks of part-time work from its coaches (or prompt engineers), a young lawyer who is a former participant of the Vienna Vis Moot competitions and a data scientist who teaches machine learning. Given the technical limitations of the current version of ChatGPT, they designed a format to break the made-up case into subsections but transition smoothly from one to the other. This structure enabled ChatGPT to produce a coherent initial pleading and to be ready to answer questions and opposing arguments within the scope of each subsection.

The element of uncertainty in the experiment was precisely how ChatGPT would react to questions to be answered quickly and in one shot. AI is known to work under a trial-and-error scheme, in which the appropriate answer is not always (or hardly ever) the one given to the first question. Even so, ChatGPT was able to provide immediate, reasoned, structured and succinct answers to all the questions it was asked. It even surprised the Tribunal when, in a moment of candour, it conceded its argument was untenable. However, it quickly recovered, and, in a later rebuttal, it argued that such concession was superseded by an alternative argument in its client’s favour.

Technical discussions that followed the experiment highlighted the current limitations of the technology and certain features that must be understood and managed, such as its tendency for hallucination – not always bad, since its bright side is creativity. One can consider, for example, using hallucination texts generated by AI during brainstorming sessions.

The broad use of generative AI tools also brings some questions concerning fairness and equality into the arbitration process. Concerns related to the equal treatment of the parties, transparency in the use of AI tools, and the possible need to disclose their use (as well as the consequences of non-disclosure) were some of the points raised during the discussions. The preservation of due process even in the presence of such AI tools was also a concern. On the other hand, considering that currently AI tools are provided by third parties and not under the control of the parties to the dispute nor the arbitral tribunal, there are also challenges concerning the protection of privacy and confidentiality of the arbitral process.

The success of the experiment, reflected in the enthusiasm of both the audience and the participants, shows that this is just the beginning. If an experiment such as this, based on a general-purpose AI tool, already shows so much promise, one can only imagine the heights to be reached by a dedicated AI trained with datasets focused on legal matters or even specific areas of practice. At least at the current stage of technological development though, it was apparent from the experiment that human traits such as flexibility, sensitivity and creativity remain fundamental to the legal practitioner.

Want to learn more about ChatGPT and its capabilities? Click here to read all about how ChatGPT can benefit the legal profession.