top of page
Search

The Possible Implications of GPT-3 to the Business of Law

By Thabo Magubane.


I. BACKGROUND

Across the globe, Artificial Intelligence (AI) has been the center of debates in the business of law, mainly to ascertain the fundamental changes it will have to how lawyers perform duties. For several years, legal departments have been looking for ways to do more in less time and still deliver the expected value to clients. It happened that the application of AI to various areas such as self-driving vehicles, automated language translation, speech generation and recognition, and predictions resulted to legal professionals looking for ways in which they can also adopt AI. In short, AI is a broad discipline under the umbrella of Computer Sciences.


Within AI, there are different fields including Machine Learning and Natural Language Processing Systems (NLP) where the field of law is primarily applicable. Recently, eyes where filled with excitement and wonder amongst computer enthusiast over the newly announced NLP software known as Generative Pre-trained Transformer 3 (GPT-3). In simple language, this article will evaluate features of GPT-3 and its applicability to the business of law. Further, too much of a good thing is bad, we will look at some of the risks that can be expected from the use of GPT-3. Automation brought by legal technology tools can have a catalytic effect to accessibility of legal services in the lives of less affording people, as this has been a century debate in the legal profession across jurisdictions.


A. What is GPT-3?

In March 2020, OpenAI researchers and engineers presented a paper introducing a groundbreaking autoregressive language model that uses machine learning methods to produce human-like text. OpenAI is a San Francisco based AI research laboratory behind the development of GPT-3, and the journey began with GPT-1 which focused on improving computer generated human language in a form of texts, and later came GPT-2 which furthered the interest of GPT-1 by eliminating human intervention with unsupervised machine learning methods that searches for undetected patterns in data to self-generate automatic responses without human supervision. GPT-3 is the latest language model developed by OpenAI, it has features that indicated significant improvement in text generation and predictions. Previous NLP models were trained to take on tasks such as answering simple questions which are difficult to fully apply in the technical field of law. However, with GPT-3 things are different as it said to be 100 times coherent and faster than GPT-2. The model is tremendously trained to a level where it could generate field-specific responses that are indistinguishable from those written by humans. Let us be technical for a moment and explore how GPT-3 is able to do this, then we will look at how this will affect the future practice and business of law.

B. Distinguishing Characterises and Functions of GPT-3 from Other NLP Models

  • GPT-3 is an autoregressive language model that generates human-like text from analyzing large amounts of data from the internet and observing which letters and words have a habit of following one another. GPT-3 can also create new correlations and come up with new discoveries and responses from the data used. From this first element, we can see that GPT-3 gives computer applications the ability to answer questions, which is not exciting, as more computer applications can do that in our days. Let us look at further distinguishing functions.

  • Contrary to other NLP models which must be trained with large amounts of data, and then tuned to perform a specific task. GPT-3 can perform various functions from producing fiction, poetry, press releases, music, and jokes to writing its functioning HTML codes. Since GPT-3 is not task-specific like other NLP models, it produces articles that are difficult to distinguish from those written by humans. Based on these performances, I think if GPT-3 becomes a full cross-domain AI technology this can unlock new possibilities for many fields, including the practice of law.

  • Before GPT-3, the largest language model was GPT-2 with 1,5 billion parameters, now overthrown by GPT-3 with 157 billion parameters. A parameter is a calculation that allows the model to understand the language in any meaningful way. GPT-3 can complete any given sentence, making it the leading language model in generality, as it is not limited to any specific field. This feature allows GPT-3 to triumph in self-supervised responsibilities and limits the need for human-intervention in tasks completion.

II. GPT-3 AND THE BUSINESS OF LAW

The business of law is about language, law schools spend thousands of hours teaching language to aspirant legal professionals. Language is one of the barriers between other NLP models and law as they are either trained on a single domain to complete specific tasks or on too general language material that makes it difficult for a model to provide various technical responses. However, the legal industry is a unique domain that requires a certain level of flexibility and adaptability on the use of language. This flexibility is one reason we still do not have a fully computerized AI systems that can render legal advice and write legal opinions on different legal issues. It is vital to mention that we are still far away from achieving a fully human-like AI technology, as what we currently have is AI technologies that are good in single domains or limited to a specific task. The use of AI in the business of law is still limited to minor tasks that can be easily automated, like document automation and keyword selection processes, and decisions analytics. There is still a need for a system that can generate legal texts and different legal documents that are not only limited to a single field of law, for example, there is quite a number of firms that make use of AI to generate contract clauses, contract management, contract analysis, document classification, and decisions analytics. However, the model used by these firms is trained on a single domain, whereby they can only execute single tasks that is based on routine and repetition. Case-backlogs, delays, lack of legal practitioners, and unaffordable legal services doubled the need the for access to justice in developing regions like Africa and Latin America which increased the demand for a system that can-do things that normally require human intelligence. For example, a system that can parallelly supplement legal practitioners with the production of essential legal documents and text generation.


A system that will increase productivity and profitability in legal departments and retain value in the delivery of legal services. Can GPT-3 contribute to the achievement of such a system? Let us look at some aspects where GPT-3 can contribute to the business of law, and perhaps play a role to the aspirations of making legal services affordable while promoting business growth.

A. Examples where GPT-3 can be Applied in the Business of Law

  • Document automation and text generation. As I mentioned, a number of firms are already flourishing in the space of document automation. However, they all seem to face a similar barrier which is their technology can only be applied in a single domain. Let's take for example, text generation and analysis technologies that can only be applied in contract automation because many current AI models can only be trained using a single field-specific data, as a result, the technology can only be able to perform contract related tasks and nothing else. Across AI disciplines, this remains a huge barrier to the aspirations of reaching generality whereby we have a single technology that can master different fields. Instead of having to train different models for different tasks such as contract management, analysis, text-generation, and litigation predictions. GPT-3 shows promises to achieve singularity. Rather than having to adopt and train different models for each task, GTP-3 can result in a single model that is used to increase productivity in different areas of law. Almost all legal departments have weekly or monthly newsletters, and there is a possibility that GPT-3 can achieve a fully automated newsletters to update the general public on specific legal or non-legal topics.

  • Legal research is another area where GPT-3 has the potential of making an impact. The ability to answer questions and provide legal opinions has always been central in the business of law. GPT-3's ability to provide responses on vast topics indicates its potential of being used in the arena of legal research, writing and reasoning. I think it is too early to say GPT-3 will entirely takeover this space, but rather what I can predict is that it will have some impact, either by assisting lawyers to save time on research related tasks or by generating researched documents on its own. Companies like Casetext and ROSS admitted that GPT-3 will not replace what they are doing but it creates a new building block which they are more than willing to adopt, as GPT-3 is compared to electricity where at first it was exciting but eventually everyone started using it.

  • Language translation. GPT-3 can be used to convert legalese language to plain English and vice versa. It is common that large firms comprise of different departments and amongst all departments there is a language barrier. In most cases, it is time consuming for non-legal departments within a law firm to translate legal documents into simple understandable language that can be shared across the firm.

However, it is still early to assume that all the translation duties will be outsourced to GPT-3, instead there is room for GPT-3 to have a role to play within the context of text translation. Language translation can go a long way on access to justice matters (A2J) as GPT-3 can be used to translate legal text for people who cannot afford to consult a lawyer. Such services can be scaled through online services, for example, the use of chatbots and automated responses to answer general legal queries.

III. ALARMING CONCERNS OVER THE USE OF GPT-3

Too much of the good thing is bad. Like all artificial intelligence methods, there are possible ownsides to the use of GPT-3 in the business of law and some of these concerns are familiar in the legal technology space. These downsides are not final as GPT-3 is still a new software that has not been fully explored. More will unveil themselves as the time goes and more people get access to the software.

  • Privacy and confidentiality concerns. GPT-3 is a model that predicts the next word from previously used words. Even though GPT-3 is quite unique when it comes to creativity and text variety, it still shares some commonality with other language models, which is the use of data to predict outcomes. This is a common barrier when it comes to the use of data in the business of law as some clients will not find comfort with the use of their scenarios and legal documents to predict future outcomes of other clients. Which is to say, in the business of law quality data that will assist GPT-3 is still difficult to access, also some legal departments are still skeptical with the use of data to predict outcomes.

  • Unfairness and biasness. What gives GPT-3 its unique feature is that it is trained from a wide range of information, from easily accessible internet sources to trustable sources, social media posts, blog articles, news articles and more. The sad reality is that internet sources can be full of biasness as many are completely from individuals expressing their views and preferences, either over a product or a group of people. Employing a biased tool can turnout disastrous for legal businesses as no firm would allow their ethical reputation to be jeopardized. Even with GPT-3, careful selection and classification of data is still vital for an unbiased and fair model.

  • How sure are we that GPT-3 will not begin the spread of fake news and information across the web? As mentioned above, GPT-3 can write blog articles, poems, songs, essays without human intervention. One exciting scenario we looked at is the automation of newsletters. As much as this sounds exciting, it is also dangerous. Throughout the world, societies are looking for ways to fight fake news and the damage it causes to democracies. Since OpenAI used a variety of internet sources it launched a toxicity content filter API that rates all content created by GPT-3 on a toxicity scale from one to five, and anything above two is a red flag.

Again, the effectiveness of this strategy has not been fully explored as GPT-3 was just introduced in March 2020. OpenAI seems skeptical to make the software available to everyone as access is restricted by an application made to OpenAI, and with approval one can then make use of the software. Seems like ethical clearance is mandatory because GPT-3 is quite a powerful tool and it can cause significant damage if it ends up on the wrong hands.

IV. IN CLOSING

Higher-order cognition required by practice of law still falls outside current AI systems, however, some computational approaches to automation may produce results that are good enough in certain legal contexts. The end-goal of applying technology to the business law is not replace lawyers but to assist them with increased productivity and profitability while delivering value to their clients. Technology is to automate burdensome tasks so that a lawyer's cognitive efforts and time can be conserved for tasks that are more likely to require higher-order legal reasoning. GPT-3 is a piece of technology that shows great potential for fields turning to automation for repetitive tasks, including the business of law. GPT-3 is still a new software and experiments are still ongoing to make assessments on how and to what extent the tool can be used to improve business productivity and profitability. There is hope that the software will add value to the business of law.


Selected resources for further reading

 

About the Author Thabo Magubane is certified legal technologist, a final year LL. B candidate, and Alibaba Cloud Specialist Trainee. He is one of the recognised voices in the field of legal technology and innovation in the Africa region. The African Union endorsed his latest work for development policy on court modernization using Artificial Intelligent and 5G. The views reflected are of his own and does not represent those of any institution or organisation he is part of.


#ThaboMagubane #legaltech #AI #legalinnovation

bottom of page