OpenAI is an Artificial Intelligence research start-up which was initially funded and supported by Tesla’s Elon Musk and Y-Combinator’s Sam Altman with a billion-dollar combined investment in 2015 when it had started. The main goal of OpenAI which comprises both for profit and not from profit divisions is to build a friendly AI that benefits humanity. From excelling in video games and using AI to compose music OpenAI gained traction when Microsoft invested $1Billion to strengthen its cloud capabilities using OpenAI’s research. Ever since OpenAI has thrived. Their best work is considered to be the GPT, a family of AI driven Language models aimed to simplify NLP problems. Recently launched GPT-3 is the third version which is considered to be the best language model trained on 175 Billion parameters which is approximately 10 times powerful than Microsoft’s Turing-NLG and 100 times than that of Google’s BERT.

At the early stages of development of GPT-2 predecessor of GPT-3 the point that the research papers revolving around it were trying to make was that people in NLP spend tremendous amounts of time working on clever logics that can be done for detailed fine tuning to achieve different benchmarks for different problems. GPT-2 did reasonably well with tasks for which it had not been trained. It was purely trained as a language model which was only intended to predict the next word or token given what it has seen or been trained on so far. This was because it was not trained on a good quality text but chunk of any text that it could get, GPT2 performed really well on some benchmarks. This was an indication that more could be done and the limit was not yet reached, this assumption by OpenAI built the very first theoretical approach for GPT 3.

With some clever engineering, which requires clusters of machines to run, which is 117 times bigger than GPT2, GPT3 is supposed to be OpenAI’s first commercial product and is expected to be launched later this year as a paid service through cloud. GPT3 after its beta  launch has flooded social media and the developer community with positive sentiments. People are exploring the APIs that OpenAI is providing , few applications include using plain English to write computer programs, design using plain English etc. But larger applications are being explored in detail across various industries. One such industry with adverse potential to adopt GPT-3 is the Legal Industry which includes and handles mammoth amount of unstructured text data and most in terms of handling and creating documents.

The problems in the legal industry are vast and complex. This is the very reason why the adoption of technologies like AI and automation is very slow. A typical automation on contract reviewing takes 1000s of documents to train with no surety of better accuracy. In countries like India with poor document qualities of court documents and unstructured formats in contracts it could have been impossible to automate and build ML models around it. Also vernacular language applications and translations are next to impossible due to lack of training dataset.

The whole idea and intention of building GPT-3 is the very purpose of solving complex NLP problems on unstructured text. GPT 3 makes it possible for creating complex ML problem with small training datasets. This can easily solve problems relating to automatic contract drafting, building better and intelligent Knowledge Management Systems and create better computational Legal models.

GPT-3 and Contracts

With capabilities of understanding complex languages, GPT-3 has a capability of turning a complex legal English used in contracts to an understandable and normal English, making it easier for a layman to understand. This application can help make contracts easier to read and bring transparency among the parties. One other application that extends can be term sheet drafting. A reverse process of converting plain English texts to legalese can also be tested for easier term sheet drafting. GPT 3 can also help in real time translation of contracts in different regional or international languages which again is a tedious task for legal teams in the contemporary world.

Analysing clauses and risk associated with a contract is one of the most important tasks to do. On a mere training of some 100 examples GPT-3 could quickly identify patterns in clauses, classify similar contracts and also identify risks associated according to a firm’s custom playbooks making it really useful at the time of due diligence.

GPT-3 In Information retrieval and Knowledge Management Systems

Over a period of time with collection of huge amounts of data and information managing knowledge will be a difficult task. Organizing information that will comprise a variety of information across documents, videos, images, tables and finding the right information will be a challenge. The legal industry today deals with numerous types of information from court documents to contracts and filings. The problem with the legal data is its unstructured handling and creation which makes it difficult to organize, identify and even impossible to retrieve the correct information at one search query.

Over the past few years with the evolution of Language models like Watson, BERT etc a lot of clever approaches have been used to solve these problems, which did not turn out to have a very good success rate. Here comes the significance of GPT-3, it can understand and identify trend, structure and insights from large chunk of unstructured data in a contextual manner i.e. a human intuition or human like understanding on developing a common sense on English language and its structure, This understanding eventually makes it easier for GPT3 to extract answers on a random questions that someone might ask.

Say for example GPT-3 can answer questions like,

What are the roles of an auditor?

When can a death penalty be relaxed?

Now if GPT-3 is implemented on the primary legal research engines finding case laws and judgments that answer questions would be so easy and convenient.

Now imagine a similar capability that can be installed on the internal Knowledge Management Systems of a company. It will be so convenient in quickly identifying the correct information by merely asking questions, it is like an advanced chatbot that can answer your queries.

These are just a few applications that can pave the way for legal innovation, but other applications can include simplifying e-discovery, simplifying drafting and document creation.

Apart from its applications for law firms and corporates, some major problems of judiciaries and courts such as docket management, identifying similar cases and extracting relevant information from evidence can be easily solved by building applications and solutions using GPT-3.

But every good thing comes with a price and it will be important to see the pricing of such a powerful model knowing the fact that it can be a game changer. Well whatever it will be, the use of technology has now started to unfold in the legal industry and it will keep evolving with years to come.