ChatGPT is a cutting-edge internet sensation, launched by OpenAI in November 2022 and had over one million users in its first week[1] and is estimated to be worth 1 billion dollars by 2024.[2]ChatGPT uses Artificial Intelligence (AI) to interact in a conversational way and respond to questions; it works by using a large language model trained by OpenAI. It has been specifically trained to generate human-like text based on the context of the conversation in real-time.[3]
This is not the first time we have seen technology of this nature, with chat widgets (Chat boxes) widely used on websites enabling interactions between a potential customer and the brand, but what makes Chat GPT special?[4] ChatGPT was fine-tuned over an improved version of OpenAI’s GPT-3, known as ‘GPT 3.5’.[5] The fine-tuning process leveraged both supervised learning as well as reinforcement learning in a process called reinforcement learning from human feedback.[6] In essence, the models use human trainers to improve the models’ performance.
Challenges the legal field faces using this technology
ChatGPT is extremely user-friendly and requires no training to be used; however, the easier the technology is to use, the easier it is for errors to arise. It is important for the lawyers and users of the technology to understand the limitations it suffers from.
Artificial Intelligence Hallucination: when ChatGPT is asked a question it does not know the answer to, it will not say it does not know the answer to it but, instead, it has the intendancy to lie and occasionally gets the answer very wrong. As ChatGPT writes incorrect but plausible-sounding incorrect answers, it is harder for the user to identify the incorrect answers, which is a fault with this type of technology.[7]
Prompt Engineering: one of the most interesting things about generative AI is that it gives rise to the need for a new skill called ‘prompt engineering’. The answers provide Large Language Models (LLMs) that are only as good as the data upon which they draw to provide those answers and the way the question is phrased.[8] Users have not been trained on how to delegate and phrase questions to an AI model and, therefore, the kinds of prompts that will ensure a good, reliable answer from the application.[9]
Lack of security: ChatGPT uses reinforcement learning to improve its outputs as it continuously learns from human interactions.[10] Therefore, any information you input into ChatGPT is likely to be adapted and used for training; accordingly, the information used is not confidential.[11]
Algorithmic Bias: ChatGPT takes on biases in its training data, which are revealed when it responds to prompts including descriptions of people. In one instance, ChatGPT generated a rap indicating that women and scientists of colour were inferior to white and male scientists.[12]
Impact on the legal field
The new generation of LLMs are ground-breaking and despite, the issues lawyers might face, we will see monumental changes in the legal practice, as a result of this technology. The founder of legal technology consultancy, Killer Whale Strategies Zach Abromowitz, stated that ‘ChatGPT direct application for legal would initially be helping lawyers tackle their daily workload’.[13] This could be by formulating emails, searching for specific data in contracts and accelerating learning topics.[14]
As this type of technology is disruptive and is continuously chasing increased productivity, lawyers will not be able to use the technology for many tasks; however, the technology will pull out certain elements of inefficiency. Research shows that fine-tuning small models works 100x better than a bigger model, and legal organisations will be able to harness the power of these models and train them on their own data for more honed results.[15]
The most pronounced effect of ChatGPT on the legal industry will come indirectly via OpenAI selling access to their models as a service. Law firms will be able to train, run individual models, and create their own forms based on their own data. We are already starting to see this pairing within the legal industry, with the law firm, Allen and Overy, collaborating with a start-up backed by ChatGPT creator, OpenAI, to introduce a chatbot intended to help its lawyers with a variety of legal tasks.[16] The tool named Harvey has been implemented across its 43 offices to automate and enhance tasks including contract analysis, due diligence and regulatory compliance.[17] David Wakeling, head of Allen and Overy markets innovation group, described the technology as a game changer that can work in multiple languages and across diverse practice areas for the delivery of unprecedented efficiency and intelligence.[18]
Thoughts
There is no denying that this type of technology is disruptive. However, like any other new technology, we need to learn as professionals how to properly evaluate its nature and select tools, which are safe and reliable. This technology is still in development, and ChatGPT is not a human lawyer, nor is always accurate, and as such, lawyers’ ethical obligations will always take precedence over convenience. Whilst the current ChatGPT would be unsuitable to use in legal organisations in its current state; with the correct fine-tuning and adaption, the technology could make lawyers more efficient, thus allowing them to produce high quality work and spend more time on high-value parts of their jobs. Of course, only time will tell what role ChatGPT will play within the legal profession.
-
Endnotes
[1] Jonathan Vanian ‘Why tech insiders are so excited about ChatGPT, a chatbot that answers questions and writes essays’ (CNBC, 13 December 2022) accessed 3 April 2023. [2] Dennis Hilleman ‘ChatGPT-Legal challenges, legal opportunities’ (fieldfisher, 9 December 2022) accessed 20 March 2023. [3] Stephanie Wilkins ‘ChatGPT is Impressive, But Can (and should) it be used in legal?’ (Law.com, 15 December 2022) accessed 20 March 2023. [4] Beata Stefanowicz ’10 Best Chat Widgets for your Website’ (TIDIO, 15 March 2023) accessed 22 March 2023. [5] ‘Models’ (OPENAI) accessed 22 March 2023. [6] OpenAI ‘Introducing ChatGPT’ (OPENAI, 30 November 2022) accessed 22 March 2023. [7] Lak Lakshmanan, 'Why large language models (like ChatGPT) are bullshit artists' (Medium, 15 December 2022) accessed 02 February 2023. [8] Nicola Shaver, ‘ChatGPT: a warning for Lawyers’ (Legaltech Hub, 27 February 2023) accessed 22 March 2023. [9] Ibid. [10]Ibid. [11]Ibid. [12] Perrigo, Billy ‘AI Chatbots are getting better but interview with ChatGPT reveals their limits’ (TIME, 5 December 2022) accessed 20 March 2023. [13] Wilkins (n3). [14] Ibid. [15] Ibid. [16] Victoria Basham, ‘Allen & Overy integrates Chat-GPT-style chatbot to boost legal work’ (The Global Legal Post, 16 February 2023) accessed 20 March 2023. [17] Ibid. [18] Ibid.