You have probably heard the buzz, you might have seen it trending, and some of you may even have beaten the queues and had the opportunity to try out the latest artificial intelligence tool, ChatGPT. For those of you who have not been so lucky, or if the ChatGPT hype has completely passed you by, this article will consider and explain what it is, how it works and how it may affect the legal sector and your interactions with it.
What is ChatGPT?
ChatGPT is a powerful chatbot launched by San Francisco-based artificial intelligence research laboratory, OpenAI. According to OpenAI itself, ChatGPT is a tool that optimises language models in order to interact with its users in a conversational way. Put more simply, you ask it questions and it answers them in a manner that would make distinguishing its responses from a human’s quite difficult indeed.
The model was trained using what is known as Reinforcement Learning from Human Feedback (RLHF), a process spread over three phases. The initial phase consisted of inputting conversations during which humans took part in both sides of the interactions, with model answers being used as prompts to compose responses.
The second phase saw AI trainers rank responses given by the chatbot during conversations in a process described as a reward model, which was then repeated as part of a fine-tuning exercise.
In the final phase, the resulting dialogue dataset was paired with that of another of OpenAI’s models, InstructGPT, a tool that aligns language models to follow instructions. Eventually, in November 2022, ChatGPT was launched to the public in what is described by OpenAI as a ‘research preview’ but what is, presumably, its final learning phase through which it has access to more users and more questions than OpenAI’s server capacity can currently handle.
Applications in the law
ChatGPT has only been live for three months and there are already myriad examples of its real world application, including in the legal sector. It was recently reported that a judge, Juan Manuel Padilla Garcia, in Cartagena, Columbia used ChatGPT to assist him with solving the key issues at play in one of his cases. In this instance he asked it legal questions and used the answers it gave in full; however, it is important to note that each of the responses given was fully fact checked and were only included along with Judge Garcia’s own opinion, providing the necessary context in what was a sensitive case.
Judge Garcia clarified that use of the chatbot was to “assist and speed up the process – not replace judicial expertise”. However, there is concern about the impartiality of the responses chatbots such as ChatGPT provide, due to the fact they can only be based on the datasets they have access to, thus running the risk of perpetuating any biases or possible agendas that its trainers or users hold.
For example (and famously), a chatbot unveiled by Microsoft, ‘TayTweets’, went from professing that “humans are super cool” to tweeting racist and misogynistic messages, all based on Twitter conversations and interactions with other users and all within 24 hours.
So, will ChatGPT replace lawyers? Well, let’s hear it straight from the chatbot’s mouth.
“Can you replace my lawyer?”
ChatGPT: “As a language model AI developed by OpenAI, I am not capable pf replacing a lawyer as I am not a licensed legal professional and I cannot provide legal advice or representation. It’s best to consult with a licensed attorney in your area for guidance in legal matters.”
“Why can you not provide legal advice?”
ChatGPT: “As an AI language model developed by OpenAI, I am not authorized [sic] or qualified to provide legal advice. Providing legal advice requires specialized [sic] knowledge, training, and experience in the interpretation and application of laws, regulations, and procedures. This type of expertise can only be obtained through formal education and practical experience working in the legal field. Additionally, providing legal advice often requires an understanding of the specific laws, regulations, and procedures that apply to a given jurisdiction. Since I am not a licensed attorney and I do not have direct access to up-to-date legal information, I cannot provide reliable or accurate legal advice.”
ChatGPT is impressive – in fact, it has passed the Watson Glaser Test; a critical reasoning exam set for prospective trainees at several Magic Circle firms.
On its website, however, OpenAI freely admits ChatGPT’s shortcomings, including that, it can provide “plausible-sounding but incorrect or nonsensical answers” and that it is sensitive to tweaks to input phrasing, providing the correct answer in one instance and an incorrect answer after a slight rephrase. The tool is also “excessively verbose” and does not ask clarifying questions, instead “guessing what the user intended”.
Finally, and perhaps most crucially, ChatGPT’s dataset ends in 2021. The law is constantly changing; both from a statutory and common law perspective and a tool that is not aware of these changes will eventually begin providing information that is out of date. Relying upon such information exclusively could have catastrophic and costly real-world consequences. In researching this article, ChatGPT was asked why its data set stopped in 2021 and whether it would ever be able to respond with up-to-the-moment information in real time. Its response was that it lacked the processing power to work at the necessary speed at its current stage of development, although work to improve its performance is always ongoing.
Speaking of which, Google’s answer to ChatGPT, Bard, is being launched in the coming weeks, and Microsoft has plans to integrate ChatGPT with its search engine, Bing. This means that the race is on for AI supremacy and with that race will come the exponential growth and efficiency of these systems.
The future may well be an AI one, but for the time being, Thomson Snell & Passmore remains here to assist you and your businesses with clear, practical legal advice. Please do get in touch if you need our help firstname.lastname@example.org.