Banner artwork by thinkhubstudio / Shutterstock.com
Cheat Sheet
- Language processing and deep learning. Interdisciplinary advancements in machine learning models have achieved complex AI engines capable of generating human-like text.
- A stalwart administrative assistant. With its haste and ability to process large amounts of data, ChatGPT can be an excellent tool for drafting, proofreading, and research.
- Garbage in, garbage out. ChatGPT’s output is only as meaningful as its input, and users must take care in verifying the information they feed to, and extract from the model.
- Ethical implications. This nascent world of AI modeling raises questions in realms such as data privacy, intellectual property, and carbon emissions.
The explosive emergence of ChatGPT as a consumer tool has catapulted Artificial Intelligence (AI) and its subfield, natural language processing (NLP), to the technology stage. As AI and NLP continue to evolve, the use of AI-powered tools, such as the Generative Pre-trained Transformer (GPT), in the legal industry has become increasingly prevalent. Many lawyers are experimenting with AI and grappling with its applications to streamline practices and improve efficiency.
While GPT and other AI-powered tools can potentially revolutionize aspects of the legal profession, it is important to consider the current limitations and potential pitfalls in implementing these technologies.
An overview of AI and GPT
At its core, AI is an interdisciplinary science, with multiple applications, focused on developing smart machines that can perform tasks that typically require human intelligence.
While the focus on GPT is recent, AI is not new and is foundational in numerous systems we use daily, such as:
- Natural language searches and processing
- Banking tools and applications to streamline payment processes
- Digital smart assistants like Siri, Google, and Alexa
- Facial recognition software
- Maps and navigation tools
- Search and recommendation algorithms that deliver customized search results and ads
- Self-driving automobile technology
- Some social media applications leverage AI tools to monitor content, suggest connections, and serve targeted advertisements
- Website chatbots
Advancements in machine learning and deep learning are shifting the way people live and work across industries. A significant NLP development is the Generative Pre-trained Transformer (GPT), which is a large language model that uses deep learning to generate human-like text.
Why all the fuss about ChatGPT?
ChatGPT exploded onto the world stage and elevated the discussion about AI. It became the fastest-growing consumer application ever launched, garnering 100 million active users in its first two months. Users are lured by the tool’s novel advanced capabilities, while others are concerned with its numerous implications, ranging from ethics, privacy, intellectual property, and disruption in various industry sectors.
Despite its deceptively sentient presentation, GPT does not feel or perceive. Open AI’s GPT-3, on which ChatGPT is built, is a predictive artificial model trained on more than 175 billion learning language parameters. This vast amount of text data allows it to generate coherent conversational text. Essentially, the system has a neural network machine learning model that can take your text query and generate a result that it predicts will be most useful.
Artwork by Ascannio / Shutterstock.com
Get the nomenclature right: GPT-3 vs. ChatGPT
GPT-3 (Generative Pretrained Transformer 3) and ChatGPT are both language models with different uses and capabilities developed by the company, OpenAI. GPT-3 is a state-of-the-art, deep learning model that has been trained on a large corpus of text data to generate human-like text. It is capable of performing a wide range of language tasks, such as translation, question answering, and summarization. GPT-3 has received a lot of attention for its advanced capabilities, but it's also quite large, with over 175 billion parameters, making it challenging to use for certain applications.
ChatGPT, on the other hand, is a smaller version of GPT-3 that has been fine-tuned specifically for conversational text. It is designed to generate responses to text inputs in a conversational tone, making it well-suited for use in chatbots and other conversational interfaces. ChatGPT has fewer parameters than GPT-3, making it more accessible and easier to use in real-world applications.
Potential uses of AI tools
Originality and precision
Because ChatGPT generates an original response that is so specific to the construction of the query, it is impossible for two people seeking the same information to receive the same response unless they use an identical query. This means that a lawyer using ChatGPT for drafting will receive a unique result to a query that no one else can replicate without using the identical verbiage. For example, compare the two responses provided by ChatGPT when presented with two queries that differed solely by inclusion of the words “an employer’s” in the first query:
Query: Draft an introductory paragraph for an employer's hybrid work policy.
ChatGPT response: As our world continues to evolve and technology advances, we at [Company Name] recognize the need to adapt to the changing demands of work. That's why we're excited to announce our new hybrid work policy, which offers the best of both worlds by combining the advantages of working in the office with the flexibility of remote work. This policy allows our employees to choose where they work based on what works best for them and the tasks they need to accomplish. Our goal is to create a work environment that is both productive and supportive of our employees' well-being.
Query: Draft an introductory paragraph for a hybrid work policy.
ChatGPT response: In today's rapidly changing world, the traditional 9-to-5 office environment is no longer the only option for how we work. At [Company Name], we understand that a one-size-fits-all approach to work may not be suitable for everyone, which is why we're proud to introduce our hybrid work policy. This policy allows our employees to choose the best combination of working in the office and remotely, providing the flexibility and balance needed to achieve both personal and professional goals. Our aim is to create a work environment that supports individual needs while fostering a productive and inclusive culture.
Meet your new virtual assistant
For law departments with lean or shared administrative support, ChatGPT can perform many functions of a virtual personal assistant. As a conversational AI model, without integrating into other platforms, ChatGPT can handle routine tasks such as developing initial drafts of letters, memos, emails, policies, handbooks, procedures, contract provisions, term sheet responses to frequently asked questions, training presentations, and other routine documents. It can also generate content within a requested word count.
Improved drafting and review
AI can streamline and automate many time-consuming tasks. ChatGPT provides responses in real-time. So now, you can generate the initial draft of a brief, policy, procedure, or letter, while running errands or picking up your children from school.
GPT can be trained on a specific set of legal documents to quickly generate highly accurate and legally compliant drafts. Additionally, GPT can be used to review legal documents for errors, inconsistencies, or omissions.
Accelerated research
Legal research companies have long used AI-powered legal research tools to reduce the time it takes for lawyers to efficiently find relevant case law and statutes, allowing them to focus on client service and other higher value tasks. These AI tools can also analyze vast amounts of legal data and identify key trends and insights, such as a judge’s leanings based on past decisions segmented by practice area. They can also predict outcomes of legal cases and automate routine tasks such as document review and contract analysis, reducing time, cost, and identifying key points in a case to help clients make informed decisions.
Furthermore, GPT can also be used to generate summaries of legal documents and cases, making the information more accessible and digestible for clients.
While it does not provide legal advice (and will tell you so if it infers that you are seeking legal advice), GPT can answer general legal questions, describe a legal process, and provide legal information. For example, a query such as, 'What are the elements of criminal conversion in New York' will generate a response purportedly based on the elements set out under New York Penal Law section 165.15.
Pro bono applications
Organizations that recruit lawyers to provide pro bono services can serve more indigent clients with fewer lawyers and increase the speed and efficiency of the guidance. They can also streamline the provision of responses to repetitive, basic questions.
Transcription services
AI is being used to increase the speed and efficiency of transcription services. For example, Otter.ai is a California-based technology company that uses AI and machine learning to provide speech to text transcription services. Consumers can upload audio files and receive a transcript of the file with quick turnarounds, depending on the size of the audio file. The company’s software, called Otter, also provides closed captioning for live speakers and writes and summarizes meeting notes, in addition to generating written transcriptions of the speeches. As with any transcription service, it is important to review the notes and ensure accuracy.
Limitations and risks of ChatGPT
Risk of misinformation and bias
If you ever used a dictation software, you might remember how scary it was that rather than providing a glaring misspelled word when the system was uncertain about what you said, it simply rattled off a string of grammatically correct gibberish that seemed plausible, even if conceptually ridiculous. Hence, you had to read your documents closely to confirm that the software actually wrote what you said. The risks of GPT are similar.
The maxim, garbage in, garbage out, is highly relevant to GPT. Because the system is trained on massive amounts of data, its ability to manipulate that data and respond to queries is only as reliable as the original data it received. Hence, before you cite that New York conversion law, you might want to verify that it is the most recent version and check the citation. Any biases and limitations in the underlying data will be reflected in the responses the system provides.
In theory, the more data the model is trained on, the more effectively the tool can detect patterns and predict relevant responses. However, increasingly vast amounts of data do not correlate to improvements in empirical accuracy. In other words, feeding the tool more data is no guarantee that erroneous information, bias, and inaccuracies, will be reduced. Vigilance is critical in verifying the accuracy of information generated by GPT before using it in court or in legal documents.
Because the system is trained on massive amounts of data, its ability to manipulate that data and respond to queries is only as reliable as the original data it received.
Inconsistent availability
As a reminder not to rely too heavily on ChatGPT, its availability is sporadic due to capacity issues. When unavailable, you will be treated to jokes, limericks, and other attempts to make you smile rather than being annoyed at the unavailability of a tool none of us had just a few short months ago. Notice also that this is a free test version while OpenAI refines the platform and determines how best to monetize it. In other words, the functionality could conceivably disappear at a moment’s notice, so don’t get too comfortable.
The end of lawyers?
Naturally, the question of displacement rears its head when considering whether AI tools can replace support staff and junior associates, resulting in job losses and economic disruption. GPT is not a substitute for human lawyers — and I’m not just saying that because I am one! The model cannot replace the critical thinking and legal expertise that human lawyers possess. For example, knowing the elements of criminal conversion is not a substitute for determining how to proceed against an employee accused of embezzling company funds.
Ethical considerations
In much the same way as autonomous vehicles raised questions about how to assign fault and whom to insure, the use of AI in the legal profession raises questions about ethical accountability where decisions and guidance are provided by autonomous systems. For now, the American Bar Association continues to place on practitioners traditional ethical obligations, interpreting the rules in the context of the use of AI.
Lawyers who rely too heavily on GPT may risk committing legal malpractice by failing to exercise the requisite care and diligence required under the ethics rules. Lawyers should understand their ethical obligations, both in the careful use of AI and their potential obligations to understand and use it where it benefits their clients.
Lawyers who rely too heavily on GPT may risk committing legal malpractice by failing to exercise the requisite care and diligence required under the ethics rules.
Intellectual property questions
As models developed by OpenAI, ChatGPT and DALL-E* raise some important concerns about intellectual property.
* DALL-E is another machine learning model created by OpenAI. Its name is derived from the artist, Salvador Dali, and an animated movie character, WALL-E. DALL-E generates images from textual descriptions.
Firstly, the vast amount of data that has been used to train these models raises questions about ownership and ethical use of that data. Additionally, the generated text and images produced by ChatGPT and DALL-E can potentially infringe on copyright and trademark laws. OpenAI has stated that it takes these concerns seriously and is actively working to address them, but it's a complex issue that requires ongoing attention and monitoring. The use of language models like ChatGPT and DALL-E is still in its infancy, and as the technology evolves, it will be important to continue having conversations about their impact on intellectual property rights.
Privacy concerns
GPT and other AI-powered tools require large amounts of data to be trained and operated. OpenAI fed ChatGPT an estimated 300 billion words that it systematically scraped from the internet. This data included websites, blog posts, books, articles, and whatever personal information was available.
User queries can intentionally or inadvertently disclose confidential information that enters the public domain (think, a lawyer on their work computer at a public company submitting queries about conducting mass layoffs or the timing of stock sales). The information is now part of ChatGPT’s database for use to further train the tool and can be incorporated into responses to other people’s queries.
The good news is that you can ask ChatGPT to provide a first draft policy for your employees on the proper and improper use of the tool.
According to ChatGPT, the system processes queries in real-time, with no long-term storage of the input (although your searches are readily visible and available to you for reference when you log in). OpenAI discloses that it may retain anonymized query logs for research purposes and to improve the model's performance. Feeding sensitive information into third-party AI models may inadvertently lead to improper disclosure.
Carbon footprint
A somewhat obscure but important drawback to the rampant use of certain AI tools is the increased carbon footprint.
Query: How much carbon emissions does ChatGPT emit?
ChatGPT Response: The exact amount of carbon emissions generated by ChatGPT is unknown, but it likely contributes to the carbon footprint of the data centers where the AI models run and the electricity used to power those centers. Data centers can consume significant amounts of energy, and the use of AI models like ChatGPT can contribute to that energy consumption. To reduce its carbon footprint, OpenAI is actively researching and developing more energy-efficient AI models and infrastructure.
Parting thoughts on usage
Whether you are still trying to hide from it or have dived head-first into the AI pool, lawyers should expect AI and GPT to continue revolutionizing the legal industry and generating novel issues and opportunities for litigation along the way.
Despite the proliferation of phone, fax, text, email, chat, and video conferencing, these tools have yet to adequately replace in-person contact. Similarly, lawyers are advised to be aware of the limitations and biases of GPT and use it as a tool rather than a replacement for human expertise. With the right implementation, AI and GPT could greatly benefit the legal industry, but it is important to be cautious and responsible when using these technologies.
Lawyers are advised to be aware of the limitations and biases of GPT and use it as a tool rather than a replacement for human expertise.
The risk of misinformation, legal malpractice, outdated data, and data privacy disclosure must be carefully considered before using GPT in legal practice. It is essential for lawyers to be vigilant in verifying the accuracy of the information generated by GPT and not to rely solely on it, but to use it as a tool to enhance the work. It is also crucial for lawyers to keep themselves informed and educated about the developments in AI and NLP, so they are aware of the new tools, new ways of working, and new risks and opportunities they bring.
In its current form, GPT does not replace your wisdom, experience, or expertise as a lawyer. It is especially useful in giving you a place to start other than a blank sheet of paper for drafting documents. However, it is not recommended that you rely on it as a primary source for learning areas of law with which you are unfamiliar, and do not rely on it to provide case citations.
