Custom legal GPTs are a new and emerging technology that can help lawyers automate and optimize various tasks, such as drafting contracts, summarizing cases, conducting research, and writing correspondence, pleadings, legal memoranda and briefs. However, they pose some challenges and risks, such as ethical, legal and technical issues. In this article, we will provide an overview of what custom legal GPTs are, how they work, and what benefits and drawbacks they have for lawyers. We will also discuss some best practices and tips for using them effectively and responsibly.
A custom legal GPT (Legal GPT) is a variation of the original GPT model that has been fine-tuned or adapted to the specific domain of law and legal practice. A Legal GPT is designed specifically to assist lawyers with legal research, drafting, and analysis. It is built on a foundation of large legal datasets and fine-tuned on a lawyer’s specific practice areas and needs. The legal datasets that it is trained on include statutes, rules, regulations, court opinions, contracts, and academic papers. They are basically trained on the same legal dataset that the traditional online legal research vendors utilize.
Key features of Legal GPTs include natural language processing capabilities that allow lawyers to query the system using plain English questions and receive direct, concise answers. They understand legal terminology, concepts, and arguments. Legal GPTs also have advanced legal research skills that can quickly surface the most relevant cases, statutes, regulations, and secondary sources on any legal issue. They continuously scan new legal information to stay up-to-date. An additional benefit is drafting assistance. Legal GPTs can generate first drafts of legal documents such as contracts, briefs, memos, and more based on a lawyer’s specifications. The system learns from revisions to improve its drafting over time. The AI assistants can provide legal reasoning and analysis functions that allow lawyers to propose hypothetical scenarios and arguments to evaluate their strength and practice for court. Legal GPTs can identify strengths and weaknesses, counter arguments, and precedent cases. One of the best features of a Legal GPT is that it can be programmed for customizations based on practice area, jurisdiction, clients, and an individual lawyer’s preferences. The more it is used, the more tailored its outputs become.
The goal of a Legal GPT is to augment human legal work, not to replace lawyers. It can handle time-consuming research and drafting to allow lawyers to focus on higher-value legal analysis, strategic thinking and client interaction. With an assistant like a Legal GPT, lawyers can work more efficiently and effectively.
Law firms can create their own unique Legal GPT system trained on its own internal data. A law firm may collect and aggregate documents the firm has already produced (work product), such as client correspondence, legal pleadings, memos, briefs, contracts, etc. Customizing a Legal GPT to a specific firm’s data, practice area, and needs provides advantages over a generic legal AI assistant. Many firms already have existing letter templates, previously researched and written pleadings, and brief banks. These documents form the training data and dataset for the AI system. To create the custom Legal GPT for a specific firm, the dataset will need to be preprocessed by sanitizing or anonymizing client information and confidentiality. The dataset will also need to be restructured so it is usable by a machine learning algorithm. Then the AI model can be trained on the firm’s dataset. Lawyers supervising the training can provide examples of good outputs to guide the training. It can also be adjusted to focus on practice areas that are most relevant. The dataset can be updated by feeding new documents, research, and legal documents authored by the firm into the system. Continual learning will allow it to adapt to new practice areas and a specific lawyers’ style. The model will probably need to be continuously refined with ongoing feedback from a skilled, experienced lawyer. Lawyers should identify errors, and suggest improvements to further customize the system. With training and customization unique to the firm, or a lawyer’s practice area, the resulting custom firm specific Legal GPT can provide lawyers with tailored, relevant suggestions drawing on the firm’s existing work products and knowledge base. Finally, a custom GPT can be trained on the specific domain and style of the law firm, to ensure that the generated texts are accurate, consistent, and compliant with the legal standards and norms.
Taking this concept one step further, a custom legal GPT has the ability to provide case analysis and predictive outcomes, based upon the firm’s past success (and fail rate), litigation history, settlement amounts, billable hours, etc. The custom GPT could be used to analyze this legal data to identify patterns and trends for planning and resource allocation.
There are some drawbacks and concerns to consider with a Legal GPT assistant. First and most obvious are the ethical concerns. AI assistants may encourage human over-reliance on the AI, rather than independent and objective legal analysis. Lawyers must retain accountability. There are also risks perpetuating biases in data training sets. The system may also be unclear on how it arrives at responses. If Legal GPTs are used by non-lawyers, there is the potential and risk for the unauthorized practice of law. Consent should be obtained if the confidentiality of client data is used for training. Without ongoing human supervision by a lawyer, the responses may become outdated or inaccurate over time.
To address these concerns, rigorous protocols around ethics, data practices, security, and quality control by a lawyer are essential. Lawyers should retain responsibility for verifying AI outputs before acting or advising clients. Transparency around capabilities and limitations is critical. Further policy and regulation will help guide proper and beneficial use of legal AI tools. Lawyers should actively supervise the system to catch any errors.
Several companies are already offering Legal GPTs to lawyers and to the public at large. A simple search on Google produces multiple vendors. Most of the traditional legal research platforms are offering this same service. Harvey AI is actually working with law firms to create custom Legal GPTs based upon a firm’s prior work product and statutory and case law. See Generative AI for Legal Research, MSBA Blog August 24, 2023.
A Legal GPT trained on a law firm’s own data can provide many benefits to a law firm, but also carries important ethical, legal, and technical concerns. Benefits include gaining a legal research assistant that quickly produces relevant firm knowledge and documents and provides drafting support to improve efficiency. This allows a law firm to leverage its own expertise, learned knowledge and previously researched and prepared legal pleadings, memoranda and briefs to generate high-quality legal texts for various purposes without “reinventing the wheel.” It can also generate summaries of legal cases, opinions and multiple documents by extracting the key facts and arguments from the original texts or previously produced work product. However, lawyers must retain responsibility for verifying all system inputs and outputs before advising clients. Risks include over-reliance on AI, perpetuating biases, reducing transparency, and possibly legal liability if used improperly. Rigorous training and testing, security protections, quality control protocols, and ongoing lawyer supervisions are essential to ensure the custom Legal GPT responsibly augments, not replaces, lawyers and their human legal analysis and judgment. Lawyers should understand both the advantages a tailored legal AI can offer as well as the diligence required to address the associated risks and limitations. A custom legal GPT is not a plug and play solution, but rather a tool requiring thoughtful lawyer strategy and analysis, and management. As with all AI used by lawyers, lawyers must ensure data security and privacy. Nuances of the law and the understanding of human behavior require human oversight of all AI tools. Lawyers may want to be mindful of their ethical obligations when using these tools.