The use of AI in the legal space is not new at all. There are hundreds of legal tech AI vendors and thousands of law firms of all sizes and corporate legal departments (CLDs) that are using specialized legal tech tools that include the latest Generative AI tech, or older Machine Learning AI tools. However, in a 2023 American Bar Association survey that questioned attorneys whether they were currently using AI or considering purchasing AI-based tools, 58% of the respondents indicated they didn’t know enough about AI to answer the question. This shows that many attorneys are still unaware of the role AI plays in their work or the possibilities it presents, indicating that the topic of AI in the legal field is still unfamiliar. 

Of course, there was a lot of negative publicity when two US attorneys were censured for using ChatGPT 3.5 to “save time” and ended up using fake legal precedents in their pleading. However, what made this case unusual was that there are many specialist legal AI tools that can actually access extensive databases of legal precedents and case law. It seemed that the individuals did not seem to understand that the version of ChatGTP they used only contained data from its Large Language Model (LMM) training and was not connected to the web in general, never mind specific sources of legal information.  

In other words, Generative AI can be as useful in a legal context as it is in any other when the right guard rails and the right governance is put in place. This comes down to the phenomenon of “garbage in equals garbage out.” In other words, if you don’t feed your AI systems with good-quality data, how can you expect them to provide you with good answers? So don’t let the case of the two attorneys misusing ChatGPT deter you from leveraging gen AI in the legal field it’s all about using the correct tools and knowing their limitations.  

Many specialist legal tech AI tools use the same underlying models as Microsoft uses with their AI tool, Copilot, but there are thousands of LLMs and now Small Language Models, image-to-text, text-to-image, audio-to-text and other types of Generative AI models. In fact, hugginface.co has over 75 thousand models!  

It’s important to understand that a commercial LLM is trained on a massive set of data, but whether embedded in a legal tech product or accessed via Microsoft Copilot; it is being pointed at your private data, which it can understand and process to answer your questions. This is shown in the diagram below, and generally speaking, you don’t train a commercial model using your data; it’s already trained. However, if you wanted to, you could create your own custom LLM by training it on your company data, and Microsoft provides the flexibility to do that via Copilot Studio.

What the diagram above does not show is any of the security boundaries, the firewalls and permission boundaries that can keep your private data very safe even when applying Generative AI capabilities like Copilot to process it. Microsoft puts considerable effort into protecting your data in your tenant when you use Copilot and provides a lot of online resources for you to dive into to satisfy yourself, your senior partners or your KM lawyer on the security, privacy and confidentiality of your data.  

In summary, both old–fashioned machine learning AI and the new and exciting generative AI capabilities can be safe, suitable and valuable for lawyers, paralegals and others working in law firms and CLDs. The combination of Microsoft SharePoint Premium and Copilot, along with Power Automate for workflow and integration and Purview Information Lifecycle Management capabilities, holds tremendous potential value in legal industry scenarios. It’s crucial to keep governance and data quality elements top of mind to prevent garbage from being fed in and garbage coming out. Now that you’ve gained a foundational understanding of AI in legal, it’s time to learn more about how these mentioned tools can address the specific requirements of law firms and CLDS.