These are the topics and questions that I hear the most from my law firm clients as they explore bringing AI tools into their environments. I will note that there are many references here to “GPT” as one of the primary AI models. There are others, including Google Gemini, that are popular. But all of my firms thus far are working with GPT, so that’s what I will speak to.
Is AI just the latest buzzword?
There’s no question that AI will be a force in the legal industry — as one lawyer told me, “AI won’t replace lawyers, but lawyers using AI will replace lawyers who aren’t.” I’ve seen enough use cases over the last couple years that I believe firms should be leveraging AI on behalf of their clients. But now, it feels like every tool is coming out with “AI features”. In my experience, many of these new features are “vaporware” — existing features that are repackaged as AI, or features that don’t really move the needle. So I would encourage you to treat what you hear from vendors regarding AI with a bit of skepticism, and to really lean on them to prove how they’re using AI to benefit firms and their clients.
“GPT” vs. “ChatGPT”
We hear people confuse these two terms a lot. Bottom line is that GPT is the underlying AI engine, while ChatGPT is a user interface everyone knows about that creates a conversational experience over the top of GPT. When you use GPT-4 as part of a technology platform, or when you “bring your own” GPT (see below), you are using a specific instance of the GPT engine, closed off for your firm. You’re NOT using ChatGPT; in fact, most firms have policies that forbid lawyers from putting client-sensitive data into ChatGPT (if they allow its use at all on firm computers).
Use their AI, or bring your own?
It’s really important that you ask any vendor what their approach is here. Larger platforms from the bigger vendors will include access to AI models like GPT-4. Other tools will allow you to build applications or workflows and then point them at your firm’s existing AI instance. Firms can go to Microsoft and get their own Azure instances with AI enabled – an instance of GPT that is closed off to the firm only, giving you the power of AI while keeping your data secure. I imagine it’s cheaper to bring your own AI than to essentially buy it as part of the bigger package. Of course, if you go the DIY route, your IT team needs to be able to source that AI instance, manage it, and so forth.
“Does the AI learn?”
Lawyers will often ask if the AI is “learning” based on the types of questions they’re asking it. When you use a tool like GPT-4, each question you ask it is a new question, and it’s as though a new person answers it each time. It would be like if you sent out a question to a distribution list and any paralegal could grab it. Just as with a distribution list, you may get slightly different answers each time, depending on who answers, the complexity of the question and the amount of context you provide. The way you improve the answers you get from AI is to improve the “prompts” that you provide it, including how you define the AI’s job, how you provide context, and how you describe the output you’re expecting. As you develop an AI-enabled application, it’s basically “trial-and-error”: you are likely to consistently iterate on your prompting, to improve the results based on how the application performs each time you run it. Once you’re satisfied that it is returning suitable responses, you may still return to the prompting over time to include new context or cover edge cases.
Client requirements
Just like anything you do with client data, you’ll need to refer to your client’s Outside Counsel Guidelines to see what is allowed and not allowed when it comes to AI. Many businesses are no further along in understanding and leveraging AI than many firms, which provides an opportunity for firms who get AI and are using it to provide guidance to clients as to how it works and how it can help.
Integrations
With any new piece of software you buy these days, it’s important to know that it can connect to something else. Some integrations are “native”, meaning a connector has been built into the tool specifically to integrate with another tool (e.g. a connector to iManage or NetDocs). Others use APIs (“REST” API being a main one), which generally means that if each tool has API connectivity, you can connect the two tools together. Then there’s sort of a middle ground: tools that leverage REST API but build little connectors that make it easier to make inbound or outbound connections. Make sure that any new tool you purchase, AI-enabled or otherwise, can be integrated with other tools, just in case.
Use cases
Does the firm have a solid plan for leveraging AI, and the stomach to pay for a larger platform that can launch many new AI-enabled applications? Do your existing tools offer AI features as an upgrade? Or do you have a specific use case that you want to solve, or at least start with? Certain vendors price their tools on an application-by-application basis; this may (or may not) be more expensive on a per-application basis, but can be a good way to get started without committing too much. The per-application model also generally comes with unlimited user licenses, so you aren’t punished for growth. Just make sure that you know whether you can scale a single application to multiple clients (e.g. you build it once for Client A and then duplicate it for Clients B and C) without incurring additional per-application costs.
Portability
With any technology you bring on board, it’s important to know how you can pull out your data, should you decide to switch tools later. If you’re hosting client data or documents in these tools, it should be relatively easy to download and/or export those items whenever you want.