Four steps for lawyers to harness AI without the risks

Lawyers live in an exciting time. Generative Artificial Intelligence (AI) suddenly appeared on the scene in Fall 2022 with OpenAI's ChatGPT. In no time, attorneys began making headlines for filing briefs with fake case law ("hallucinations"). And before lawyers knew it, they were being bombarded with new AI products or learning that products they already used now had AI. But no matter how you feel about it, AI is here to stay. Indeed, under Rule 1.1, attorneys have a duty of competence with respect to new technologies. Here is how to incorporate AI into your practice without ending up as a headline.

Step 1: Become fluent in AI

Before you can harness technology, you need to understand it. For example, most of the AI attorneys are interacting with are large language models (LLM), which are designed to forecast word sequences after being trained on massive datasets. ChatGPT is the now-famous example. What this means is that an LLM is not designed to be accurate; it is designed to guess what the next word in a sentence would be if written by a person. When an LLM lacks the data to respond to a "prompt" (i.e., our question or request), it produces a hallucination. That is where fake case law comes from.

Retrieval-augmented generation (RAG) is an LLM that can be more accurate because it is designed to retrieve relevant information from sources you provide. For example, you can upload a PDF of an appellate record and ask the AI questions to quickly find information while drafting an appellate brief. Gemini Notebook by Google is a useful, free RAG.

Agentic AI is the latest tool. These AI bots step outside the typical LLM chat window we are familiar with and can perform tasks out in the world semi-autonomously. They utilize multistep reasoning instead of having a back-and-forth conversation. Attorneys should be wary of how they integrate AI agents into their practice right now. They have the potential to, for example, run up a huge tab on Westlaw, offend a client or court staff, or leak confidential information. But, like all AI, they also have the potential to make us better at our jobs. For example, Deep Research features are capable of producing competent legal research requests with real authority.

But to really understand AI and stay abreast of the latest developments, attorneys should seek out continuing legal education, take certification courses (e.g., CLIO The Legal AI Fundamental Certifications), demo new products, and subscribe to legal tech news (e.g., lawnext.com). Learning how to prompt well can also impact AI's output quality. Finally, trying out AI also doesn't have to be all business: Suno.com can whip together a pretty good song in a genre of your choice about anything you want in minutes.

Step 2: Go slow and steady

When it comes to AI adoption, remember the old fable of the tortoise and the hare. Take a "slow and steady" approach to incorporating AI into your practice. You have probably already seen exciting headlines and press releases about how some law firms have adopted AI across the board, how they are saving money, winning more cases, etc. But given the risks associated with AI at this time, those who rush AI adoption may get burned. Try out AI tools in limited, low-risk ways and see how it goes. Build from there.

Most importantly, create a clear written policy for how AI can be used at your firm. Attempting to outright ban AI will likely only drive usage underground where it can't be supervised or regulated. Instead, become informed, approve certain AI tools, decide what attorneys and staff can and can't do with AI, and set up firm-controlled accounts for employees. Also, track usage to see if people are getting value from the products you subscribe to. Otherwise, unsubscribe.

Step 3: Belt and suspenders confidential data

The key to avoiding risk with AI is protecting confidential information. First, know that free AI tools are likely using anything you share with them to train their models. In other words, it is not being kept confidential. But paid accounts usually claim data is not used to train. The best approach is to carefully read the terms for each AI tool you consider adopting to understand how data is handled. Even then, don't trust it or share highly sensitive information. Redact or anonymize everything else. As some courts have already said, things shared with AI do not fall within the attorney-client privilege.

Step 4: Verify, verify, verify

At least for now, human judgment remains the cure for AI hallucinations. Every AI statement of fact or law that attorneys rely on must be verified by a person. AI companies are increasingly building the ability to verify into their products by providing citations and links to sources. But knowing what sources of fact and law you can trust and using them for verification is key.

Some may counter: "Having to verify everything AI writes defeats the purpose of using it in the first place." But let's say it takes you 30 minutes to write a section of a motion, demand letter, etc. Meanwhile, it only takes AI one minute to draft something similar. It probably then takes you five to 10 minutes to verify the accuracy of what AI wrote and to edit. With this approach, you just saved 20 minutes. Using AI properly is all about the long game. Shaving minutes off some tasks shaves an hour off your day and shaves hours off your week.

Be surgical too. Piecemealing your project will result in better quality in the end than attempting to kitchen sink it. For example, instead of asking AI to draft an entire motion, go paragraph by paragraph, section by section.

Conclusion

The invention of AI is impacting society on a scale that is equal to or greater than the invention of the internet. Many attorneys are afraid that AI will replace them. It won't. AI can replace many things, but it can't replace human judgment. However, while attorneys won't be replaced by AI, the attorneys who don't use AI will eventually be replaced by those who do.

The author wishes to thank Timothy A. Loranger for his valuable feedback on this article. This article was written without the assistance of artificial intelligence.