AI is changing everything, even the notoriously cautious legal industry. While it took decades for TAR to be widely accepted and used, AI will be normalized in just a fraction of that time.

But right now it’s still very new, and many legal teams have questions about what it is and how to take advantage of it.

Here are three things to keep in mind as you get deeper into the AI conversation and start exploring how AI might help you achieve your eDiscovery goals.

1. AI means different things to different people

AI has evolved so much over the last five to 10 years that people can draw very different boundaries around what it is and isn’t. Some people consider any and all analytics to be AI. Others associate AI with the technology behind tools like ChatGPT—intensely smart and powerful platforms that use large language models (LLMs) to analyze and, in some cases, produce language with amazing competence.

Now that AI is all the rage, companies are eager to highlight anything that uses AI in any form, which can lead to some awkward conversations. If you think AI means LLMs, and the person you’re speaking to thinks it means TAR with machine learning, you can go pretty far down the road together before realizing you’re talking about completely different things.

We like to define AI in terms of its purpose, which in eDiscovery is to model human judgment.  With today’s massive datasets, people make millions of decisions, of varying degrees of complexity, throughout the eDiscovery process. AI can assist with many of those decisions.

With that in mind, we think the most important thing for legal teams to know is what they want to achieve with AI. Which “human judgments” do you want AI to support? What challenges are you trying to overcome?

Be specific about the use cases you care about underneath the AI label. This will help you find a technology or service provider with the kind of AI you need.

2. Different kinds of AI are good at different tasks

Since a variety of technologies can be considered AI, it makes sense that they’re effective at a variety of tasks.

But this is actually quite easy to miss or misunderstand today. Marketing messages often sound the same, even for very different AI products. And technology like ChatGPT is perceived as being all-purpose and all-intelligent—as if you can drop anything in its lap and it will return the answer or action you need—when its utility is actually quite specific. (Namely, a chat-based AI is good at responding to factual questions and generating sample text based on prompts or source material that you provide. We dig deeper into this type of AI in our next post.)

We’re focused on using AI to improve the quality, pace, and amount of linear review. That’s where the majority of eDiscovery costs are concentrated, so that’s where we think AI can provide the most value. Our AI is built with LLMs and can help with things like responsiveness, classifying privilege, and other sensitive information.

But, like all AI, ours isn’t designed to do everything. It’s designed for large datasets and computational tasks that require nuanced analysis. For smaller, targeted tasks like email threading, you don’t need an LLM. You need something made for that task.

The bottom line is: Don’t expect to use the same technology for every problem. You may use different tools on different matters, depending on their needs and constraints. You may use multiple tools in combination. While that may seem complex, the right partner will manage the workflows and explain them to you in a way that makes it simple and sensible.

3. AI needs to be integrated with care and intention

Legal teams are opening up to AI much more quickly than they did with previous technologies. But they still need to be pragmatic about it.

To optimize your use of AI, you should expect to augment and transform your current workflows. In the past, widespread adoption of TAR and other eDiscovery analytics was accompanied by widespread use of human experts in information retrieval and other areas. The same should happen with AI.  

The challenge is that popular examples of AI are so user-friendly and readily available—anyone can open ChatGPT and start chatting away—it feels like anyone can use them. It’s also tempting to put different forms of AI in places where they’re unnecessary, just because it feels like the thing to do.

We’re very excited about AI, but we’d hate to see the shimmer and excitement of AI woo anyone into adopting it in the wrong context or without the support they need to make it worthwhile. It comes back to the notion of where and how AI can provide value. Think about your use cases and be very intentional about where you position AI in your workflows.

Bonus tips: more ways to educate yourself on AI

We hope articles like this are helpful, but there are lots of other things you can do to get more familiar with the AI story, as well as the tech itself. For example:

• Talk to the AI experts in your team or organization. Almost everyone has a pilot or exploratory AI program now. Find those people, and talk to them about their work, outlook, etc.
• Play with the models available today. Even if it’s just bantering with an AI chat platform. It helps to get a sense of how these tools feel and what is (and isn’t) possible with them.
• Ask questions. Online, at work, with friends and colleagues. You may get some helpful information, or you may find someone else who’s wondering the same things and can join you in seeking answers.

[View source.]