AI: Risks and Rewards

I’m not an early adopter. People on the cutting edge have the fun of being avant garde, but they also have to solve all the early problems. I’d put myself at more the early/mid-adopter.

Clio recently (early 2025) published a report called AI is Disrupting the Legal Industry. In it, Clio claims that 79% of lawyers use AI in some way, and 1 in 4 use it widely. Pretty good for such a stodgy profession!

Clio’s study estimates that nearly three quarters of the work of a law firm can potentially be automated in some way—higher for legal assistants and lower (57%) for lawyers. This should in turn lead to some restructuring within law firms. Clio suggests that firms can be more client-centric, or redirect some of their resources to marketing.

According to the study the tasks (amounting to 66% of a firm’s billable work) that are the most subject to automation are:

? Documenting and recording information (memos, summaries, notes, briefs);

? Getting data or information (reviewing data, searching records, evaluating evidence, legal research); and

? Analyzing information (taking the above and analyzing it).

Clio suggests that automating these tasks could let lawyers focus more on high-level advising, which can (possibly) be billed at higher rates. (On the other hand, possibly not.)

Automation may impact the business model of charging hourly rates. Bar associations are making it clear that if the task takes 20 minutes instead of 5 hours, the lawyer can only charge 20 minutes.

This creates a problem without an obvious solution. The big players in AI for legal research charge very high prices for their subscriptions—actually a double high subscription, first the legal research subscription and then the AI subscription. Right now, it’s still a hard sell that an attorney should spend a lot of money (that generally can’t be charged to the client) to greatly reduce the volume of work that can be charged to the client. No one seems to have a good answer for how attorneys can survive this business model, including the companies selling those subscriptions. A common guess is that this will likely push more to flat fees and hybrid rates.

While Clio’s paper didn’t discuss this, one good outcome could be if AI has the result of making legal services more accessible. Currently, law firm economics are such that if an attorney charges the client enough money to stay in the middle class, an average middle class client cannot afford that attorney. (Attorneys, as a group, are squarely in the middle class, ranging from lower-middle to upper-middle, and their staff earn correspondingly less. The reason the high hourly rate doesn’t translate into more money is high overhead and the many non-billable tasks.) This unaffordability means most of the middle class and below doesn’t have adequate access to legal services.

Some forms of AI are readily accessible. Models like ChatGPT and Microsoft are affordable.

What are the risks at this early adopter stage? One big risk is that AI is still unreliable. Its work quality is quite uneven. More seriously, it makes stuff up. The technical word for that is “hallucination,” but the fact is, if it doesn’t know, it just makes stuff up. I was doing a research project with several legal issues. To save some research time, I tried asking ChatGPT4 for a summary and some cases on each issue. On several of the issues, it produced a nice summary with standard black-letter law. On one of the issues, it also produced a nice summary with case citations and a description of the holding of each case. The problem was—those cases didn’t exist. At all.

On another occasion, I was having trouble getting into ChatGPT4, and my question history had disappeared. I asked the program why, and it informed me that with a new upgrade, the content history was no longer going to be available because ChatGPT had changed its model. That was a total lie. It was that day that much of the broader Internet had gone down, and ChatGPT just wasn’t working. All the data reappeared a day later.

Another risk is the data security piece. We address that by using the paid subscription, where you can choose not to share your data for training purposes. But because we don’t know if we can truly trust the program, we don’t load up material that is highly confidential, just in case. Still, this is just a new twist on the same old problem. Is your email really secure? Is your website? Ask all those large corporations who’ve gotten hacked.

What are the rewards? Using AI has saved a lot of time. AI is great for summarizing material, creating outlines from material you give it or from its database, writing drafts of letters, memos, policies, and so on. It needs correcting and supplementing, but it definitely saves us (and clients) a significant amount of time.

David has also had fun using it to generate images for our Telios Teaches material. He has it create several versions of an image he has in mind, then has it modify and update the image until he has what he wants. I think he actually uses several different programs. What you end up with—though AI-generated—is still somehow obviously David’s brainchild.

I’m cautiously optimistic about AI—but not at all trusting. And I believe it will reshape our world—but also that the final, perfect Word has already been spoken, and His plans will not be thwarted by any intelligence, artificial or otherwise.