Caging the AI beast

While showing me the hidden treasures of Maui, where he lives, my son Damian introduced me to ChatGPT. As we drove past a flock of cattle egrets and I asked "Do they migrate?," Damian turned to the cell phone on his dashboard. "Let's ask my friend." The answer popped up in seconds.

When I got home, I started using the chatbot for legal research. It's very user-friendly. I can ask questions using normal language, and it understands me -- even if I don't use the magic words that Westlaw or Google search often need. I always check what the bot gives me, of course, because I'm the one who signs the brief.

I did have one unsettling experience. I was working on an appeal where the parties' contract said that it was to be construed under the law of Washington State. So I asked ChatGPT for Washington cases that applied one particular contracts rule. Up popped several cases, quoting some very helpful language from each. I downloaded the cases from Westlaw. Lo, behold -- no such language! Indeed, the cases had nothing to do with the question I asked! What the hell?

I called Damian, who informed me, "It's called a 'digital hallucination.'" What's that? His answer was way over my low-tech head. But he assured me that it's real, though rare.

Though it would never occur to me to take the bot's word for it and not check the cases, apparently that does occur to some lawyers. News reports dribbled in about furious judges scolding and sanctioning lawyers who submitted briefs containing citations that did not support what their briefs said about the cases.

I was glad to see a tough judicial response. Lawyers who fail to check their cites should be slammed. I've always checked up on cases cited by treatises that I've usually found reliable -- like Witkin and Restatements. We're a profession, with huge responsibility for our clients' lives and fortunes -- plus a related duty to support a justice system that maintains the confidence of the public. It's crucial that we do things right.

You go to a doctor, kvetching about chest pains. The doc examines you, then advises: "My AI site says surgery is a bit risky, but medications are riskier. So based on that, we'll slice you open tomorrow morning." Shouldn't the state medical board do something about this guy?

Which brings me to my latest encounter with AI.

I'm working on an appeal, and perusing the trial court record. I come across a ruling by the trial judge, which includes the following: "Counsel contends that his position is supported by two cases: Smith v. Jones (1995) 134 Cal.App.4th 28, and Shmeckle v. Shmuck (1980) 197 Cal.App.3d 876. However, neither of these cases exist."

"Neither of these cases exists"! I've been doing law for a long time, and I thought I've seen everything. But this?

My attention grabbed, I kept reading the order -- expecting the hammer to fall. But much to my surprise, the judge said nothing more about it. No sanctions, no threats -- not even a mild scolding.

Why? Did her honor bestow mercy on a young, newbie advocate? Did she know the lawyer from before she took the bench? Or did she want to avoid spending time dealing with threatened sanctions and the inevitable excuses and apologies?

I fear something more insidious. It's called "habituation." This trait "is a fundamental type of learning where a person or animal gradually stops responding to a repeatedly presented stimulus that is not threatening. This desensitization occurs because the brain filters out constant, non-threatening information, allowing the individual to conserve attention and energy for more important stimuli. For example, a person living near a busy street will eventually stop reacting to the constant noise."

Here's another example. Try to imagine a President who is offered a gift from a foreign government -- say a megabuck airplane for his personal use. This comes from a country with whom the President must negotiate tough deals in his official capacity, to get the best results for the American people. But he accepts the gift, saying "Why wouldn't I accept a gift?" This gets headlines, and we, the public, are shocked. It gets worse. Seeing what happened, foreign officials drooling to curry favor with our Prez start offering business deals to the Prez's companies, to his family, and to his buddies and bootlickers. And they take it! These incidents too are reported, for a while -- but then they migrate towards the back pages, near the sports sections and comics. Then they disappear.

New incidents of corruption occur, but now hardly anyone notices them -- we've been "habituated." And they spread -- even Supreme Court Justices begin accepting goodies from sleazeballs who have cases pending in the Court.

Of course, this is pure fantasy. In the real world, none of this could ever happen. Indeed, back in the 1950's, President Eisenhower's chief of staff -- Sherman Adams -- accepted a single gift (a $700 vicuna coat) from a guy under SEC investigation. Public outrage forced him to resign. (How quaint.)

I do not want us to become habituated to the AI-sourced citation problem. We should nip this possibility in the bud -- "we" being either the State Bar or the judiciary, or both.

For starters, a new Rule of Court should be adopted, saying something like: "If there is reasonable cause to believe that a brief cites authorities that were generated by artificial intelligence without proper checking by counsel, the court shall issue an order to show cause directing counsel to explain why sanctions should not be imposed."

Note the "shall." The judge should have discretion regarding what to do to the lawyer -- but the judge must do something.

As time goes by, AI will become stronger and more widely used. We need to cage the beast now. We can't habituate ourselves to accepting misuse by a few misguided attorneys.