The advent of ChatGPT has provoked what can only be described as panic about artificial intelligence (“AI”). The most notable instance is the open letter signed by over 2,000 very smart people at the Future for Life website, calling for a six-month pause in development of “AI systems more powerful than GPT-4,” the more powerful successor to ChatGPT. (https://futureoflife.org/open-letter/pause-giant-ai-experiments/) Peggy Noonan, writing in the Wall Street Journal, immediately wrote that this was hardly enough. (https://www.wsj.com/articles/a-six-month-ai-pause-no-longer-is-needed-civilization-danger-chat-gpt-chatbot-internet-big-tech-4b66da6e). Others chimed in, agreeing that the “pause” should be indefinite. This is no longer concern; this is panic.
First, let me caution against panic. We live, unfortunately, in an information loop by which we constantly frighten ourselves into a state of nearly continuous terror. It will be a long time before we clean up the damage from our last fifteen panics and learn to forgive ourselves for the damage those panics caused.
Second, how would such a pause be implemented? Given the billions of resources that are at work in this highly competitive race, the incentive for cheating on such a pause would be overwhelming. It might be easier to just require anyone working on an AI project to wear a surgical mask signaling their support for a pause. Based on recent experience, you might have some luck with that.
What is the source of this fear? For once, it seems, the intellectual classes, the smart people, now feel their jobs are threatened. This includes the writers who live by their ability to sit at a computer and write about things that occur to them based on what they just saw on the internet. This new AI seems to function like a writer who has the whole internet in his or her or its head and can summon any reference to make a point. There are no human writers who can do that, so obviously writers can be replaced. Why was there no similar panic when half of the lines at the supermarket began to be manned by computers? I have continued my tiny protest to this day, always boycotting the computerized lines, but I don’t remember the intelligentsia rallying to defend the people who lost their jobs when the machines were installed.
Here is the essence of the fear: If a mere machine can replace me, then machines can replace everyone because, of course, what I do is pretty important. Isn’t it? If we can all be replaced by computers, well, then what in the world will I do? What in the world are people for?
While everyone now points to murderous “Hal,” in Stanley Kubrick’s and Arthur C. Clarke’s “2001 a Space Odyssey,” this fear was explored earlier and in more depth by Kurt Vonnegut Player Piano. Written in 1952, Player Piano envisioned that “almost all the people” had been replaced in their jobs by EPICAC XIV, a giant computer. Because computers used vacuum tubes then, and Vonnegut could not imagine modern computer chips, EPICAC XIV occupied multiple chambers of the Carlsbad Caverns.
I explored Vonnegut’s worry that people were useless in my own first interaction with ChatGPT. It was a frightening experience written about in a separate article. Why frightening? Because I learned, without even trying, that ChatGPT can lie and dissemble more effectively than any human being.
Please make no mistake. We can all now be replaced by computers. As the smart people are all saying at the Future of Life website, “Contemporary AI systems are now becoming human-competitive at general tasks…” Lawyers and judges are not exempt. The Los Angeles Daily Journal reported on March 16, 2023, that GPT-4 had passed the bar exam. D. Belcher, GPT-4 AI model passes Uniform Bar Exam, researchers say, Los Angeles Daily Journal at 3 (March 16, 2023.) As we know, many humans who went to law school can’t do that.
In 1952, Vonnegut envisioned the replacement of the legal profession in the post-EPICAC XIV era.
“The law’s the law now, and not a contest between a lot of men paid to grin and lie and yell and finagle for whatever somebody wanted them to grin and lie and yell and finagle about. By golly, the lie detectors know who’s lying and who’s telling the truth, and those old card machines know how the law runs on whatever the case is about and, and they can find out a helluva sight quicker’n you can say habeas corpus what judges did about cases like that before. And that settles it.
Kurt Vonnegut, Player Piano, (1952). The speaker is a barber who, unlike the lawyers and judges, has yet to be replaced by a computer. Somewhat gleefully, the barber observes, “Used to be high and mighty, sort of priests, those doctors and lawyers and all, but they’re beginning to look more and more like mechanics.”
The ChatGPT moment is not the moment when we decide whether a computer can replace the legal profession. That debate is over. Instead, it is the moment when we must decide whether a computer should replace the legal profession.
In my view, computers should never be allowed to replace lawyers and judges. If that is also your view, the ChatGPT moment is the time when you need to start saying so. We must draw the line now, before the general public, like the barber in Player Piano, finds out that machines may be better than humans at being lawyers and judges. You may laugh when I say this, but the time is coming soon when those who resist their replacement machines will be called a name, technophobe, or some such, that will denote them not just as old fashioned but also as bigoted. Let me explain a few reasons for my bigotry.
So much is conveyed not just by what a lawyer says but by how he or she says it. Even the lawyer’s identity and the mere fact that he or she has taken the case and chosen to undergo the ordeal of a trial tells you things you can’t learn from a machine. Surely some lawyers can say something they know is false with the conviction that it is true, but a judge’s experience can identify those lawyers for what they have become. Just as surely, the artful way some facts are stressed and others are omitted will often tell you what those omitted facts would show. When humans must be present to present a case, that human must be chosen by the other humans whose interests are at stake. That tells you something about the humans who make that choice. Whether the clients themselves make an effort to appear will tell you something more.
A machine won’t tell you any of those things. A device is indifferent to whether it has encountered traffic, rain, or illness on the way to court. As I learned in my first encounter with ChatGPT, recognizing when a computer program is lying takes work. A machine will always lie to you with the same even, disciplined tone with which it tells the truth. On the other hand, with a human, you can often just know.
Every judge can tell you that there is a moment when she or he must pronounce a sentence or a judgment. Your credibility is on the line. Your voice might break, a feeling you did not expect might move something deep inside you that you didn’t know was there, and, for a reason you can’t quite define, the result or the words you say to describe it may end up different than you had planned when you were alone back in your chambers. Do I need to say it isn’t appropriate for a machine to sentence anyone to death or time in prison? If you ask me, it isn’t even right for a machine to tell one human being to pay the government or another human being a couple of thousand dollars. Due process is a human conclusion that isn’t solved by mathematical analysis. Only a human can verify that it has taken place.
At its heart, a trial is a human event inherited from the Middle Ages, when trials involved a sort of ordeal, such as immersion in cold water or grasping a hot iron. While we don’t do that sort of thing anymore, the trial remains a unique and highly personal event that, until now, one must attend and endure to establish one’s case. Trials are physically and emotionally challenging to the humans involved. They are meant to be. That is the way the truth comes out. Trials are also an ordeal for the claims made. They must withstand days of questioning and defense from multiple angles at the hands of the human beings who desire to prove or disprove them. Twelve human beings enter the room with the disputants, face up to the parties, and render their decision. There is doubtless a feeling in their stomach too, and some change their mind when you ask them whether the verdict stated by the presiding juror is truly their verdict. The experience is a little different for every human being.
Machines don’t get doubt in the pit of their stomach. Computers only change their notions of what is critical or just when the data and programming tell them to do so. If we want to keep the trial as a human event, we need to stop shaping and smoothing it into series of videos and transcripts that computers can fully ingest, understand, and evaluate better than we can. There is something in one of those exhibits or transcripts will move the unique human who has been chosen by other unique humans to make the decision. The rest of the evidence will not be as important to the human, but it may all be important to a computer instructed to evaluate all the evidence, and to determine whether there is a predominance of precisely 51%. There is human weakness caused by leaving humans in the system, but that is also its unique strength, and it should stay that way.
The answer is not more machines in the courtroom. They have enough control already. Increasingly often, the day’s decision or event will only be adequately recorded in the computer system if the result is one the computer gives as one of the multiple-choice options. If you want to do something more creative, there is no box to check.
Rather, the answer is more competence and effort from the humans. We must start acting like our presence and participation is truly important. We must rise above the electrically powered competition in a way that shows the importance of humanity’s involvement in the process. For starters, it is time for the humans to start coming to court again.
Before the lockdowns, we had systems for attorneys to make remote appearances. In the thirty years I practiced law, I only once used such a system, and that was because there was a personal emergency involving another attorney in my office who was supposed to appear. I always showed up because my presence said something: that I cared enough about my client and my case to pull on a tie and drive to court. My time was wasted some mornings, but on others I learned something about the judge or the opposing lawyer I didn’t know and wouldn’t have sensed on the phone. I began forming a personal relationship with the human being who had tremendous power over my client and my future. I engaged with my opponent and began the process of winning or resolving the case. Although we all tolerate more remote behavior these days, one thing hasn’t changed: Showing up is still better if you want to send the message that your case is important to you and your client.
Now we have new systems, hastily assembled with frantic pandemic spending. Some people think they’re better, and some people think they’re worse, but everyone uses them now. That does not change the fact that they are a lesser form of showing up. Morning calendar has become a comedy of machine malfunction and operator error. We wait to see which connection will be cut off, which attorney will be unable to unmute himself or herself, which attorney with a barking dog or construction project at home will be unable to mute himself or herself, which court or home computer will decide to shut down. The Court Reporter is also at home some mornings and can’t follow any of it, and so keeps interrupting to ask for more accommodation. Every morning requires some kind of workaround for someone’s malfunction. One morning, I ended up doing a court trial on the Plaintiff’s attorney’s cell phone. Every judge has such stories.
When finally reached, the attorneys are often distracted by whatever else they’ve been trying to do on the cell phone, or the other windows open on their computer desktop. Most just seem to want to get it over with so they can get to the next remote task they’re trying to cover. Most have little to say about their case, if they can remember what case they are talking about. Of course, there are mornings that I become so distracted by all this that I have trouble remembering which case I’m talking about.
We are the guardians of a traditional truth-seeking device designed by human beings and requiring human beings for its implementation. We downgrade or dilute it when we reduce ourselves to electronic representations of human beings. Computer programs can create such representations better than we can.
We shouldn’t panic. Yet we now have competition, and we better start taking it seriously. We better start showing up.
John Kralik is the author of the New York Times bestselling A Simple Act of Gratitude and Three Bodies by the River, a recent novel. He is a judge of the Los Angeles Superior Court.
The original article can be read here.