MY AI NIGHTMARE
I’m no hard-core tech expert, but I do handle tech appeals (three now pending). So I’ve been following the recent developments in artificial intelligence. Many advances, plus lots of fear.
Fear of what? Well, coincidentally, I’ve been pushing my 12-year old son Ari to elevate his tastes. So the other night we watched The Terminator together. (You ask, “That’s elevating? What kind of crap has he been watching?” Don’t ask.)
The flick begins in the future, where machines are battling humans for control of the world. The machines seem to be winning, but the humans have a leader named John Connor who poses a real threat. So the machines send one of their kind back to our present to “terminate” John’s mother before she conceives him. No mommy, no Johnny.
The Terminator is a “cyborg”—made to look human, but underneath all metal and wires. Strange, but he has an Austrian accent, and looks like our erstwhile ex-governor, Arnold what’s-his-name. He is extremely intelligent, and stores a vast amount of data. And he is very determined. His whole existence is formed by a single goal: to kill John’s mother-to-be.
That night, after the movie, I had a dream. Actually, a nightmare. The Terminator comes back as—O the horror, the horror—an appellate lawyer!
I arrive for oral argument at an appellate court with a stack of stuff: my notes, my briefs, the trial court record, and a few key cases. I represent the respondent, and we’re first on the calendar.
In walk a woman and a big guy. She looks like an ordinary lawyer. The man looks like Arnold. He’s wearing a three-piece suit, a red power tie, and Clark Kent glasses (with no lenses). He sports a few grey hairs, giving him an air of maturity.
She carries a small purse. He carries nothing.
The woman slips a small disc into a slot in the back of Arnold’s head, then sits down and starts reading a spy novel. I saw what was written on the disc: “Goodguy v. MegaCorp.” That’s my case!
The presiding Justice calls the case, and Arnold gets up with a broad smile. “Good morning, Your Honors. I’m Arnold Sy Berg, representing MegaCorp. This case is about . . . .” When Arnold slurs a few words, the woman takes out an oil can and squirts a few drops into Arnold’s jaw. Then she goes back to reading her book.
Arnold proceeds to present a succinct, persuasive summary of his arguments for reversal. Brilliant.
How will he handle the Justice’s questions? Quite well, it turns out—though the questions were predictable, easy to program answers to.
But then one of Their Honors tosses a curve ball from left field, as often happens. This curve is quite unpredictable. In my head, I quickly think up an answer, though I know of no authority to back me up.
Arnold pauses for a microsecond. His fake eyes flicker and roll. He then comes out with, “Good question, Judge. It arose in a 1935 New Mexico Supreme Court case, Coyote v. Roadrunner. Here’s the citation, and here’s how the court answered it. The court said . . . .” He goes on to quote from the court’s opinion—word-for word!
I woke up screaming “Noooooooo!” How could I possibly win an appeal against that kind of performance? This “thing” had come back from the future to terminate me. I might as well retire my shingle—or buy my own cyborg.
The dream haunted me all the next day. When I’d calmed down, I began to consider the bottom line: who would win more often—Arnold or me? That question took me to: what does it take to persuade an appellate court? Can a machine do it as well as a human?
I don’t think so.
OK, I admit I’m not qualified to foretell the future in tech matters. All sorts of things will happen that no one—let alone me—can predict,
But here’s the key: the target. As long as the target of one’s persuasive efforts is a human being, a human being will have the best chance of doing that job successfully.
A machine would do a better job of “persuading” another machine. If our appellate judges were machines, a smart engineer could figure out how the judge-machine comes to decisions, and then program a lawyer-machine to feed into the judge-machine’s system whatever it needs to come to the “right” decision. (Of course, the lawyer-machine would need to have the right record on appeal and the right supporting statutes and cases. He could not just make these up. If the engineer could not provide these, he might advise the client to drop the appeal.)
Human judges are much more complicated. Yes, they have a logical side that follows the record and legal authorities, just like a machine does. But they also have non-logical sides. Each one has had a different life experience. Their disparate upbringings will affect how they see the world and the issues brought before them. Each judge has a different sense of “justice.” Each judge has different personal tastes, in brief-writing and oral argument styles. And each case is decided by a panel of multiple judges.
When I orally argue a case, I’m very attentive to what the Justices say. But I’m also focused on how they say it. Their tone, their variations in pitch, their facial expressions, their body language. All these tell me something about which way they are leaning, which issues they care about, and where I should focus my argument.
And here’s another very important cue. By custom, appellate judges tend to ask questions rather than make comments. But simply answering their questions literally is a big mistake. Behind every judge’s question lurks a viewpoint. When the judge asks, “Is the language of the addendum to the contract consistent with your interpretation, counsel?” it’s likely that she’s thinking, “It’s not consistent, and that’s why I’m going to rule against you.” When this happens, it’s time for me to drop whatever argument I had pre-planned to deliver, and immediately switch all my remaining time, energy, and intellect into trying to change that judge’s mind about the addendum. If I don’t, I’ll lose the appeal.
There’s a name for our ability to catch these cues that go beyond the raw text of the words spoken by another person. It’s called “intuition,” as in: “How did you know she was going to do that?” “I dunno, just intuition, I guess.”
Intuition is the cumulation of our subconscious perceptions of what the other person is thinking, feeling, and wanting. It is so complex that we have a lot of trouble reaching into ourselves, extracting and dissecting each separate component, and explaining how we came to a certain conclusion—usually within seconds of receiving the cues. We’re too lazy and clueless to do all this, so we say: “I dunno, just intuition.” We can easily explain our conscious reasoning, but it’s very hard to explain our subconscious intuition.
All humans have this “intuition,” in varying degrees. (The lack of such intuition is often a key symptom of autism.) Children don’t develop the ability to “put themselves in another person’s shoes” (called “theory of mind”) until age four or so, but after that, they learn it instinctively, without much overt teaching.
Why do we have “intuition”? Because we’re social animals, and eons of evolution have honed this ability to attune ourselves to the thoughts of others, so we can function in society, get along with other people, and survive and prosper (and reproduce!) in a group setting.
I use intuition when I write a brief—even though I don’t have the visual and aural cues that will be available when I’m speaking with my targets. I’m constantly strategizing—thinking about how my arguments might affect the judges and law clerks who will be reading them. When I do legal research, I seek cases that include juicy, persuasive language that I can quote. I might describe a case or statute in a way that is accurate, but slanted towards a policy that helps my argument. My statement of facts is supported by the record and the appropriate standard of review, but it tells the story in way that I think will be viewed favorably towards my case. If I represent the appellant, I include rulings and reasoning from the trial court record that makes the trial court judge look bad. And my appellant’s opening brief might set up a trap, luring the respondent’s counsel to make an argument I can knock out of the box in my reply brief.
All these stratagems turn on my ability to predict how judges, law clerks, and opposing lawyers will read what I write. It helps if I know something about them as individuals. But even without that, I can use my experience with similar people as a touchstone.
It’s worked pretty well for me. Sure, I employ the same logical arguments that Arnold uses. But experience teaches me that, in many appeals, logic ain’t enough. So I reach into my intuition bag with every sentence I write and every word I utter.
Trial lawyers do the same thing, perhaps more so. When deciding whether to use a peremptory challenge against a prospective juror, the lawyer must make a quick judgment based on slim information—a jury questionnaire, the juror’s answers to mostly non-penetrating questions of voir dire, visual and aural cues—and a guess about who might replace a bounced juror. All about people you’ve never seen before, based on split-second “intuition”.
Could Arnold do these things? Certainly not now, but can artificial intelligence ever be developed to replicate a human’s ability to intuit?
I don’t think so, but what do I know? There are some hyper-smart engineers working on artificial intelligence, and they are well-funded by some very large tech companies. If it can be done, they’ll do it.
But—at least for now—I’m pretty sure The Terminator won’t get me.