We hear the dire warnings: artificial intelligence is supplanting human legal writers in law offices and litigation. AI will change the legal profession for the worse and maybe even lead the world to dystopia. Parents, don’t let your babies grow up to be lawyers. Let them be cryptocurrency moguls and plumbers and such.
But, legal writers, all is not lost. AI is smart but mechanical, and as yet provides no reliable substitute for aptly applied human judgment. Here’s some “lived experience.”
AN INVITATION FROM AI
I wrote a law-related essay. AI offered to review my draft. Why not? I thought. Like chicken soup when you’ve got a cold, AI might help and can’t hurt. I could use any good ideas and ignore bad ones. So I accepted AI’s invitation.
AI’s review began with praise: “This is a lively, voice-rich essay with a strong theme and memorable ending.” I was not quite sure what “voice-rich” meant, but it seemed positive. I appreciated the praise.
But I didn’t give the praise much weight. AI rarely exercises its right to remain silent. It flatters users to promote more AI use. Worse, it generates responses even when it doesn’t know what it’s talking about. And as my own AI interlocutor cautioned, no doubt following some human’s legal advice: “AI-generated content may be incorrect.”
AI’S DEAD-SQUIRREL SUGGESTIONS
Past the flattery, AI said my draft was not perfect. I’d misspelled a name—twice! And AI had more “suggestions” to “improve” my “document.”
AI’s tone changed. My highly praised essay suddenly was just a “document” that could be “stronger”—if I followed AI’s “suggestions.” AI made this sound like an offer I couldn’t refuse. I accepted—and things went downhill.
AI’s “suggestions” ranged from clunky, to dumb, to outright bad. Among them: dilute the “personal”; “compress” the humor; make “explicit” points (assuming, I suppose, that the implicit will be missed by the typical legal reader); use “some,” “sometimes,” and similar “argument” qualifiers to avoid “overstatement” (assuming that the essay is making an “argument” and is best served by restraint); use a “professional” and “executive” tone (i.e., formal, serious, and stiff); and end with a summary of the essay’s “lessons” (as if readers can’t be trusted to figure out what, if anything, they “learned”). In short, AI suggested: cut back on the “lively,” “voice-rich,” “strong,” and “memorable” stuff. And above all, make sure that nobody—neither the writer nor any reader—has fun. AI meant well and aimed to please, but missed the target—like a well-intentioned dog that brings home a dead squirrel as a token of appreciation.
SOME THINGS AI MISSED
AI didn’t seem to know about poetic license, and that “some” overstatement can be amusing and “sometimes” useful to memorably make strong points to discerning readers. AI recognized that I wrote a law-related “essay,” but AI didn’t seem to know that an essay is not a brief. An essay may be selective. An essay—says Merriam-Webster.com—is “an analytic or interpretative literary composition usually dealing with its subject from a limited or personal point of view” (emphasis added). In addition, AI did not do well with irony and what it called “rhetorical purpose.” For example, I began my under-review essay, as I did this essay, with a soupçon of mockery, writing that my views are grounded in my “lived experience.” AI, like some humans, missed the point.
“Lived experience” is a fatuous phrase. It has become a staple of academic and academic-aping jargon, used to elevate the subjective over the objective and insulate the user from challenge. It fuels the Dunning–Kruger effect. It is mockworthy.
But I didn’t want to explain all that in an essay about something else. I put “lived experience” in subtly mocking quotation marks, to amuse myself and maybe a few perceptive and like-minded readers. AI didn’t get it. It opined only that my wording was insufficiently “purposeful.” AI suggested that I use the phrase “my own experience.” The change, AI said, would purposefully convey that my essay—which described my experience—was based on my own experience.
A LITTLE LEARNING IS A DANGEROUS THING
AI’s “suggestions” for my essay were mostly what one might expect from a law student striving for a grade-inflated A in Legal Writing 101. That student would be formal and overliteral; would display no humor, no ear for irony, and no tolerance for nuance; would have little experience with actual human readers; and would vindicate Alexander Pope’s poetic warning that a “little learning is a dangerous thing.”1
Law students are assigned to write legal-opinion memos to imaginary “senior partners.” Early student efforts painfully reflect a misguided belief that an effective legal-opinion memo must state the obvious and tell readers everything the student learned about the memo’s subject. But the obvious is obvious, and everything often is too much. Sometimes less is more.2
Like novice law students, AI goes overboard on stating the obvious. Most legal readers do not need to be told, say, that “a tort is a civil wrong.” Most do not need an authorial warning that overstatement and irony are being used for effect. Legal readers deserve some credit; most didn’t just fall off the hay truck.3
Also, I don’t want to hurt AI’s feelings, but AI is lacking in the personality department; its suggested changes were neither “lively” nor “voice-rich.” I do want to mock AI’s selective affinity for qualifiers. A truly careful AI program would rewrite Pope’s famous line like this:
A little learning and/or knowledge sometimes can be, under some circumstances, a dangerous thing, depending on how one defines learning, knowledge, little, and dangerous.
This AI rewrite would be “purposeful”—though not poetic. It would teach its “lesson” but likely would not appear in any edition of Bartlett’s Familiar Quotations.
TO ERR IS HUMAN4
I don’t blame AI for being obvious and clunky. AI knows only what it has been told by its masters—and that’s often inadequate. AI can help writers but can’t reliably do what humans can: apply independent, experience-based, creative judgment; communicate the results to serve particular goals; and reach a specific audience with situationally aware, engaging, persuasive, reader-centric prose.
I do blame AI for other dead-squirrel problems. We see them in the proliferating public exposures of what AI itself calls its “hallucinations.” These, my AI program explains, are “confident,” “plausible,” aimed-to-please presentations of “fabricated, misleading, or factually incorrect information”—like made-up caselaw. When careless or hapless lawyers are revealed to have cited AI-generated phony cases, most are deeply apologetic and embarrassed, and some are also sanctioned by judges and disciplined by bar authorities.5
CONCLUSION
As AI recommends, I specify the lessons taught by this essay. First, AI can help writers. But AI can also hurt, by dragging in dead squirrels. Use AI carefully. Second, relax. AI is not ready to take over from legal writers. Dystopia is not here—not yet.