Stop the AI scold; focus instead on the substance of what the AI makes https://x.com/i/article/2117699564929200128
Against AI-shaming
As AI becomes more pervasive, and more people begin to use it to write, to read, and to organize their daily life, there has been a rise among enlightened sorts to condemn those who use AI. A certain self-righteousness has welled up in our society, demanding that those who deploy the technology identify themselves. Writers are told they must reveal the methods they used and the technologies they consulted. The consequence of those disclosures is never fully — or reliably—itself disclosed.
All this is idiotic and unfair. That’s not to say that there aren’t contexts in which the use of AI ought to be condemned. It is simply to say that the idea that you should condemn the technology simply because it is used is idiotic. We need to recognize just how deeply unfair and inegalitarian—not to mention wildly inefficient—this anti-AI bias has become.
Let’s start with the context where AI should be condemned. I teach in a law school. Students come here to learn how to become lawyers. To become a lawyer, you need to learn to read and write legal material. Writing is difficult. To read a case requires extraordinary care. To read a statute can be a real struggle. To write a convincing legal brief or a compelling argument takes skill and practice. As I tell my students, if you come to law school and use AI to read and write, you are wasting your money. You will never learn to do or think like a lawyer if you’ve never practiced the actual craft of lawyering. Claude can’t go to the gym for you.
Yet there are a million contexts in which we read and write that are not meant to be teaching us anything, and instead, are contexts in which we could be helped if we could read or write more efficiently. That’s especially true if, for example, someone is writing in English and English is not their first language. Or even if they’re writing a brief in a legal case, and they know they cannot write as quickly or as clearly as Claude can. In those cases, and a million others, using AI well makes perfect sense. And shaming AI use is, well, again, idiotic.
And even worse, hypocritical. Because we already have plenty of contexts in which we all understand a writer is not a writer, and in which our measure of that writer-who-is-not-a-writer is what s/he wrote, not how s/he wrote it. Consider, for example, one clear context in the law in which we recognize today that “writers” are not actually writers: judging.
Among judges, there are at least three types. There are judges like the judge I had the honor to clerk for, Judge Richard Posner. Every single word of the appellate opinions that Judge Posner wrote was a word that he himself had typed. (And he was by far the fastest typist I had ever seen, but that’s a different matter.) He used law clerks for the purpose of critiquing his writing. He would hand us a draft. We would read the draft and give him comments. He would take the comments, and he would rewrite the draft. The final product was as much his as any novel by Stephen King is Stephen King’s.
Other judges use clerks differently. The other judge I had the honor to clerk for — Justice Antonin Scalia — had clerks draft his opinions. But Scalia worked through every draft meticulously, changing practically every sentence to conform to his understanding directly. So that writing was not all his, but it was ultimately his. And no one who reads a Scalia opinion has any doubt that it is the opinion of Justice Scalia.
But Scalia’s practice was not the practice of all justices on the Supreme Court. Indeed, without naming names, there were some who would change four words in an opinion written by a law clerk and then publish it under their own name. Those law clerks, no doubt, were outstanding, but no President had nominated them, and the Senate had not confirmed them. Nonetheless, they had effectively written a Supreme Court opinion, which the other justices then signed on to.
At least among lawyers, none of this is news. Yet no one critiques the opinions of justices by asking whether the opinion was clerk-written or not. No doubt, some, including Posner, criticize that opinions are clerk drafted. But in general, the question is whether the reasoning is good. An opinion stands or falls based on the seriousness of the argument. It doesn’t stand or fall based on who actually typed the words. Thus, in the context of judging, we’ve succeeded in recognizing that authorship can be attached to an author, even while recognizing the author is not ultimately responsible for every word that was written.
Now, of course, this is not to say that in some contexts, we aren’t right to insist that a human do the work and all of the work. Bridget Dooling’s brilliant article, Ghostwriting the Government, makes a compelling case for contexts in which we rightly demand a decisionmaker make a decision rather than adopt the polished reasoning of someone — or some thing—else.
And of course, to remark that judges don’t write every word of their opinions is not to ignore the regret one might have about that fact. The most striking experience I had clerking for Posner was the number of times he went home intending to write an opinion one way, and yet came in the next morning with an opinion that came out the other way. “It just wouldn’t write,” he would tell us. That was an example of the act of writing actually constraining what the judge did, which, of course, is an ideal of the rule of law. By contrast, when clerks draft judges’ opinions, there’s always a way to reach a result. It is a rare clerk who would tell a judge that she could not, in fact, write the opinion in the way the judge, or the panel, or the Court had directed. There is always a way; clerks are smart and can always find it.
Thus, to recognize how, in fact, we recognize judges as authors is not to say that there’s nothing to criticize about how things are. It is simply to recognize how things are, and how that might well be explained, even justified, by the burden on judges today. Ninth Circuit judges average about 350 merits cases a year; they issue about 106 written opinions. To imagine the judges processing those opinions on their own is to imagine superhumans when all we have are humans.
The same point is true in other contexts. Barack Obama was, in my view, a brilliant president. He was also extremely meticulous. I saw him in just one meeting, but what I saw others confirmed: The President listened carefully, and he took copious notes as he did his job. But when he gave speeches, or issued statements, his words had been crafted by others. Not exclusively: No doubt, he edited what others had written, sometimes more extensively than other times. But Obama, like every president, did his work by being supported by some of the very best in the nation. The list of those who worked for Obama is the list of the greatest talent in policymaking in America. But again, nobody sits back and asks what proportion of Obama’s speech Obama himself wrote. They ask: Is the speech a good one? Is the argument a good one? Is the policy sensible?
Now, of course, it could be different. Indeed, at a different time, maybe it was. Just over a hundred years ago, Washington was rocked by a public lands scandal that is referred to as the “Ballinger-Pinchot affair.” Richard Ballinger was Secretary of the Interior. Federal investigator Louis Glavis had charged Ballinger with misconduct. Future justice Louis Brandeis represented Glavis. On September 13, 1909, President William Howard Taft issued a letter exonerating Ballinger of any wrongdoing. Brandeis challenged the letter, charging it had been effectively written by Oscar Lawler, an assistant attorney general assigned to Ballinger’s department. Brandeis’ charge was not quite true. Doris Kearns Goodwin has shown that though Taft’s letter borrowed from a memorandum written by Lawler, the borrowing was limited. Nonetheless, though the concern was whether Taft had reached a truly independent judgment, what’s weird is the suggestion that his writing was actually his writing.
The same point is also true beyond government. CEOs of corporations issue reports; they give speeches; they write letters. The vast majority of this work is done by others. No doubt, others that the CEO has hired, whether directly or indirectly. But what we ask about this work is not who did it, but whether the work is good; whether the policies or decisions it reflects are good decisions or policies. The question we ask is substantive, not procedural. (Except again in the contexts Dooling describes.)
The point is that we live in a society where all sorts of important people constantly use others to do their writing and speaking and preparation for speaking. All of us have adapted to that reality by ignoring the fact that the writing is not done exclusively by the person said to have written the work. All of us understand these important people have other people working for them.
So why don’t ordinary people get the same privilege? As we move into a world where anyone can recruit AI-agents increasingly just as talented as the very best assistants working for CEOs, or judges, or Presidents, why do we shame the ordinary people using AI when we don’t shame CEOs, or judges, or Presidents for using staff? Is writing, as a Leona Helmsley sort might say, just “for the little people?”
Because as anyone who has spent any time trying to understand the potential of these new AI technologies sees, they are an enormous gift to an extraordinary range of would-be creators and writers and innovators and managers. These technologies, especially the emerging technologies of agents, are giving to anybody who takes the time to learn how to use them an endless supply of extraordinary assistance. AI assistants can read through all your email and find a message you remember you sent 20 years before. They can keep track of your travel schedule, notify you of conflicts. They can draft thank-you letters. They can draft complaints. They can do an increasingly large portion of all the things that we must do in this world, and that ordinary people must do themselves, but that important people have done for them. We are increasingly in a world where it is not even a joke to say, “Have your AI speak to my AI.”
And yet, just at the moment when this technology is giving the relatively powerless enormous power, we’ve launched whole industries to ferret out the use of AI. EU governments are forcing AI companies to watermark AI output, so those who use AI can be identified. The ostensible purpose of the regulation was to address deception and manipulation. But no doubt, the watermarking feeds AI-shaming.
Again, I’m not saying we shouldn’t police that delegation in some contexts, maybe many contexts. Education is the perfect example, and not just education. We want kids and students to learn how to write and how to read. They can only do that if they do that themselves. Who knows how long humanity will really need humans to write and read. I hope forever, but it’s not my call. I am sure that right now, we need to educate people to do what computers may be able to do better than us. Commercial pilots learning to fly an airplane do not use autopilot while they learn. But literally 90% of the time a commercial pilot flies an airplane today, they are using autopilot. (Indeed, pilots must learn both how to fly without autopilot and how to use autopilot to fly. Similarly, some law schools have started to teach law students how to use AI to do legal work.) In that context—indeed, in every context where the elite get the benefit of assistants—the question we should be asking is, what's the most efficient way to do your work well?
In every context, of course, we should question the output. Of course we should ask whether it is any good. We should criticize AI-slop — because it is slop, not because it is AI-generated. We should demand the work be as good as it makes sense for it to be. But for the vast majority of what we could call purely-functional speech—by lawyers, by judges, by bureaucrats, by business people—what we should not be doing is AI-shaming. We should not be criticizing people simply because they used AI to help them in their writing.
Yet the opposite is increasingly the reality today. It’s easy to hate the technology (unless you’ve fallen for its incredible power); it’s easy to be (rightly) furious with the idiotic policy failures that have allowed the industry to grow while imposing externalities on the rest of us. And it should be easier to get people to recognize the real threat the technology could present to humanity generally. I get all that. But those easy critiques have morphed into a constant scold, just as this empowering technology is giving ordinary people the freedom that wealth and power have given the elite since the dawn of time.
Don’t get me wrong: I don’t want AI to write articles for me. By writing this essay, I am emphatically not asking for that permission. I love writing. I think it’s right that authors who present themselves as authors present work that is their work alone. I do use AI to do the sort of things that I would have used law students to do — to find sources, to help me research a question, to check sources, to fill out footnotes, to explain the reasoning of certain sources in footnotes, to critique what I wrote. (And btw, I used m-dashes long before AI.) But the text is mine, and when I write something in the text, it has come either from my fingers, or increasingly from my speaking into the extraordinary AI technology called Wispr Flow.
But the people who are using language to simply get things done, whether bureaucrats or business people, should be embracing the technology to the extent the technology helps them do their work better. Obviously not uncritically, obviously not without checking and verifying what is produced. But why isn’t it obvious that this gift of creative power should empower all of us, especially those not already given regular staff.
Yet the infrastructure of AI-shaming grows nonetheless. It is astonishing to me that the continent most responsible for the protection of privacy—Europe—is mandating a technology to out ordinary users for the ordinary use of empowering technologies. It makes no sense. Don’t shame the technology; shame bad work.
Or at least if we are, as a culture, going to start shaming ordinary people for using AI, then let’s at least be consistent. If the relatively powerless are going to be shamed for using AI, let’s shame the relatively powerful for using staff. Let’s shame judges for using law clerks to write their opinions. Let’s shame Presidents for using speechwriters to write their speeches. Let’s shame CEOs who use fancy assistants to write their reports. Let’s shame everybody who presents work as their own when we all know that their work is theirs, plus the team they lead. I doubt we will embrace that shaming. I wouldn’t support it. I would instead support the reasonable right that anyone should be free to use the legitimate tools they have available to them to do what they legitimately are trying to do—the relatively powerless as well as an enlightened elite.
I understand why this view may seem heretical, especially if the last time you touched AI was a free account, 3 years ago. It is not possible to understand how quickly the models have changed without using them. I desperately hope lawyers always read and write by hand. I am confident that if they don’t, the practice of law will be weakened. But I’m also quite confident that if every brief in the Courts of Appeals—and possibly the Supreme Court—were written by at least the best models today, the quality of briefing on average would rise substantially. If I’m right about that, then I’m confident my desperate hope for the future will not be lawyers’ future.
That doesn’t mean the law will be worse. To the contrary, it could well be much better. The law as it is is an awful system: slow, expensive, unequal in its reach. Given the capabilities we’ve seen so far, AI could render the law both worse for lawyers but much better for everyone else.
That’s speculation. What’s present and real today is this: AI can help people do their work or engage their play. Anyone who uses it is responsible for what it produces, just as a judge is responsible for what her clerk writes. We should measure the output based on the output: Is it good, or true, or honest, or fair. That an AI helped someone say something that was good or true or honest or fair is not a reason to be critical of the AI, or of them for using the AI. Could it have been better had they not used the AI? Maybe. It depends on the writer. Is it better because they did use AI? Maybe. Again, it depends on the writer. But who cares about that hypothetical in either way? What we should care about is that more good work gets done, as easily and as efficiently as possible.
Finally, a few words about technology-lament: Among my colleagues, there are some who believe we should stand athwart the technology, and yell stop, to remix William Buckley a bit. That we should fight the technology, at least in the context of the law. That we should insist on handmade law, whether by lawyers or by judges.
There are plenty of contexts in which I would join that fight. I want the New Yorker to be handmade, always. The same with the Times. I know I don’t have to worry about Stephen King novels. Were he still writing opinions, I know I would never need to worry about the opinions of Judge Posner.
But to demand that the law banish AI is to ignore the core truth about law in America today: As I’ve already said, the law sucks. If we had a well-functioning system, delivering justice cheaply and quickly, I’d be happy to fight anything, including any technology, aiming to change it. But when we have the embarrassment of a system that we have now—despite the good faith effort so many make to make it work better—we need the professional courage to acknowledge that failure, and strive to make it better, however we can. The law in America is in its late-Brezhnev period. Let us hope its Gorbachev is more politically astute, and that no Putin may live in its future.