Steven Pinker has written nine books and devoted his life to studying language, cognition, and writing. In this interview, we started with the practical stuff — his essential rules for writing — and then had a lively conversation about AI’s impact on writing.
Here are the takeaways:
The curse of knowledge is the biggest threat to clear writing. It can be harder to write about a topic you’re an expert in because it’s so easy to forget what it's like to not know something, which makes you overestimate what the reader knows.
The easiest way to fight the curse of knowledge is to show drafts of your writing to people outside your field.
Shakespeare said: "Brevity is the soul of wit." The point is that saying something in fewer words will almost always make it better because it requires less cognitive load for the reader to understand.
Ok, let's try again: Saying something in fewer words will almost always make it better.
Ok, one more time: Remove needless words.
One reason why writing is harder than speaking is there's no real-time feedback. You have to imagine the audience's reaction.
The best thing you can say about how LLMs write is that the sentence structure is sound. But the downside is how generic and banal the outputs are.
18th and 19th century writing is more vivid because the abstractions that modern writers use hadn't been invented yet. Calling somebody "pathologically aggressive" isn’t nearly as vivid as saying: "They grabbed me by the throat."
Generalizations without examples are useless, and examples without generalizations are pointless. You need to marry them both. Generalizations show the big picture. Examples make them concrete.
The more vivid a piece of writing, the more people can form a mental image of what you’re saying. Avoid abstractions: frameworks, paradigms, concepts. All those things. Get concrete, so people can see what you’re actually talking about. For example, don’t talk about a “stimulus that awakened your senses” when you can say: “I got excited because I saw a cute bunny rabbit.”
Academic writing should be clear. I mean… if the taxpayers are funding most of the research, shouldn’t they be able to understand it?
Transcript
David:
I'm here at Harvard in the office of Steven Pinker. He's written nine books and devoted his life to studying language and cognition and writing. And so what we did in this interview is we started off with the really practical stuff. We started off with his rules for writing.
What makes him unique is that he's been thinking about AI since the 1980s. So if you're interested in doing great nonfiction writing in the age of LLMs, well, this interview is for you.
I want to talk about the curse of knowledge, and I want to talk about this cartoon from your book, which says, "Good start, needs more gibberish."
Steven:
Yes. When I pose the question, why is there so much bad writing? Why is there so much academese and bureaucrates and corporaties? People's favorite answer is captured by that cartoon, namely that bad writing is a deliberate choice.
In various versions, it's academics with nothing to say dress up banal ideas with gobbledygook to show how sophisticated they are, or pasty-faced nerds get revenge on the girls who turned them down for dates in high school. People want to erect a kind of cult that no one else, outsiders can't penetrate because they haven't learned the jargon.
I don't think that's the best explanation for bad writing. Partly it's personal. I just know enough people who have plenty to say. They're brilliant people, and they have no desire to obfuscate. They're just incompetent. They just don't know how to express themselves clearly.
There's something called Hanlon's razor: Never attribute to malice that which can adequately be explained by stupidity. These are not stupid people I'm talking about, but it's a kind of stupidity in not knowing where your audience is coming from.
I illustrated with an anecdote of a conference in Technology, Entertainment, and Design, better known by its acronym TED, where a brilliant molecular biologist had been invited to present his latest findings. He launched into what was obvious to me is the exact kind of talk that he would give to his peers in molecular biology. Within about four seconds, he had lost everyone because he just spoke in jargon without even introducing what problem he was solving, why it was significant.
It was launched right into the experiments. There was a room of several hundred people from many walks of life, from entertainment, from design, and it was obvious to everyone in the room that no one was understanding a word. Obvious to everyone except the distinguished biologist, who was just clueless.
This is not a stupid man, but he was very stupid when it came to communication, namely, not everyone knows what you know.
The curse of knowledge, a term from economics, is the difficulty that we all have in knowing what it's like not to know something that we know. That is to subtract something from your brain.
Put yourself in the shoes of your audience, whether it be in public speaking or in writing. Figure out, where are they coming from? What do they know? What don't they know?
I think that's the main cause of bad writing. You get abbreviations and acronyms that no one has any way of knowing. You have jargon that is known only to a tiny little clique. You have abstractions. The level of the stimulus was proportional to the intensity of the reaction. What it really means is that kids look longer at a bunny than a truck. It's so familiar to you that you don't think it's worth explaining to people concretely what they're supposed to be seeing.
So all of these are manifestations of the curse of knowledge. It goes by other names: egocentrism, absence of a theory of mind, that is a theory of what's going on in other people's minds. If I had to identify the single biggest flaw in writing and communication...
David:
Would be that that's it. So when you're writing your books, I know you go up to Cape Cod and you write for as long as you possibly can. Now, I would assume that one way to get around the curse of knowledge is just to talk to the kinds of people who would be reading your book.
Steven:
Yeah, you show it to people.
David:
So what do you do? What do you do? Because you write really intensely. So when you're writing, how do you get around that?
Steven:
Well, I do something that is, I know is not good enough, but I do it as best I can, which is I try to imagine what it's like for someone not to know what I know. That is, I try to cultivate my sense of empathy. But the problem with the curse of knowledge is you don't know when you're subject to the curse of knowledge. Because something that seems so obvious to you that you don't even question whether other people know it turns out not to be obvious.
So anyway, I try, but at the end of the day, I show it to people. When my mother was alive, I would always show her a draft of my book. Not for the reason that most academics cite, namely referring to my mother as the epitome of an unsophisticated, not very well-read, not very bright person. My mother was extremely intelligent, extremely well-read, very sophisticated, but she wasn't a cognitive psychologist. She wasn't a psycholinguist. She didn't know what I knew.
And when I write, I don't write for just a random sample of the population. They don't buy my books. I write for people who are intellectually curious, who have some degree of education, however, not for peers in my field.
And so my mother being an example, but also, of course, when you publish for a commercial publisher, you have an editor, and the editor is typically very smart, but again, not in your field. And I show it to people in different fields who are academics.
But it's surprising how insular even academics are when it comes to other academics. Sometimes there'll be people here in this building, in my own department, sometimes my own subdivision within my own department, like students, and they'll give me their thesis proposal. And I just don't know what they're talking about sometimes in my own field because they've been immersed in like five or six people in their lab, their supervisor and the other grad students and a couple of postdocs and a research assistant. They've all been consuming the same jargon that as soon as they step outside that tiny little circle, they're unintelligible.
So even showing it to, if I'm showing it to a friend who's an economist or a historian or a political scientist or an evolutionary biologist, not being a cognitive psychologist, they'll say, I'm sorry, but I just don't know what you're talking about.
David:
Yeah. In terms of your writing, you've done so much work looking at vision and how the brain works, and it seems like a lot of your writing advice is informed by that.
Like I guess if I were to try to summarize it, it's like a lot of our brain and basically the way that we move through the world is indexed heavily on vision. And so writing well means being concrete and helping people see what it is they're trying to write. Is that a good way of describing it?
Steven:
Yes. And that would probably be the second bit of advice on the list. Find some way of getting into, get into your readers heads, but don't just depend on your ability to get in their heads. Actually get a flesh and blood person to actually read it and see if it makes any sense to them. So that'd be number one.
Number two is, I study language, and as a writer I live in language, but language is kind of overrated in the sense that what understanding consists of is not a bunch of words. It's not blah, blah, blah, blah, blah, blah, blah. But language is a means to an end of getting people to appreciate the ideas you're trying to convey, which are not just a string of verbiage.
And those ideas very often are visual, and motoric, that is bodily, emotional, auditory, but they're sensory. Or they're even conceptual, but they aren't just a bunch of vowels and consonants. And so constantly allowing your reader to be able to form a mental image based on what you're writing is the next key to good writing.
That is, don't talk about a stimulus if you mean a bunny rabbit. Don't talk about a level or a perspective or a framework or a paradigm or a concept, all of which mean a lot to you in your day-to-day work. But no one can form an image of a paradigm in their mind's eye.
David:
So how do you do that? One thing that you've said is, use, for example, all the time. What else can we do?
Steven:
So often visual metaphors are helpful. One of the reasons that often the prose of other eras strikes us as so much more vivid.
David:
I was gonna say lush.
Steven:
Lush, yes. Partly because they had the advantage of writing before. There were several hundred years of academia and intellectuals inventing terms and abstractions, and they had to appeal to images that were part of people's common knowledge. So instead of saying something like aggression or antisocial behavior, they might say, the spirit of the hawk kneaded into our flesh.
David:
Yes.
Steven:
Now, we wouldn't write like that because we can say aggression or antisocial behavior. And that's jargon that a lot of people are familiar with, but they weren't always with us. And before that, there had to be some way of referring to it in terms of an image that everyone shared. And I think that's why often the writing of previous centuries just strikes us as so much more gripping, so much more evocative and powerful that they had to appeal to visual metaphors.
David:
Yeah, I've never thought about that before because a lot of the Bible I think of, like, I don't know why the Owl of Minerva came to mind. There's just all this symbolism in animals. It always is striking.
And so what you're saying is, like, a lot of the concepts that academics and people brought in, obviously they're kind of a more efficient way to communicate, but they do lack that kind of visual quality that makes writing vivid.
Steven:
That's right. So they do make it. In fact, they're essential for doing the work within the profession. You have to be able; if you're a biologist, you have to be able to talk about things like ecosystems and species and systems and reagents and potentiation.
All these concepts that you don't want to have to go back to basics and talk about. Well, there's a certain amount of chemicals and we call that a concentration. And when it increases over time, we call it—you're beyond that after you're a freshman. And so you have more and more abstract terms that you could refer to enormous bodies of knowledge just with like two syllables. That's very good.
The problem is that then, now, when it's time to convey them to someone who isn't at the pinnacle of specialization in your field because of the curse of knowledge, you're apt to forget that these abstractions, which are kind of basic to you, are just don't even need to be defined, aren't basic to anyone else.
David:
What do we need to know about writers? About how hard writing is? And I mean it in this way. Speaking comes so naturally to us. But then writing is something that we sort of have to learn, right?
You watch a kid and you talk to a parent who has a 20 month old, they're like, yeah, you know, they're speaking now, they're like, oh, you know, you wouldn't believe it. You know, they're talking so much, they're crawling around and it's always like, yeah, it's like that. And then you talk to them who have parents who have an 8 year old kid's like, how's the writing going? It's like, well, you know, it's going a little slow. There's not that same excitement. And so it's as if, like, why is it that writing is so unnatural in a way that speaking is so natural? Like what is going on there scientifically? And then practically a number of things.
Steven:
One of them is that in conversation you don't never have two people that are parachuted on a stage and immediately have to begin a conversation. They have some common ground to begin with. They know why they're there. They're talking about something that is in the air that they're both familiar with. It was the reason for them having the conversation in the first place. They can get away with using terms that in context are perfectly clear, like this and that and the thing and what I was talking about. And she.
Whereas if you are not privy to that little social circle, you may not know who they're referring to. In writing, you're wrenched from the context. Someone's picking a book up off the shelf and they've never met you, and they may be living in a different country. You might be dead. They've got to pick up all of this detail from what's there on the page. Not what's in, not the common ground that the two people bring to the conversation.
Also, when you're speaking, you know that you're speaking to someone. You know their idiosyncrasies. You're a little bit better at avoiding the curse of knowledge, partly because you get feedback.
David:
Right.
Steven:
The furrowed brow, the quizzical expression. The what? The request for clarification. I'm sorry, I just don't know what you're talking about in real time. And, you know, in the body language, the engagement, even in a live audience, any speaker knows when people are starting to fidget and drift off as opposed to continuing to be riveted. None of that is available in writing.
David:
Absolutely. How about this one? I think generalizations without examples and examples without generalizations are both useless.
Steven:
Yes. Well, useless might be a bit strong, but generalizations without examples, I find even in my field, nine out of ten times, I just don't know what they're talking about. It's like, what do you like? Give me an example.
It's too, because a generalization erases detail. It sweeps over particulars. You often just can't really know what it's referring to. And the abstract words in a language just aren't precise enough. They often have, we have probably 100 concepts for every word in the English language. And so a particular word, especially if it's abstract, won't call to mind a particular referent.
The example pins it down to what's the general ballpark that you're talking about? What is the example the generalization is generalizing about?
Now, examples without generalizations, it's well, why don't you just tell me that? Like, you know, and your point is, and apropos of what?
David:
One of my favorite ways to think about this is that there's always a trade off between context and compression. So if I could wax poetic for the next five hours about my train ride to Boston today, but that's way too long.
Or I could tell you, yeah, it was an easy train ride, but then you didn't actually get anything from that because there's no context there, and that maybe we're always kind of balancing the examples kind of give us that context, and then the generalizations are the compression, and it's actually in the swing and the dance of the two of those that good writing and good communication happens.
Steven:
Yeah. So, for example, if I were to say that familiar words don't have to refer to the literal meaning of their parts, you can understand that. You can kind of say, yeah, yeah, okay.
And then I say, well, for example, a bathroom isn't necessarily a room with a bath, and going to the bathroom doesn't necessarily mean going to a room that has a bathroom. Breakfast isn't necessarily breaking a fast, and Christmas isn't necessarily, doesn't necessarily refer to Christ's mass.
Now, having said that, I think you now understand what I mean when I said that the meaning of a familiar phrase doesn't necessarily correspond to the meanings of its parts. Hopefully that now makes some sense to you now that I've given you the examples. Without the examples, you could nod and say, oh, that sounds plausible, but you really wouldn't have understood it.
David:
Let's roll with that. If adults commit adultery, do infants commit infantry? If olive oil is made from olives, what do they make baby oil from?
If a vegetarian eats vegetables, what does a humanitarian consume? A writer is someone who writes, and a stinger is something that stings. But fingers don't fing, grocers don't gross, hammers don't ham, humdingers don't humding, ushers don't ush, and haberdashers don't haberdash.
What's going on there? Why'd you pull this out?
Steven:
So this is a quote from Richard Lederer. I deserve no credit for that witticism.
David:
I love this. I think it's so good.
Steven:
He's written a number of delightful books like *Crazy English* and *Anguished English*. He has a marvelous ear and a fabulous collector of quirks, errors, oddities, and blunders in the English language.
That whimsical list is quite profound because it indicates a lot of truths about language, such as that over time, compositions of words—compounds or words with a prefix and suffix—can sometimes drift away from the original meaning. In the case of adultery, for example, it is related to adulterate.
To commit adultery is to introduce a foreign substance—namely semen—into a woman where it doesn't belong. But that whole connection has been completely lost. Also, in many kinds of compounds, there are a number of different semantic relationships that can hold. So, for example, olive oil is oil made out of olives. Baby oil is oil for babies. There isn't a single logical relationship between the different parts of the compound.
Language is so ancient. English itself is, depending on how you count, maybe 1500 years old or more. But it came from Germanic, which came from Indo-European, and who knows where that came from. A lot of words can completely obscure their origin. You have to go to a dictionary to find them out, but they're kind of like fossils of processes in the language that are long dead.
How the "er" got into finger, I would have to look up the etymology to tell you. But it's not the same as, say, sing-singer, which comes from the rule that "er" turns a verb into a noun for the kind of person that typically engages in that activity. That's a rule that's still alive and well in the English language, whereas we have the fossil record of lots of rules that died long ago.
David:
When you're writing, if I were to map out almost like a gradient of our conversation, I'd say curse of knowledge. I'm being very haphazard here, but I'd say equal parts left brain, right brain. And then we sort of moved into a left brain side of the conversation.
I think of what you've said about language should be a source of pleasure. I think of that in beauty. You talk a lot about beauty also in your photography; you're very interested in that. That to me is more right brain.
When you're writing, when you're actually sitting down at the keyboard, what are some of the more right-brained, maybe less verbal concepts that drive your writing?
Steven:
Well, certainly visual imagery—can I form an image? Can my reader form an image? Euphony, that is sound—is there some poetry in the prose? I read aloud or at least mumble to myself my prose. That's again a highly recommended writing tip. I didn't invent it.
Often when you read a draft of your own prose, if you can't articulate it smoothly, probably your reader won't be mentally sounding it out smoothly either. The aesthetics sometimes come from even things like paying attention to the metrical structure of language, that is the rhythm.
David:
Oh, tell me about that.
Steven:
There is a regular rhythm to language in that it's not perfectly tick-tock, tick-tock like a metronome, but there are beats. If you disrupt it too much, then it does interfere with speech, but it also interferes with reading, even though it's just characters on a page.
Even the aesthetics of sibilant sounds—generally, too many "ch's" and "s's"—make prose a little unpleasant, and I will often pick a synonym that avoids the sibilant, at least too many sibilants in a row. I'll often go with alliteration just because there's a little spark of pleasure.
David:
Sense of style.
Steven:
Sense of style, yep. You don't want to make it too conspicuous; otherwise, it starts to feel forced. But often a bit of alliteration can just make the sentence roll past more easily.
David:
Why do you think that you, as much as anybody, have been the person to kind of stomp your feet, kick and scream about the how bad academic writing has been over the years? Like, when I think of who has been like, guys, what are we doing? You've sort of been one of the frontrunners. What about it? I mean, besides the obvious, what about it has just bothered you so, so vividly?
Steven:
Part of it is just the sheer waste. There's an awful lot of really brilliant work, a lot of really smart people in academia. Why are they doing it just to entertain each other in a closed little circle? I mean, taxpayers pay for it. It should be accessible. They should give it away. They should state it in a form that it's accessible.
Even within the profession, there's just an enormous amount of wasted effort and potential for misunderstanding in bad prose. It's like, if I have to read something, a student paper, a paper in my own field, I'm reviewing a grant proposal or peer reviewing a manuscript or evaluating someone for tenure, if I have to read the same paragraph five or six times in order to know what they're talking about, I might get it wrong. What's the point?
Why should I have to read it five or six times? I'd rather be doing something else.
There's waste, there's confusion. There's also foregone opportunity for pleasure and beauty. It's enjoyable to read something that's well crafted, and it's annoying to read something that's stilted and turgid and bloated. There are many reasons, and it does get under my skin when academics devote so much brain power into the scholarship and then just blow off the essential task of letting the world know what you've done.
David:
Yeah, it's a true virus in the modern world.
Steven:
I take a particular interest in it, not just because I do try very hard, even in my academic writing, to be clear. Since I'm someone who actually not just studies stuff as an academic, but studies language, I get particularly annoyed at people in the field of linguistics and psycholinguistics because so many of them are such bad writers.
It's like, you study language. If what you're studying is what makes a sentence difficult to understand, why don't you pay attention to your own research and make your sentences less hard to understand?
David:
What is it that you love so much about children's explanations? Things like clouds are water vapor. Smoke is fire vapor.
Steven:
That was from my grandson. I mean, it's poetry. It's a new juxtaposition.
David:
There's such a freshness to children's explanations.
Steven:
Yeah, because they aren't writing in cliches, and partly because they haven't accumulated this mass of abstractions, like the writers of a few centuries ago who had to grasp for some common image. Children, without the decades of acquired jargon from academia, have to appeal to something that they can see and that other people can see.
David:
There's a guy who I follow on Twitter who I've really come to like, and in his bio it says, listen to children. They haven't forgotten how to see. I know a YouTuber, and one of the ways that she comes up with her ideas is she has conversations with her friends' kids who just ask these crazy questions like, how deep is the ocean? How high would a tower need to be in order for us to get to heaven? I have another friend who, when he was a kid, he thought that clouds were produced by those little smoke things, sort of like chimneys. So he thought that those were like cloud creators.
Steven:
Oh.
David:
There's such a freshness to just how children see and how they talk about things. Like you said, they can't possibly think inside the box because they don't even know the box exists.
Steven:
Yes, right. There have been in history various features to remind people of the originality and the freshness of kids. There was a regular feature from my childhood called Kids Say the Darndest Things from a television host named Art Linkletter.
David:
Yeah, I've heard of him.
Steven:
Oh, you have? Okay.
Steven:
In fact, what has survived him and the feature of the program is the "say the darndest things" or even the "verb the darndest nouns," which you'll often see adopted in headlines, and it's become itself a kind of formula, but he originated it. Johnny Carson, the longtime host of the Tonight Show, would sometimes have features where he'd read kids' answers to questions to much hilarity.
David:
How does humor factor into all this?
Steven:
Into writing?
David:
Yeah, exactly.
Maybe even the science of humor, like words that are funnier, because you have that great talk from like 10 years ago. I'm going to recommend it.
Steven:
There's an art to it because you don't want to. When it's strained, you get groans instead of laughter. So, it has to be appropriate.
It can't just be something that has been repeated so often that people have heard it 37 times before.
David:
Humor specifically depends on freshness.
Steven:
Yeah, it does. And it also depends. Humor does have much in common with good writing. There's a line from Hamlet, which is, "Brevity is the soul of wit."
David:
Yeah.
Steven:
Now, it's a great saying for a number of reasons. One of them is even the term "soul," as opposed to "is essential for wit," or "is important for wit," or even "the essence of wit." Those are ways that we would say it now, but "the soul of wit," you have to think for a minute, soul being the deepest essence, but it's a monosyllable, and it has so much resonance.
I guess Shakespeare really was a good writer, wasn't he?
David:
Yeah, I've heard of that guy.
Steven:
Yeah. It is true, and it's a good example of itself because it is so brief. It was reiterated centuries later in the famous style manual, The Elements of Style by Strunk and White, where Professor Strunk, in lectures at Cornell that EB White recalled and then turned into a book, would say that the first rule of writing is "omit needless words."
Now, again, it's a beautiful example of itself because there are no needless words in it. He said that sometimes Professor Strunk would be so adamant and so insistent that this was almost the only thing you needed to know about writing, that he would just not know what to do except repeat it. He'd say, "Omit needless words. Omit needless words. Omit needless words."
As with "brevity is the soul of wit," aside from being an example of itself, it is so true that sometimes when I have to compress an article to fit into some fixed space, especially if I'm writing for a newspaper or magazine and say, unlike academia, where you can blather on with no one telling you to shut up, when you have to write for a paper, they'll say, "Sorry, 800 words." And if it's 803 words, the editor will chop off three words, and you don't get to say what they are, and it can often mangle meaning.
So you have to actually learn how to say something in 800 words or however many words it is. But what I find is often just the exercise of squeezing it into that maximum limit just improves the prose, as if by magic.
It's like, "Oh, you know, damn, I have to make this shorter," and I thought I had said it perfectly. Then often when you do it, you find, you know, it really has improved in quality, having gotten rid of those needless words.
Partly it's because of mental effort; namely, every syllable, every word is more cognitive processing by a reader. If you can get the same message in quicker, that's often less work, but it also makes it aesthetically more pleasing. You have to work within the constraints of the English language, the melody, the rhythm, the music of the language. It forces you into often into concrete language instead of woolly idioms and cliches.
David:
"Brevity is a solo wit" is good. It kind of just hit me how good of a line that is.
Steven:
It's such a good line.
David:
I've heard that a thousand times, but it never punched me in the chest like it just did. I gotta ask.
Steven:
Sorry. The reason I brought it up is in humor. And it's a soul of wit. Wit used to mean a lot more than just, you know, ha, ha. It meant trenchantness and appropriateness. But wit in the sense of just raw humor, that's something that comedians know.
You pare down the jokes, you pare down the lines. When you're using humor, the shorter, the pithier, the funnier. If you drag it out, then it ain't funny. You can step on your own punchline. You can telegraph the punchline.
David:
So I went to Rockefeller Center the other day, and there's this called Like a Big Thing of Stone. John D. Rockefeller is talking about his theory of society and the good life. It was just remarkably well written.
And then it's the same thing with the Declaration of Independence. There's just some beautiful lines in that: We hold these truths to be self-evident, that all men are created equal. It's just beautiful writing.
What do you make of how older writing seems harder to read? Thank God that writing's gotten so much clearer and easier to read. But at the same time, it does feel like a kind of beauty and poetry, a kind of reverence that I have for the language itself has been stripped away from the English language at times.
Steven:
You do sometimes feel that, even again, Rockefeller is a good example because he did not go down in history as a pro stylist.
David:
Exactly.
Steven:
Thank you.
David:
Thank you. That was a very important part of that point. This was the CEO. He was an oil magnet.
Steven:
Yeah, right. And likewise, I have a quote from Thomas Edison in Enlightenment Now on how electricity will liberate women. It's just beautifully written.
David:
Beautifully written.
Steven:
This is a guy who spent hours with filaments and light bulbs, but he could write like an angel. Herbert Spencer, Oliver Wendell Holmes Sr., Charles Darwin.
All these people who were not themselves professional writers, they had something else, but they were very good at it.
So it's an interesting question. I think one reason is they were trained themselves on the classics, on great works of literature. Probably because for many of them, there wasn't the telephone, radio, let alone texting, social media. The way that you presented yourself was through your prose. You cultivated it the same way you arrange yourself in a mirror—how other people viewed you.
They were educated on the classics and had many good examples to draw on.
The third reason is one that came up earlier in our conversation, namely, since they didn't have the benefit of decades and decades of abstractions and cliches, and they were exploring virgin territory, conveying new ideas, and they couldn't easily reach for a canned cliche. They had to put new ideas into forms that their readers or listeners could understand. That forced them to draw on visual images and metaphors and vivid expressions and locutions.
There's been a process that's been going on for longer than a century called informalization. You can see it in dress, where men no longer wear hats and women don't wear gloves.
Sometimes you look at a picture of someone on a hike from a century ago, and the men are wearing ties, and the women are wearing long dresses, and they're going hiking. It's inconceivable to us.
The fact that people used to address each other with Mr. and Mrs. and were all on a first-name basis.
David:
I got in trouble in the third grade for calling my teacher by her first name. I went to detention for calling my teacher by her first name, and that would not happen now.
Steven:
That's certainly familiar to me from my childhood. It would be unthinkable. I wouldn't have gotten that detention.
David:
Actually, it's been 20 years, and I still feel the shame that I felt in that moment. That's crazy.
Steven:
It just wouldn't have occurred to us. You couldn't have done that in a million years. The fact that taboo language, profanities, are commonly woven into speech, which again, would have been unthinkable a century ago.
But there is a process of greater familiarity, of less hierarchy, partly as a byproduct of democratization, the erosion of traditional hierarchies of class and education. There's a kind of romantic ethos where authenticity, spontaneity are valued, as opposed to putting on airs, thinking carefully before you do anything.
All of these have led to fancy language being seen as more pompous, stilted, distancing, whereas the cultural value has changed toward spontaneity, intimacy, naturalness, authenticity.
So I think many of us, if we had to put the effort into crafting prose the way they may have a century or two centuries ago, we'd feel, "Oh, people are thinking I'm being too fancy." You might even perceive you as being too fancy. And as a result, the conversational vernacular, as opposed to the elevated, the well-crafted, has come to characterize modern prose.
David:
So I gotta ask, you were talking about being able to pull a cliche or an abstraction. AI takes that completely to the next level.
And do you feel like AI then enhances our almost inability to think originally, for lack of a better word? Like, does it keep the trend that we're seeing? Or are LLMs like a new form and then maybe actually create some new way of communicating altogether?
Steven:
Yeah, the output of LLMs is peculiar in one sense. It's well written in the sense that it tends not to be in academies, in jargon, the sentence structure tends to be pretty plain and sound. Even the progression of ideas tends to be orderly. There tends to be an introductory sentence and concluding sentence. So in that sense, it's good writing.
It's bad in the sense that it is so generic and prosaic, and you can almost recognize the output of a large language model. It's so banal now, perhaps it can be trained, and perhaps if you prompted it, "Don't be prosaic, don't be plain," it'll be interesting to see whether it would come up with any kind of style or freshness, but it's not the way it's designed. It's designed as a mashup, as a pastiche of literally billions of examples out there.
And it's an interesting question why it should be so, at least why its prose style should be sound. I mean, I think that's the best you could say about it. Whereas the prose style of most academics, most lawyers, most bureaucrats is not sound, why is it better?
One possibility is it's just been hammered into shape through the fine tuning and the feedback, that is instead of just regurgitating an amalgam of the reinforcement learning.
The other is this is completely speculative, but we know in visual beauty, often a composite is more attractive than the elements that went into the composite. So that if you take a bunch of faces and you morph them together, the non existent human being that comes out of the morphing is more handsome or prettier than all of the men's faces and women's faces that went into it. Really take a high school yearbook and you mash together the several hundred faces and the result is pretty attractive.
David:
Is that right?
Steven:
Now whether that can be true of prose style, that is that if you were to eliminate all of the god awful convoluted constructions and just came up with kind of the generic sentence structure, it wouldn't be beautiful, but it would be clear. Hmm, that's a hypothesis.
David:
You've written nine books, you're working on your 10th now with LLMs and the way they're going. If you were to almost rewrite those books starting now, how would you have written those books differently?
Steven:
Knowing what we know about.
David:
Knowing what we know about LLMs, and also your sense of where the world is going and how it's changing—how the world of writing is changing—would you have written the same books?
Steven:
Probably not. I think I'd have to give greater weight to the power of abstracting patterns from massive amounts of input.
In the approach that I was trained in and that I then developed of computational cognitive science, Chomsky and linguistics, classic AI, it was much more organized around rules, algorithms, logic, kind of logical programming. It was hard to imagine how with enough sheer input and training, an associative neural network could extract sensible ideas and prose out of this huge unstructured mass of input.
Now, I don't think it shows that the human mind is a large language model because it would be the equivalent of a child listening to language for 30,000 years before they could put up their first sentence. Also, kids don't just need to have massive amounts of text pour into them. They also are situated in a world where they can figure out what the people talking to them are trying to say, and they're interacting with the world. That's a very different style of learning than just processing massive amounts of text.
Still, I think I certainly would have had to reconcile the intelligence of large language models with human intelligence, with more attention to the power of pattern extraction from large input corpora than I did.
David:
Well, thanks very much for doing this.
Steven:
Thank you. It's been a pleasure.
Share this post