In 2005, Princeton University philosopher Harry G. Frankfurt published a book that holds the key to understanding much of what’s going on in the 21st century. On Bullshit pulls no punches, right from its opening line: “One of the salient features of our culture is that there is so much bullshit.” It goes on: “Everyone knows this,” and acknowledges that we all contribute to the rising tide of bullshit. But what, he asks, exactly is bullshit?
This slim volume, really more of a hardbound essay than an full-blown book, goes on for the next 67 little pages to define the term in a way that’s surprisingly useful for understanding what’s going on right now with generative AI, and also in social media, politics, and my own field, known, with some bullshittiness, as “content.”
Frankfurt makes a crucial, useful distinction between lying and bullshit. The liar, he writes, attempts to “insert a particular falsehood at a specific point in a set or system of beliefs, in order to avoid the consequences of having that point occupied by the truth” (p. 51). This requires focus and craftsmanship. The liar needs to know what they believe to be the truth, and design their falsehood carefully around it.
The bullshit artist, by contrast, is completely untethered to the truth. “His focus is panoramic rather than particular. He does not limit himself to inserting a certain falsehood at a certain point, and thus he is not constrained by the truths surrounding that point or intersecting it. He is prepared, so far as required, to fake the context as well” (p. 52). Sound familiar? This enables creativity that is “expansive and independent, with more spacious opportunities for improvisation, color, and imaginative play.”
Of course, everyone loves a bullshit artist, as long as they aren’t bullshitting us too directly. The audacity! The humor! The colorful use of language and the shameless inventiveness — such freedom and entertaining outrageousness!
This freedom is a direct result of being completely disconnected from any concern for the truth. “For the bullshitter,” Frankfurt writes, “all bets are off: he is neither on the side of the true nor on the side of the false. His eye is not on the facts at all… He does not care whether the things he says describe reality correctly. He just picks them out, or makes them up, to suit his purpose” (p.56).
To me, this sounds like a perfect description of how generative AI works. A chatbot produces text based on patterns of language that have emerged across its recursive statistical analysis of billions of source documents, picking out or making up things that sound coherent, without any reference to the facts at all.
A generative AI model is a predictive text machine that uses neural networks (computer algorithms that use probabilities to simulate neurons). These networks are trained using examples of coherent human language, and through iterative training, the networks evolve and produce outputs that seem increasingly coherent and legible to us. The result is what we call a “large language model,” like GPT-4 or Claude.
This works well for language, which is relatively predictable. If you feed a language model a huge amount of language, like the contents of the entire internet, it can get really, really good at creating text that sounds like a human wrote it.
It’s an impressive trick, and the fact that the results are often useful as well as intelligible is a remarkable testament to the fact that the internet is not completely crap, after all. Actually, it is mostly decent. If you vacuum up billions of web pages, published articles, and books, what you get are text patterns that by and large correspond to somebody’s facts.
Dylan Tweney
But when the text generated by a chatbot misrepresents the truth, it’s not correct to call it a hallucination. The AI is just doing its job, which is imaginative confabulation. Its job is literally hallucination.
In other words, generative AI is a bullshit machine. A bullshit bot, if you will.
Whether what it produces has some truth value or not is absolutely beside the point. The text that’s more or less correct is just as hallucinatory as the text that’s fantastically wrong.
I suspect that no amount of language analysis will get AI engines any closer to the ability to analyze logic and facts and make assessments of the relative truth values of different statements. That will require a different kind of AI. For now, it seems, the chatbots are doing what they do best, which is spewing out huge quantities of plausible-seeming bullshit — and everyone seems quite happy with that.
Applying this insight to the world of politics is left as an exercise for the reader.
Here comes the flood
What about us, though? Will generative AI fully replace the need for writers and editors like us?
If all we are doing is creating “content,” then yes, I’m sorry to say, the AI-generated writing is on the wall.
As the WSJ’s Christopher Mims reported on June 21, freelancers are seeing their gig work disappear and their rates plummet. Writers were the worst hit of the freelance categories studied by Upwork, whose numbers the WSJ published, with a 19% decline in the freelance fees for “low value” writers since November 2022. (“High value” writers saw their rates increase by 1%.) Note that in the same time period, inflation took away another 5.2% of the value of those fees.
At the moment, it seems clear that the companies needing content are finding that GPT and similar large language models are compelling alternatives to human writers. LLMs are fast, they don’t mind working nights or weekends, they don’t fuss about rights, and they are incredibly cheap.
If you have a content-shaped hole you need to fill, you’re going to choose the least expensive filling that meets your requirements. If you’re not too fussy about truthiness or originality, generative AI is the obvious solution. Sorry, freelancers.
The corollary is that we’re about to see a lot more bullshit content.
What if we’re solving the wrong problem?
Generative AI looks like a solution if you are thinking in terms of content. But that’s the wrong way to approach these things. “I need an image for the top of this blog post — what’s the fastest, cheapest, and easiest way I can get something to illustrate my point?” DALL-E or Midjourney to the rescue. “I need 50 blog posts to enhance the SEO of my website — how can I get those up there as fast as possible?” Enter ChatGPT or Claude.
The problem with that approach is that it’s missing the point. It’s just creating bullshit, and the reader isn’t fooled. “Oh, another AI-generated bullshit image,” the reader thinks, as they scroll rapidly past your hero image. “Oh, another bullshit SEO post,” the reader thinks, scrolling past tons of SEO-optimized verbiage just to get to the recipe they actually want.
Dylan Tweney
In fact, there’s growing evidence that AI-generated content actually hurts websites’ search engine rankings. A couple of significant sites that experimented with mass-producing content through AI have actually seen their traffic decline by 30% or even 50%, my former employer reports. Another study found that human-written content delivered 5.44 times more traffic than AI-generated content.
The problem is that creating “content” is not, by itself, a goal worth aiming for.
What we really want, I think, is to touch people somehow — to move them.
Maybe you want somebody out there to read your work, recognize another like-minded human, and send you a comment or an email. Maybe you want to convert some readers into customers by winning their trust. Maybe you want to persuade someone to adopt your point of view through a strong argument.
For these kinds of objectives, what you’re really looking for is human connection — a meeting of minds and hearts.
For that purpose, AI is not close to being the right tool. It’s not even in the same problem space.
An AI could produce the most stunningly convincing impression of a Beyoncé song, but it’s not going to win the adulation of fans because it’s missing the most critical element: Beyoncé. Fans don’t just want to hear a “Beyoncé song,” they want to feel a connection to the artist they love. They want to understand her better. They want to feel understood, even seen. They identify as Beyoncé fans because she gives voice to things they have also experienced, and they reward her for her talent with their love and trust. It’s a personal connection.
When I read something written by another human, I'm not just consuming words, I'm connecting with the writer. If they’re a good writer, I want to read more from them. I want to learn about how they grew up, when they started to write, and what made them so good. I want to hear their views on writing and on life. I form a live, personal connection with them, even if they’ve been dead for a hundred years.
AI cannot replicate that experience because there’s nobody there to connect with. AI-generated writing is dead from the start.
This is also why SEO-oriented writing is so dead: Because it’s written for an audience of search engine robots, not other humans. In that case, the human is missing from the reader’s side of the equation.
Either way, a missing human means dead writing.
(Ironically, the AI powering Grammarly suggested several improvements to the previous paragraphs, rewriting the first two sentences to make them stronger and fixing numerous spelling errors and weaknesses in my draft copy. I’ve still got my hands on the keyboard here, though, and I’m reviewing and editing every suggestion.)
Writing is life
Take humans out of the writing, and you might have some kind of digital networking, but you don’t have a social connection.
I’ve come to believe that the best writers create connections, not content. Connections with readers — connections among ideas — connections between communities.
Good writing is good communication – and good communication makes life better.
This is why writing with others, in series or in parallel, is such a powerful experience. Writing together taps into the social dimension of writing, and writing is social because, at the end of the day, most writers are writing for someone else. The audience is other human beings, making reading and writing deeply social activities.
To connect with other people, bullshit is the opposite of what you need. You don’t need “content.” I am tempted to say that you need authenticity, but that word has been so overused that it, too, has become a kind of bullshit.
What you need is some kind of truth. Write your own truth. Investigate the truth skeptically, journalistically, or experimentally. Write about your company’s widgets or the services you’re offering. But whatever you write, do it with honesty and integrity.
And then share it.
Any kind of shared writing — be it a newsletter, a blog post, or a letter you send to your father — is furthering the sacred mission of connecting us with others. If you want to become a better writer, share your work. Publish it somehow, formally or informally — but get it into the hands of a reader, or readers.
Writing has the potential to knit us together, letting us see and understand another’s point of view. Let’s not cede our storytelling to the bullshit bots.
