The comments here made me think of this wonderful passage from Wendell Berry that my pastor recently cited, where Berry talks about why he resisted using a computer to write his books.
“ a computer, I am told, offers a kind of help that you can’t get from other humans; a computer will help you to write faster, easier, and more. Do I, then, want to write faster, easier, and more? No. My standards are not speed, ease, and quantity. I have already left behind too much evidence that, writing with a pencil, I have written too fast, too easily, and too much. I would like to be a better writer, and for that I need help from other humans, not a machine.”
That is a beautiful quote. Our culture has come to worship speed, ease, and quantity. I esteem people who treasure Berry's priorities. I was talking about this with my husband, and he said, "The one thing that AI writing can never have in its algorithm is the spirit of God....when God's spirit inspires a writer with ideas and direction." That was a comfort to me.
It's funny because after reading this, while I noticed many of the same things as my beautiful wife, Heather, it actually led me to the opposite belief -- that the 1st version was AI and the second wasn't.
As a computer programmer and software engineering manager, I've spent a lot of time working with various AI models, which I think has given me a lot of insight into the "tells" of most LLMs.
The most obvious thing I would look for is moralizing to put a bow on a story. Most LLMs are trained to want to find a lesson. I suspect it's because many of them are trained on articles and blog posts on the internet that have a tendency to end on an uplifting note.
AI also doesn't tend to be comfortable with unfinished thoughts or open ended imagery. For example, why did his secretary insist there were no first class options? Why couldn't he have coffee before take off? Why was he annoyed by a woman pulling a scarf over her head? I don't think AI would leave any of that nuance.
I envy your background! You've hit the nail on so many AI "tells," many of which I hadn't consciously realized before—the wrap-it-up-ribbon and the aversion to open-endedness. I'm not going to reveal which is which just yet—I have too many thoughts, and I think I may have to put them into another article.
Its hard to say at this point. In my field, I've definitely noticed that junior developers tend to not understand how their code works because they didn't have to struggle with it like they used to.
I have seen it give us some productivity boosts, mostly around things that don't require a ton of context, which AI tends to struggle with.
I also have concerns that as AI is creating more and more content, it will delute its training data until it sort of becomes a copy of a copy of a copy, which will make its results less intelligent over time.
I appreciate your nuanced take on this issue. It’s one I want to pretend doesn’t exist because it frightens me, but I know that my denial can’t last forever.
I agree that human creation is still valuable, even crucial. What concerns me is the potential for being tricked into thinking something is human-created when it’s not. I may choose to listen to the radio over going to a live concert, or purchase food or clothing made in factories and not by human hands, but in those instances I’m aware of the substitution for “the real thing.” With AI, we might assume we are connecting with others on the level of their art and have no idea that we are interacting with technology—I feel that dehumanizes both the potential creators and the consumers.
I spent far too long trying to discern which piece was written by you, I kept second-guessing my decision! Ultimately I went with the second choice as the one you’d written because it felt less choppy but it was a full-on guess and not a confident assessment. I liked both and didn’t LIKE that I liked both and that I could be so easily fooled.
I cannot agree more on this: "With AI, we might assume we are connecting with others on the level of their art and have no idea that we are interacting with technology—I feel that dehumanizes both the potential creators and the consumers." It makes me wonder how much AI content I've mistaken for human-made.
The first sketch was more image-driven than the second—which, in my mind, indicated human skill. The first conveyed Gerald’s state of mind through images (“Gerald clenched his jaw,” “He exhaled sharply,” etc.)—showing rather than telling. The second sketch seemed to tell rather than show (“Rich hated how small and scratchy the seats were in coach”).
I also noticed that your AI prompt didn’t ask that the protagonist learn a lesson at the end of his experience. As such, it seemed unlikely to me that AI would go to the extra step of creating Gerald’s moment of self-reflection at the end of the first sketch.
I could be wrong about my hunches, of course. Will you share which sketch was yours later?
Perceptive as always, Heather! I love how you and your husband came to different conclusions. When I saw how close the AI output was to mine, I knew I had to share it even though, honestly, it stung. BUT, mine was a first draft—literally rolled off my fingers with no revising or polishing. And you know better than most what first drafts are like... So, if an AI's first draft is comparable to an amateur author's, what are the implications? Or is that the wrong question? 😂
I was surprised how challenging this game was! I can’t tell for sure, but I voted for the second one. Two things about the first one made me question its human origin: those few short paragraphs and the little “lesson” seemed a bit forced.
YUP! Me too. It was humbling to see that AI can produce a similar first draft to a human, but then again, I'm not a pro by any means. I didn't feel comfortable using another author's work, so I offered up mine on the altar. Haha!
I didn't vote b/c I couldn't tell the difference based on the skill of the two pieces (face palm). But I LIKED the first piece BETTER because it had a moral to it. The character was learning something profound, and that was important to me.
It’s crazy how close it is but at the same time not nearly close enough to the real thing. I’m worried that 5-10 years from now it will be almost impossible to tell the difference.
I'm also wondering about the implications! Like Kendra said in her comment, nobody likes to be duped into thinking they are connecting with human-made art only to realize they are connecting with a robot.
What felt human to me in the first one: referring to the guy as Cargo Shorts, making the descriptor a name. If AI is doing that, then I'm impressed. The way the smell of tuna came back at the end. I wouldn't expect AI to have a detail recur. "Loudly." used as a complete sentence. The mix of very longer and very short paragraphs. Putting "Gerald clenched his jaw." as its own paragraph feels like an interesting creative choice. The second had paragraphs of fairly standard lengths, none shorter than two sentences, and those are fairly log sentences at that.
But I have spent almost zero time looking at AI writing, so I might be off in what I'm looking for. I really want to know if I'm right.
But we’re in a brave new world. The eventual consequences of AI are exponentially greater than the internet and may even rival the Industrial Revolution, the last place where human history shifted forever.
I find it telling that no author I know of has been able to imagine a hopeful or good future from this AI revolution, most fictional projections revile or embrace a handful of transhumanist nightmare scenarios… but only true believers in transhumanism believe that any of these scenarios are good.
I have a question for humanity, if we can’t even imagine a good future then why are we embracing this?
The only good response I’ve heard is that it’s an AI arms race and we have to beat other totalitarian countries (China) or they will use the technology for dominance.
But that just makes AI the newest nuclear weapon… an inherently dangerous technology that must be embraced to avoid an even worse scenario.
*
Oh and I chose number 2, because of greater use of metaphor which AI tends to avoid but certainly can imitate… honestly it was harder than anticipated. AI is getting really good
That is wild as it is hard to tell which one AI authored! I knew it was getting better, but I didn't realize it was able to write fiction. I mostly use it to summarize long research articles or if I'm looking for quotes, but the latter I have to verify because it's not always accurate.
I'm not AI-savvy, but I've heard from many folks who don't need to hire content writers for blog posts, social media copy, and even sales copy because they can get similar or better results with AI. Many are using it to write novels, songs, poems...
Yup, we writers aren’t going to be totally eliminated but the art form will be forever changed, as painting was changed forever by photography but even more so
Tragically I’d guess that AI will outcompete humans in fiction and poetry but not in non-fiction.
So just as still lifes and portraiture mostly ceased with photography, so too in a generation I imagine most human fiction and poetry will be a boutique industry…
The comments here made me think of this wonderful passage from Wendell Berry that my pastor recently cited, where Berry talks about why he resisted using a computer to write his books.
“ a computer, I am told, offers a kind of help that you can’t get from other humans; a computer will help you to write faster, easier, and more. Do I, then, want to write faster, easier, and more? No. My standards are not speed, ease, and quantity. I have already left behind too much evidence that, writing with a pencil, I have written too fast, too easily, and too much. I would like to be a better writer, and for that I need help from other humans, not a machine.”
That is a beautiful quote. Our culture has come to worship speed, ease, and quantity. I esteem people who treasure Berry's priorities. I was talking about this with my husband, and he said, "The one thing that AI writing can never have in its algorithm is the spirit of God....when God's spirit inspires a writer with ideas and direction." That was a comfort to me.
It's funny because after reading this, while I noticed many of the same things as my beautiful wife, Heather, it actually led me to the opposite belief -- that the 1st version was AI and the second wasn't.
As a computer programmer and software engineering manager, I've spent a lot of time working with various AI models, which I think has given me a lot of insight into the "tells" of most LLMs.
The most obvious thing I would look for is moralizing to put a bow on a story. Most LLMs are trained to want to find a lesson. I suspect it's because many of them are trained on articles and blog posts on the internet that have a tendency to end on an uplifting note.
AI also doesn't tend to be comfortable with unfinished thoughts or open ended imagery. For example, why did his secretary insist there were no first class options? Why couldn't he have coffee before take off? Why was he annoyed by a woman pulling a scarf over her head? I don't think AI would leave any of that nuance.
I envy your background! You've hit the nail on so many AI "tells," many of which I hadn't consciously realized before—the wrap-it-up-ribbon and the aversion to open-endedness. I'm not going to reveal which is which just yet—I have too many thoughts, and I think I may have to put them into another article.
Tyson, what’s your projection about the impact of AI for the next generation?
Its hard to say at this point. In my field, I've definitely noticed that junior developers tend to not understand how their code works because they didn't have to struggle with it like they used to.
I have seen it give us some productivity boosts, mostly around things that don't require a ton of context, which AI tends to struggle with.
I also have concerns that as AI is creating more and more content, it will delute its training data until it sort of becomes a copy of a copy of a copy, which will make its results less intelligent over time.
I appreciate your nuanced take on this issue. It’s one I want to pretend doesn’t exist because it frightens me, but I know that my denial can’t last forever.
I agree that human creation is still valuable, even crucial. What concerns me is the potential for being tricked into thinking something is human-created when it’s not. I may choose to listen to the radio over going to a live concert, or purchase food or clothing made in factories and not by human hands, but in those instances I’m aware of the substitution for “the real thing.” With AI, we might assume we are connecting with others on the level of their art and have no idea that we are interacting with technology—I feel that dehumanizes both the potential creators and the consumers.
I spent far too long trying to discern which piece was written by you, I kept second-guessing my decision! Ultimately I went with the second choice as the one you’d written because it felt less choppy but it was a full-on guess and not a confident assessment. I liked both and didn’t LIKE that I liked both and that I could be so easily fooled.
I cannot agree more on this: "With AI, we might assume we are connecting with others on the level of their art and have no idea that we are interacting with technology—I feel that dehumanizes both the potential creators and the consumers." It makes me wonder how much AI content I've mistaken for human-made.
The first sketch was more image-driven than the second—which, in my mind, indicated human skill. The first conveyed Gerald’s state of mind through images (“Gerald clenched his jaw,” “He exhaled sharply,” etc.)—showing rather than telling. The second sketch seemed to tell rather than show (“Rich hated how small and scratchy the seats were in coach”).
I also noticed that your AI prompt didn’t ask that the protagonist learn a lesson at the end of his experience. As such, it seemed unlikely to me that AI would go to the extra step of creating Gerald’s moment of self-reflection at the end of the first sketch.
I could be wrong about my hunches, of course. Will you share which sketch was yours later?
Perceptive as always, Heather! I love how you and your husband came to different conclusions. When I saw how close the AI output was to mine, I knew I had to share it even though, honestly, it stung. BUT, mine was a first draft—literally rolled off my fingers with no revising or polishing. And you know better than most what first drafts are like... So, if an AI's first draft is comparable to an amateur author's, what are the implications? Or is that the wrong question? 😂
I almost don’t want to know…
I was surprised how challenging this game was! I can’t tell for sure, but I voted for the second one. Two things about the first one made me question its human origin: those few short paragraphs and the little “lesson” seemed a bit forced.
YUP! Me too. It was humbling to see that AI can produce a similar first draft to a human, but then again, I'm not a pro by any means. I didn't feel comfortable using another author's work, so I offered up mine on the altar. Haha!
I didn't vote b/c I couldn't tell the difference based on the skill of the two pieces (face palm). But I LIKED the first piece BETTER because it had a moral to it. The character was learning something profound, and that was important to me.
Thank you for reading along and for your thoughts! This has been such a thought-provoking exercise.
It’s crazy how close it is but at the same time not nearly close enough to the real thing. I’m worried that 5-10 years from now it will be almost impossible to tell the difference.
I'm also wondering about the implications! Like Kendra said in her comment, nobody likes to be duped into thinking they are connecting with human-made art only to realize they are connecting with a robot.
What felt human to me in the first one: referring to the guy as Cargo Shorts, making the descriptor a name. If AI is doing that, then I'm impressed. The way the smell of tuna came back at the end. I wouldn't expect AI to have a detail recur. "Loudly." used as a complete sentence. The mix of very longer and very short paragraphs. Putting "Gerald clenched his jaw." as its own paragraph feels like an interesting creative choice. The second had paragraphs of fairly standard lengths, none shorter than two sentences, and those are fairly log sentences at that.
But I have spent almost zero time looking at AI writing, so I might be off in what I'm looking for. I really want to know if I'm right.
Thank you for reading, Melanie! I have loved this discussion, and I think I may want to reveal the answer in a future article.
That’s a fun game…
But we’re in a brave new world. The eventual consequences of AI are exponentially greater than the internet and may even rival the Industrial Revolution, the last place where human history shifted forever.
I find it telling that no author I know of has been able to imagine a hopeful or good future from this AI revolution, most fictional projections revile or embrace a handful of transhumanist nightmare scenarios… but only true believers in transhumanism believe that any of these scenarios are good.
I have a question for humanity, if we can’t even imagine a good future then why are we embracing this?
The only good response I’ve heard is that it’s an AI arms race and we have to beat other totalitarian countries (China) or they will use the technology for dominance.
But that just makes AI the newest nuclear weapon… an inherently dangerous technology that must be embraced to avoid an even worse scenario.
*
Oh and I chose number 2, because of greater use of metaphor which AI tends to avoid but certainly can imitate… honestly it was harder than anticipated. AI is getting really good
I agree with you on so much of this. It's moving so quickly, and we don't have time to THINK. And that cannot be good.
Honestly, theologically I just can’t avoid how often transhumanism embraces the concept of “Ye Shall be as Gods!”
In fact, much of its literature uses that idea frequently and unironically.
I voted for the first one because it had more details, which implied that there's a human behind it. But maybe I'm totally wrong ;-)
Thank you so much for reading and joining in! This certainly was a thought-provoking exercise for me.
That is wild as it is hard to tell which one AI authored! I knew it was getting better, but I didn't realize it was able to write fiction. I mostly use it to summarize long research articles or if I'm looking for quotes, but the latter I have to verify because it's not always accurate.
I'm not AI-savvy, but I've heard from many folks who don't need to hire content writers for blog posts, social media copy, and even sales copy because they can get similar or better results with AI. Many are using it to write novels, songs, poems...
Yup, we writers aren’t going to be totally eliminated but the art form will be forever changed, as painting was changed forever by photography but even more so
Tragically I’d guess that AI will outcompete humans in fiction and poetry but not in non-fiction.
So just as still lifes and portraiture mostly ceased with photography, so too in a generation I imagine most human fiction and poetry will be a boutique industry…