Was I an AI all along?

Over the weekend, we started a cult. It was one of those things that happen when you decide to stop drinking for a few weeks. It’s an avocado-oriented octopus cult called Octo-guacamolians. But that’s not important right now. What is important is that we fed the parameters of the cult into GPT-4, and it came up with a number of remarkably coherent and reasonable rites, rituals and, well, all other aspects of cultiness. Later the same weekend, I was reflecting on what I’ve done for much of my life and started to wonder whether I was a proto-AI for a while.

A publisher saw the post and asked if I wanted to write a book about macro photography. I said yes, then had to learn how actually to do macro photography.

You see, I found myself in an interesting situation for a while there. I wrote a stack of books about photography, starting with one on macro photography, followed by a bunch of others. Portraiture, travel photography, etc. In the weird times before we settled on “mirrorless” for the current slew of image-capturing devices, there was a hot moment when nobody really knew what to call them, so we were referring to them as Compact System Cameras (CoSyCa) and so on — and I brilliantly thought I would name the entire genre EVIL — electronic viewfinder, interchangeable lens. The publisher agreed, and we even ended up publishing a book called EVIL photography. Those were the days of experimentation, a brave, brand new world where nobody really knew what was what, where was where, and whether anything really mattered anyway.

The story of my macro photography book was particularly interesting; back in 2005, I built a macro lens out of a Pringles can (really, it was just a convenient tube to reverse a 50mm lens, but… it worked!), which ended up on slashdot, digg and all of those pre-Reddit-era link aggregation sites. In the process, I racked up a ton of traffic and got to learn the hard way why we use AWS and such these days — back in the mid-naughties, we ran our own servers, which meant that my host was very grumpy that there was suddenly a ton of traffic to this one particular page, and threatened to unplug the server because it was taking up too many resources. Ah, the heady days of the early internet.

In any case, a publisher saw the post, and asked if I wanted to write a book about macro photography. I said yes, then had to learn how to actually do macro photography. The great thing about book publishing is that it takes a long time. So I bought a new camera, wrote a letter to Canon saying that I was writing a book about macro photography, asking if they wouldn’t pretty please send me a few macro lenses, and went to my local bookstore and bought a bunch of books and magazines about macro. I spent the next four months looking at porn (that was my day job — long story) and taking macro photos, teaching myself and taking notes as I went along. In the end, I distilled all my notes and learnings into my first book.

All the subsequent books were, essentially, the same thing: I learned everything I could about a subject, distilled it and wrote it up, illustrating it with photos I took along the way. Are they the best macro/travel/portrait photos you’ve ever seen? Probably not. But my publishers would tell me that was okay; I had a rare gift of being able to learn the thing, do the thing, then teach the thing, and string sentences together that mostly made sense in a way that book-buying audiences seemed to enjoy.

In addition to writing my own books, I’ve ghost-written a bunch of books as well (most of which I can’t say anything about because NDAs and contracts, ssssh), but they were the same thing but in a different way. I would soak up as much information as I can, then distill it into a coherent form, matching my style as far as possible to something that the other person would have plausibly written. I’ve written books about venture capital, photography and some far more esoteric topics.

When I started learning more about how the AIs work — how they can emulate the forms of other artists and even the writing styles of other writers — I found myself having an existential crisis. This thing I had been doing for 20+ years could now be done by computers.

As a university student, I had a number of internet friends who were word-playing dilletantes, crafting the most obtuse prose we could, in the styles of Franz Kafka, Werner Herzog and the OMG! EXCITING! KAWAII! styles that were meandering across the internet. It was fun and educational, as a writer, to get a sense of how much flex there is, stylistically and creatively, in the writing of words.

I have to admit, I was scared to ask GPT-4 to try out some of the writing prompts I remember from back in the day.

I asked it to “Write 100 words about the joy of kittens in the style of Werner Herzog”:

The joy of kittens is a fleeting, visceral experience, a mere glimpse into the chaos of existence. As they frolic, we are reminded of our own primal desire to play, to explore the unfathomable cosmos. These tiny creatures embody the fragility of life, their innocence bearing the weight of the relentless march of time. Yet, in their eyes lies an indomitable spirit, a force that defies the entropic nature of the universe. It is in these ephemeral moments that we are granted a reprieve from the crushing indifference of the cosmos, our hearts swelling with warmth, as we are engulfed by the rapture of their presence.

I asked it to “Write 100 words about the joy of kittens in the style of Franz Kafka”:

Within the shadowy corners of existence, one might find solace in the unassuming presence of kittens. Their delicate purrs, woven into the fabric of reality like an intricate tapestry, are an enigmatic paradox of innocence and primal nature. The joy they evoke, a fleeting moment of respite from the existential weight we carry, is an elusive treasure that eludes our grasp like a whisper in the wind. The dance of feline innocence, reflected in their vibrant eyes, a window into the depths of their souls, fills the void of human longing with a warm, ephemeral glow, rendering the unbearable burden of existence momentarily bearable.

And my heart broke a little.

God, this is good.

This unassuming little AI was able to do, seemingly effortlessly, what my compatriots and I had been slaving over for hours. The ramifications of that are terrifying on so many levels. Fake news, fake reviews, fake everything, can now be cranked out on an industrial scale, and all good enough to be just-about-plausible.

It is often said that humans will do jobs until machines can come along and do it well enough for cheaper. That’s true for most of the field of robotics — it is possible (maybe) to build a car by hand, but robots have taken over much of manufacturing, for good reason.

In being asked to write books on various topics, I’m reflecting on how similar my instructions were to the kinds of prompts I give to GPT now. “Hey Haje, could you write a book on travel photography? Same kind of style as your portraiture book, but focused on lighter gear and taking photos on the go?” Well. Yes, sir, I can.

And in the midst of all of this, I’m reminded of the time I received a copy of a book that had my name on the cover but that I didn’t write.

And in the midst of all of this, I’m reminded of the time I received a copy of a book that had my name on the cover but that I didn’t write. My book publisher had re-purposed most of the copy I wrote for a book that flopped, had an editor re-write parts of it and then re-published it with a completely different title and angle. Somehow, that also happened while the publisher was being acquired, or sold, or went through a management change of some sort, which resulted in nobody actually telling me this was happening. The first time I knew this book existed was when a FedEx box with five copies of it turned up at my doorstep, followed (a few days later) by (ironically) my advance check. Legally, they were in the clear — I had signed away the rights to my book, and they were able to turn it into cartoons or bobbleheads or movies or dance performances. At least, that’s how I read “any media,” — but it turns out that “any media” could also mean “another book.” Silly me.

In today’s world of generative AI, all of the above doesn’t sound completely outrageous. But I’d be lying if I said I didn’t have a huge, Texas-sized helping of déjà vu as I’m watching AI unfold in front of our eyes.

What I’m left with is the question of whether humans are still better than our robot overlords. And that’s the thing that truly scares me. A generative AI with full access to the internet can out-research, out-write and out-post any human. It can emulate any writing style, and can even throw typos into the mix to throw you off its scent, should it decide it needs to. There are still some supposed tell-tale signs of AI-generated text, but as a writer, I have to admit I don’t know what I’m looking for, and I’m as likely to be fooled as anybody else at this point.

The truly scary thing is how we got to GPT-4 so quickly, and that there’s a lot of numbers left before we hit infinity. In other words: The shortcomings of GPT-4 will be shredded by whatever comes next, and as a writer, I feel like I have every reason to worry.

Oh, that cult we formed? We co-created it over a delicious meal of pallet wine (yes, really — wine bought by the pallet, which I did not partake in, as per my earlier comment about not drinking) and Mexican food. I introduce you to The Octo-guacamolians: A group dedicated to the worship of an eight-armed avocado god named Guacamoctopus. They hold weekly rituals that involve the creation of intricate guacamole mandalas and underwater dance performances.

I mean. If it didn’t exist before, it exists now.

Was I an AI all along? by Haje Jan Kamps originally published on TechCrunch

Source link