Comparing LLMs like ChatGPT to an auto-complete algorithm is like comparing an F-16 to a Ford Pinto. Far different tools, for different purposes. ChatGPT can enhance our communication, allowing us to communicate our unique life experience more effectively, and get more done, even enabling people who are less-abled to do things they couldn't before. But like any great technology, it can also be used for harm. How we choose to use it seems to be a more important question than whether artists can use it as another paintbrush or typewriter in their toolbox.

Likewise, the Hallmark card is still meaningful when given to another person, even though you didn't write it, and even though many others are receiving the very same card, with the very same words.

Or, it allows us to get a lot more done in the same amount of time, making us far more efficient producers, creators, writers, etc. Depends on how you look at it. In coding spaces it has been recently estimated that programmers are at least twice as efficient at coding with an AI assistant as they otherwise would be without one. Many technologies help us to do things better, more quickly, more efficiently, with better performance, and it seems generative AI is no exception.

Again, we grossly underestimate exponential growth and development of technology. Hardly anyone saw ChatGPT coming until it was basically here. This underestimation of exponential growth will likely catch many people, companies, and institutions off guard for many years to come.

Do we cope with unfamiliar situations better? Many people clam up in unfamiliar situations, paralyzed in fear or discomfort.

So can a human. And it might actually be less likely that an autonomous vehicle would crash into it, given the number of cameras and sensors are far greater than any human. Humans may take their two eyes off the road for a split second, and boom.

Depends on what you mean by "efficient." A computer can play a game thousands of times in a second, whereas it could take a human hours to play the same game one time. It took AlphaZero only NINE hours to play those forty-four million games and master the game of chess better than any human (or other computer program) in history. Which is more efficient?

I wouldn't be so sure. We always underestimate exponential development.

Perhaps we should be asking ourselves why we're generating such documents and bulleted lists in the first place.

And using your arms in a warehouse to move pallets of boxes would be likewise the wrong way to approach that job. One isn't there to increase their arm strength, but to move pallets, and the use of the forklift would be appropriate.

If the point of the writing is to develop critical-thinking skills, then it is important to do the writing one's self. But that writing is for the writer, not for the reader. Different purposes.

And yet, isn't that what teachers tell their students to do, to not simply cite them in full, but "put it in your own words"?

Again, giving an example of word for word plagiarism as what AIs do. That's not how they work.

It doesn't do so spontaneously.

Why can't humans use LLMs or image generators to communicate something?

Yes, but ChatGPT isn't talking by itself. There is a human user that is facilitating the conversation. The prompt, if nothing else, encapsulates the intention of a real human.

That is true. It takes a human user to use the LLM as a tool, to use the tool to help them communicate. LLMs are not generating language or writing by themselves, to each other, or to humans. Humans are guiding them, instructing them, using them as a writer might use a typewriter to record their writing on a piece of paper, or a keyboard into a computer, but with many more possibilities attached, many more options to communicate, facilitating that communication. There is human intention behind everything written by a LLM or created with an image generator. Why discount these intentions?

That's quite the accusation. But as explained earlier, AI doesn't copy pixel for pixel, or word for word, from those who came before. It learns associations, so that it can generate novel ones along similar lines. And this is where what constitutes creativity might be addressed. The phrase "there is nothing new under the sun" is a proverb that suggests there is nothing truly novel in existence. Every new idea has some sort of precedent or echo from the past. It is impossible to create something absolutely new or novel, despite of the use of the term to describe book-length fiction. It is always a remixing of ideas in different settings, with different characters, with different plot twists. Language itself is a remixing of letters and words in different ways. So the idea that AI uses what has come before should not be a shocking indictment. All artists do the same thing, whether they are conscious of it or not.

And yet, does the recipient gloss over those words as meaningless? No, they read them, and often appreciate the sentiment expressed. The giver of the card didn't write it themselves, and yet the recipient is still often grateful for it.

If the whole point of a writing it to express one's heartfelt thoughts and emotions to someone, then of course it makes sense for one to write it themself. In that case, it probably does more for the writer than for the recipient. But not all writing needs to be so directly heartfelt, as noted by Jung in the note above that I shared. Someone can be quite detached personally and emotionally from what they are creating, allowing it to almost create itself. They can be personally divested from it, allowing it to stand on its own. Art doesn't need to be an ego trip.

No. When I read something, I'm not thinking about the great effort the writer put into it. That doesn't determine whether I'm interested in the writing or not. I don't think about that at all, really. What matters is the writing itself. Is it good? Does it capture my attention? Is it interesting? Does it make me think? Does it take me on an adventure? Does it inform me? Do I learn something? These are the things that matter to me, and I suspect most people, when they read. Who wrote it, and how, may be entirely beside the point.

AI also affords great expressive potential, even multiplying the efforts of an individual artist many-fold.

Again, Chiang seems woefully ignorant of the state of the art of creating art with AI. There is quite a bit of work that still goes into the medium. It is different work, to be sure, but work nonetheless. The AI doesn't write stories or generate art by itself. Someone pushes the buttons and pulls the knobs and foot pedals, just as on an organ to create music.

I doubt that.

Not all would agree that these countless small-scale choices are a requirement of art. Much of art may be produced unconsciously (as Chiang himself noted). Carl Jung once observed:

"To feel that you are the creator is a terrible burden, hellish anguish, provided of course that you are creator enough to feel it consciously. The creator is usually like a child that just plays with the gods and can produce the most awful monster without seeing it. Many artists can only produce because they don't know what they are producing; the moment they know, the creation is completely stopped. For then they begin to reflect; then they feel responsible and cannot play like the gods, unless they fulfil the psychological demand that they dissociate themselves from the creation, from the archetype, from the creative impulse itself. If they can do that, they can go on creating; then they can allow the god to play." (Jung, Nietzsche's Zarathustra)

Perhaps by not getting caught up in endless details, AI does allow the artist to unleash their creativity, to not get caught up in making such endless small-scale conscious choices. They let the god play and see what happens.

Has the definition of art changed from making a lot of choices to the amount of effort put into it? Tools are called such because they help us accomplish things more efficiently, quicker, easier, with less sweat, and that goes for a paintbrush or a typewriter too. I don't think art is defined by the amount of sweat pored over it, but by how skillfully the artists wields their tools to create an insightful or inspiring piece.

I'm not as familiar with how authors are using AI to help them in their writing, but I doubt that it is by generating a hundred thousand words of prompts. They are likely using it to help them brainstorm, with organization, get ideas for plot twists, for character development, etc. Authors are probably not using these tools to generate all the text, but to help them refine the text, to improve it, to help them organize it, etc. Or the AI might generate a first pass rough draft, and then author goes back over it and edits it, and then puts it back into the AI for further embellishment, and this could go through several such rounds of refinement. Could that user perhaps deserve to be called an "author" by Chiang?

That may be true for a tool such as DALL-E, but it isn't true for professional artists using AI in their workflow. It often takes significant effort to produce what the artist is looking for.

Professional artists using AI as a tool are not using "hacks" on DALL-E to create their art. Their tools have evolved significantly.

The state of the art has moved far beyond DALL-E.

I'm glad that Chiang concedes that even if there were just many text prompts used in the creation of art, that such a creator could still be an "artist" (and the output still "art"?). But that's not the way professional artists are using AI.

Again, this is just ignorance of the state of the art. Much more than text prompts are used in the creation of this art.

Chiang makes the exact same error that people did with photography when it was first developed. He thinks that all there is to making AI art is a short text prompt. This is just ignorance of the art form, the same ignorance he recognizes in photography when it was first developed. There are hundreds, if not thousands, of settings, configurations, options, models, techniques, algorithms, processes, edits, inpainting, outpainting, etc., in the creation of art using AI as a tool in the artist's toolbox. Just as it first seems that a camera doesn't offer many choices, when you compare an amateur's use of AI to a professional's, you also see a difference. The artist is not simply entering a text prompt. That's not to mention the millions or billions of choices the AI is also making behind the scenes (unconsciously). Here is an example of a professional artist at work using AI in the creation of their art: https://youtu.be/K0ldxCh3cnI?si=23LFRJb4DAu7PV-b

Indeed! Artists were up in arms that this new medium was the death knell for traditional art. If anyone could take a photo, then who needed an artist? But, of course, art continued, often in new and interesting ways (beyond realism, into Impressionism and other forms), and photography itself was soon seen as an art form in itself, as Chiang notes.

Perhaps this gets at Chiang's real argument against using an AI tool in the creation of art. The artist can't claim total credit for "every detail." Art is an ego trip? It is only art if the artist can claim credit for every detail? What about other influences? Teachers? Can artists not give credit to their own progenitors that inspired them? Can we not create art on the shoulders of giants who have come before us?

Contrary to popular belief, image generators are not simply borrowing elements from similar paintings found online. They are not merely collage-makers. While they are trained on many millions and billions of images and paintings, they also make decisions when generating new images, and rarely make a simple copy of what they were trained on. Indeed, such photocopying effect is usually frowned on in AI training as overfitting, which degrades the AI. We don't want it to simply regurgitate what it was trained on, pixel for pixel, or word for word.

Again, Chiang seems to not recognize the AI as making an enormous number of decisions here too, far more than the human. Letting the program "do the rest" glosses over a mindboggling number of decisions, choices, calculations, configurations, even tens of thousands of decisions that humans made when they trained the AI.

Yes, you can prompt it to do that, which sometimes makes for very interesting derivative works. The entire Creative Commons is built on the idea of "derivative works," as is fair use more generally, building on the work of those who have come before. Remixing is another example.

Perhaps in Chiang's view. The output of tools like ChatGPT are not often bland (which is why many people use it), and they are not exactly the same each time, even for the exact same prompt (which they would be if it was a simple average). There are many settings, configurations, various algorithms like temperature that help introduce novel and creative choices, unlike anything that the AI has been trained on.

This is an extreme oversimplification of what the AI does when it makes choices. It is not simply an "average" of choices that other writers have made. That would make for very bland text indeed.

But the AI is making quite a few choices, even millions or billions of choices as it processes the prompt (and other settings), they are just not human choices. Given Chiang's definition of art as "something that results from making a lot of choices," this would qualify quite well, much more than a human writer. Even if we narrow the definition to "human choices," Chiang noted that when writing fiction some choices are "unconscious." Are these still human choices? How are unconscious human choices different than choices made by a computer? In neither case is the human actively or consciously involved.

Again, who presses the buttons? The calculators don't work on their own.

This is a common misunderstanding of AI, I think, that it will get "better than us," and so then replace us. No, I think we will use it, humans will collaborate with it, use it as a tool to help us write better fiction, create more and better paintings and movies. It won't create fiction or make paintings or movies on its own. Why would it?

Perhaps depends on who you ask. I have personally created fiction with it that I thought was quite good, and I know of writers who regularly use it to help them brainstorm new ideas and work out storylines.

In this story, the use of the invention multiplies the creative output many-fold, and rather than being uninteresting and dull, contrary to Chiang's appraisal here, is popular. Such flourishing of creation seems to add to the culture, rather than diminish it.

Is it the product of the invention alone? No. The engineer had to create it and operate it.

There's the rub. It still requires a human to operate it. The human must manipulate the handles and foot pedals in order for the machine to operate. It doesn't operate itself. The comparison to driving a car or playing an organ is apt; these are also machines that require human operators to function. And an organ generates music when a human does so, which we all recognize as a form of art.

Alien? All artificial intelligence does it make choices, millions or even billions of choices in a split second, far more than a human can do, and much faster too.