3 ms·

Personally I would draw the line where the tools I'm ultimately relying on to create art don't function at all without having millions of works by other artists

Personally I would draw the line where the tools I'm ultimately relying on to create art don't function at all without having millions of works by other artists baked into them. You can say that's how my brain is doing it anyway—and sure, you might be right to a first approximation—but isn't that the whole point? If I'm not even capable of executing that synthesis in some medium X myself, but rather have to outsource it to a computer program, then I'm more technician than artist. It's something I don't expect people who have never seriously engaged with an artistic process to understand. Art is interesting when the artist discovered things in creating it, and you are not going to discover much by typing in a few sentences of prompt and maybe touching up the image (or story, or whatever) the model makes for you. "Cheating" is a loaded and generally unsuitable word based on an infantile understanding of why art matters. The universe doesn't give a shit whether you do it yourself or get a machine to do it for you, and neither do I, really. Modulo speculative impacts on economy and culture, these are obviously useful tools and I don't have any moral issue with people using them. But I will say that if those people are aiming to create art rather than engage in a process of design, then by using these tools in their primary creative workflows they are depriving themselves of a priceless journey. That's why it matters. It's a bit like sitting in your car at a trailhead and flying a drone out to the end of the trail and back rather than doing the walking yourself. Not only are you potentially ruining the experience for others (see: the recent Clarke's World closure to submissions), you're vastly cheapening it for yourself too. Most people in today's discourse are focusing on the former, but I think the latter is more interesting and important.

YeGoblynQueenne 4y ago
Great point, but then, I went for an art degree before I turned to computer science, so I know exactly what you mean. The act of artistic creation itself is rewarding, in a way very different to looking at art created by someone else. As a counterpoint, though, I note that I've done some work on language generation and that, too, is something that can engage the user, and can be almost as rewarding as e.g. writing a story. Though that's much more so if the language generator is your own piece of work. I'll also add that we appreciate artists who develop their own, unique personal style much more than artists who copy those of others. So far, neural net image generators have not been shown capable of creating their own styles, unless of course one considers artifacts of their generation process (e.g. monster hands) as a "style". People severely overestimate the "creativity" of such systems.
- caconym_ 4y ago
  > As a counterpoint, though, I note that I've done some work on language generation and that, too, is something that can engage the user, and can be almost as rewarding as e.g. writing a story. Though that's much more so if the language generator is your own piece of work. Writing fiction is my thing, and I've seriously considered using AI to generate descriptive passages and other relative "filler" as a force multiplier. Either way, I'd be interested to see how well I could get a model to imitate my style. I think I'm waiting until it's easy to run a GPT-3-class thing locally, which IIUC isn't currently the case. I don't like building workflows that rely on "cloud" stuff. "Force multiplier" is the key, though. Another example might be somebody writing a story and then using AI to generate illustrations for it. I can also imagine future image generation AI allowing much finer control over the results, which would (IMO) make it much more akin to photography (and legitimize some of those arguments that have been popping up everywhere, including in this thread). Letting the computer provide the vision and direction as well as the technique is fun, for sure. I had a lot of fun with Stable Diffusion. But it's as you say—the current state of the art is highly imitative, and I think these capabilities will ultimately frustrate serious creators of art and serious consumers of it.

YeGoblynQueenne 4y ago

Great point, but then, I went for an art degree before I turned to computer science, so I know exactly what you mean. The act of artistic creation itself is rewarding, in a way very different to looking at art created by someone else. As a counterpoint, though, I note that I've done some work on language generation and that, too, is something that can engage the user, and can be almost as rewarding as e.g. writing a story. Though that's much more so if the language generator is your own piece of work. I'll also add that we appreciate artists who develop their own, unique personal style much more than artists who copy those of others. So far, neural net image generators have not been shown capable of creating their own styles, unless of course one considers artifacts of their generation process (e.g. monster hands) as a "style". People severely overestimate the "creativity" of such systems.

caconym_ 4y ago
> As a counterpoint, though, I note that I've done some work on language generation and that, too, is something that can engage the user, and can be almost as rewarding as e.g. writing a story. Though that's much more so if the language generator is your own piece of work. Writing fiction is my thing, and I've seriously considered using AI to generate descriptive passages and other relative "filler" as a force multiplier. Either way, I'd be interested to see how well I could get a model to imitate my style. I think I'm waiting until it's easy to run a GPT-3-class thing locally, which IIUC isn't currently the case. I don't like building workflows that rely on "cloud" stuff. "Force multiplier" is the key, though. Another example might be somebody writing a story and then using AI to generate illustrations for it. I can also imagine future image generation AI allowing much finer control over the results, which would (IMO) make it much more akin to photography (and legitimize some of those arguments that have been popping up everywhere, including in this thread). Letting the computer provide the vision and direction as well as the technique is fun, for sure. I had a lot of fun with Stable Diffusion. But it's as you say—the current state of the art is highly imitative, and I think these capabilities will ultimately frustrate serious creators of art and serious consumers of it.