I return with a little reluctance to the theme of AI and its possible replacement of human artists and writers.
Let me say, first of all, that I am not completely convinced that AI-generated writing or art has the capacity to really satisfy in the same way that art created by humans does, and that it always seems to lack some sort of je ne sais quoi (okay, let's call a spade a spade - I think it lacks soul). Looking at these images, for instance, I am forced to confess that it is remarkable that an AI simply came up with them starting from a few text prompts. But I still don't like them. They look to my eye like something merely created by a robot aping a human artist, which I suppose, in the end, is what they are.
Let me also say that I'm aware of the argument that goes something like: 'AI won't replace human creators - it will just be a neat tool to supplement human creativity, allowing it to go in new and possibly wonderful directions.'
Let me also say that I'm aware of the argument that goes something like: 'People who are against the development of machine learning technologies are just like blacksmiths being against the development of the car. They are standing athwart human progress and making a lot of fuss about nothing.'
I know about those arguments, and can't really refute them; they may very well be right.
I have, nonetheless, arrived at the position that I would never use AI-generated art or writing in anything I created, and that I think it is important that humanity takes a stand against it. If human beings have anything special about them, anything that separates them from animals, it is the fact that we have innate capacities to create art and to tell stories. Painting a beautiful painting or writing a great novel is not like being a blacksmith; the latter is not really distinct from a beaver building a dam or a chimp using a stick to catch ants, but the former is in a different universe to anything any other species can produce. Undermining that through technological innovation is thus a step not only towards making human creativity obsolete, but towards depriving human beings of what in fact makes us human at all.
That might seem quite a grandiose concern for a part-time creator of niche RPG materials. But since that is what I am, that is where my own personal stand has to take place.
I think we will increasingly see creatives use a Fair Trade-style label to indicate that no AI/ML was used in the creation of work. Like Fair Trade, consumers will be empowered to vote with their wallets.ReplyDelete
That said, I wouldn't want to be a digital artist in 2022+. The economics are what they are, and workflows, production deadlines etc will be much easier to control when the art department is just a guy/gal who knows how to write decent prompts.
For my part, I agree that ML-generated writing isn't close to "there" yet. I disagree about the art though. I've seen ML art that is stimulating (and perhaps even sublime) in a way that human-created art isn't. Knowing a machine generated the art is it's own, new category of fascinating.
Yes, but what we gain in fascination isn't enough in my view to justify the consequences. I also don't like the use of the word "sublime", because it suggests transcendence or spiritual significance, which AI art doesn't by definition have.Delete
Interesting point about the 'Fair Trade' label. That would be a good innovation!
I was using "sublime" in the Romantic sense, of having an emotional response to the natural world, rather than anything religious. Maybe AI art isn't "natural" in the same sense a mountain vista is, but if it isn't human-created, then it's a third category of thing?Delete
It is human-created in the sense that it is part of our 'extended phenotype' - I would put it in that category. Like a train or tool.Delete
I think that human artists are probably safe for the moment, primarily because I have seen some examples of AI art, particularly on DeviantArt, and as your page of examples shows, the AI may capture the texture and lighting, but it has little understanding of the actual forms it is supposed to be depicting. Human figures in AI often have distorted or absent features and limbs. AI seems to have particular problems with arms and hands. Similarly I have seen AI attempts at sailing ships, and although the planking and carving on the sides looks excellent, the masts and sails are twisted and stunted. The AI does not know what the object it is depicting should look like. Until this happens, AI art will not be a serious threat to human art.ReplyDelete
You should see what it does to biplanes!Delete
My concern in this regard is the rapidity of the development. How long have we been playing around with this technology? 5 years? What will happen if we fastforward 10 years from now?Delete
As both an artist and a proponent of AI art, I would say that human artists aren't going to be replaced entirely any time soon (unless they make generic stock art maybe) but that's not because there's some kind of inherent 'soul' in human art that is hard to replicate, nor because of coherency problems (which may be improved significantly in the next few years).ReplyDelete
Problem 1: Knowing what you want. The beta testing for most of these models is done in a discord channel to monitor the content being created. Having watched hundreds of people creating images for hours at a time, I can say that 90% of the material created is awful mainly because of the prompts people give. If genies existed there would be professional wishers, because most people don't actually know what they want and can't describe it.
Problem 2: Referencing concepts with precision. A picture is worth a thousand words because it's nearly impossible to specify certain objects textually. Try describing the exact shape of the coca cola logo and you'll see what I mean.
Current models work great for making pretty pictures, but if you try to create an image to a specific brief it fails instantly.
Personally I'm experimenting with interfering with the image diffusion process to manually guide it in the direction I want and fix errors as I go. I think a fusion of human art direction and AI grunt work will produce the best of both worlds in the end.
With regards to the human-ness of art and creativity, I'm not convinced that an artist is any different to a blacksmith. An artist is essentially (but not entirely) an architect of experience. One learns a craft and solves problems just like anyone else. I didn't become an artist to express my personality, I had some ideas in my head that I wanted to instantiate in reality. Any tool that helps me do that is a bonus, and if there was a way to pull those ideas directly out of my brain rather than grinding eggshells and scratching things with burnt wood then that would be ideal.
For reference, I guess, here are some images I've been messing around with. The first set are for my dying earth D&D setting. These are done with just the text prompts and no manual interference. I hope those will be much better and more coherent.
Interesting comments! I love the idea of the professional wisher in a world where there are genies, by the way. Now that is a novel waiting to be written.Delete
The images you shared look great but have that same odd (and in its own way fascinating) quality of looking as though they were created by something inhuman. This could just be because I know they were AI-generated and I'm projecting my own ambivalence about that concept into the pictures themselves. I'd like to run a 'taste test' to see if it is possible for people to spot AI-generated from human-created art.
Your post comes at an interesting time for me. I’ve only just started to experiment with these technologies and my own opinion is still nascent and unformed. My instincts run something along these lines: It’s fine to use the output of these technologies, but I would draw the line at charging anyone money for it. For example, if you want to generate a picture to help your players visualize something and you show it to them at your weekly gaming session, I have no issue with that. If you want to release something for free that uses AI generated art, I have no issue with that. You want to put a few pieces of AI art on your blog, that anyone can look at for free? No issue there, really, either. In any event, I think if you use this tech, you should be transparent about its use.ReplyDelete
There are other uses I object to not at all, and in fact think could be a net positive. For example, I think it could be incredibly helpful in commissioning a piece of artwork – you can give the artist something semi-close to what you would like them to do, or vice versa, they could use an AI to generate a few different images to show to you as prototypes for what they actually do.
But something feels slightly dishonest about putting together a commercial product that relies heavily on AI generated writing or art, in spite of the fact that it may have taken hours of playing with prompts to get something you wanted the same way it would take hours of manual work for a human to produce an image you wanted.
I don’t know why I feel this way yet – again, I’m still exploring this technology and have not really come down on one “side” or another. A lot of my professional work over the last 15 years has been in the application of machine learning to legal documents, and I wonder how much that influences my thoughts on this topic. I certainly understand some of the concerns that AI art raises, though part of me is saying “well, computers have been changing or removing the need for certain jobs for a long time now, why should creatives be treated as a sacred cow?” I wonder if some of the way we see this may come from whether we approach our own creation of art as amateurs or as professionals. For whatever it is worth, I’ve remained an amateur artist / writer / musician / what have you my entire life. I like to paint, for example, though I’m not very good at it. I will go on painting in spite of the fact that there are machines out there that can make stuff that is technically much more skilled than my creations, and I suspect a lot of other people will do the same – for many of us, we HAVE to do this stuff, there’s such a strong compulsion that it’s not really even a choice. So I don’t think that the advent of AI art is going to stop humans from creating things. But it may put people out of a job, and that does bother me, though I’m not certain why it should bother me more than say, automated check out machines or the phone tree systems you get when you call support, or something like the driverless car.
I certainly understand the feeling of being a luddite. I have often thought that the internet has ruined everything, that social media in general is a mistake (I miss antisocial media!), and recently I have been seriously considering throwing my smartphone into the garbage. Sometimes I wish I could take the entire world back a ways technologically and kind of freeze it there, though I’m not sure if that is perhaps pure nostalgia on my part.
It is a complicated topic, and I applaud you for taking the kind of stand on it that will encourage discussion. I’d love to hear what some of your other readers have to say about it as well.
I agree wholeheartedly about the importance of transparency and not charging money.Delete
You bring up an important point about automated checkout machines and driverless cars. I do think a case can be made that there is a fundamental difference between creating art and manning a till or driving, though it would ultimately be a religious/theological one. But in a sense I'm not sure it matters, because I'm also against automation of those other activities as well! I don't use automated check out machines precisely because I think it's dehumanising and I don't want people to be put out of work, and I wouldn't get in a driverless taxi or whatever for the same reason.
Silicon Valley innovators would make the case that automation has always happeend and has always put people out of work, and that ultimately the net result has been positive. I'm much more ambivalent about progress, but I also think that our most useful technological innovations have been transformative rather than automative. Penicilin, cars, planes, containerisation, the radio, the train... These were not just more efficient automatisations of pre-existing activities. They were new technologies. We're kidding ourselves if we think automated check-out machines or driverless cars are in the same category.
Yes, the bit about the net result being positive - I agree with you that I'm not so sure that's true, and to me it seems to be becoming LESS and LESS true at this point. And I should clarify as well that I absolutely agree that the most useful technologies have not been improvements in automation, but brand new technologies altogether. And the point about things being dehumanizing is well-taken, I agree a lot of these automation technologies are dehumanizing as well. Who wants to deal with the phone tree for god's sake? Ugh!Delete
Ultimately I may very well wind up holding essentially the same opinions you express in your original post. I haven't decided quite yet, but I think you make some strong arguments.
You say something below about translation and how that has become something that computers do and how this will happen in field after field. I do think you are right about this. It's true that while these technologies aren't in the same class as something like the radio, they have the ability to be just as if not more disruptive because of the way they disenfranchise people. It used to be that new tech might result in a loss of jobs in one area but an increase in jobs in other areas, but just because that was a pattern for a while doesn’t mean it will continue that way. It’s not a natural law or anything! Perhaps I am just a pessimist!
Yes, this is the point exactly - just because previous innovations helped to produce as well as destroy jobs doesn't mean we should be blithe about this wave of technological innovation, which seems qualitatively much different to all those that have come before.Delete
My feelings on this are close to yours. As an artist (an amateur musician and amateur writer, mostly of D&D campaign materials) I find it all very unsettling. The technology may be somewhat crude now, but what about in 5 or 10 years?ReplyDelete
There are uses of AI that are absolutely brilliant, and there are uses that are dehumanizing and the epitome of capitalist mechanization. I read that the Peter Jackson "Get Back" film used a custom AI program to separate out vocal and instrument tracks from very rough recordings so they could be remastered properly. AI can be used to pull out individual voices from a recording of many people talking at once. AI can increase image resolution or upscale old video to HD. All of these things are tools that can be used in the service of human creativity.
But then you have the dehumanizing side. What happens when AI programs can simply create a new Beatles song using young John Lennon's voice? Are we moving towards a point where human artists are no longer necessary, except as raw material for the machine training data?
I feel like the ethics and philosophy are lagging behind the technology here. (As per usual?) The potential for "deep fakes" is troubling too.
Yes, I agree, and you make an important observation about the ethics and philosophy lagging behind the technology. This is always the case, as you point out. But the problem is really acute, here. The pace of technological change is now such that we still haven't begun to figure out the ethics of smartphones or social media, and yet here we are faced with another epochal transformation. We're not lagging behind one technological big leap forward, but several.Delete
By the way, humans becoming unnecesary except as raw material for the machine training data has already happened in the field I used to work in - translation. Translation is a dying profession. This will happen in field after field.Delete
Hell's teeth, I thought that was fairly safe. Fine, maybe the lucrative-but-dull business of translating tumble dryer manuals into Latvian would go, but the elements of nuance in speech or literature one would think would remain. Even the relatively simple stuff in a news cycle tends to have a certain amount of cultural reference or sledgehammer irony involved that you might want a bit of human fine-tuning for.Delete
Yes - by and large what happens is that the translation is done by machine and then a human comes along and 'edits/proofreads', correcting mistranslations along the way, for a tenth of the pay they were once given for translating. For most translation work this suffices. For the high end technical, legal and literary work there are still some translators out there making a living, but vastly fewer in number than there were 10-20 years ago.Delete
I agree that AI art has many limitations (although it can do many things remarkably well), but if it’s cheaper to produce “good enough” AI art then that is probably going to impact a lot of jobs. In many fields it’s unlikely to matter whether customers prefer AI work or not - unless a brand’s audience is intimately tied to its artists by name (for example, Magic: the Gathering cards) then speed, cost, and control are likely to win out.ReplyDelete
Whatever the benefits of industrialization felt by future generations, the luddites were correct that the machines were going to impoverish and immiserate them.
I increasingly find that my viewpoint is similar to yours, but I'm a lot less hopeful that it will be of much help. In visual art in particular I think the robots are winning the war. Did you see the story about Jason Allen's AI generated artwork "Théâtre D’opéra Spatial" winning first place in the Colorado State Fair’s digital arts competition? Again it's a rather soulless image - or at least, with hindsight it appears to be so - but just as vast majorities of populations will gladly watch movies by "Peter Jackson" (surely some sort of machine manifestation?) I think that the unstoppable combination of economics and "looks OK to me" mean that AI will soon be putting a lot of visual artists out of work.ReplyDelete
As for writers - probably them too, and for similar reasons, albeit slightly further down the line. That "fair trade" scheme can't come too soon for my liking.
My "nanodeities" project, a book of tiny gods conjured up via a human-AI collaboration (and most definitely illustrated by a living, breathing human), is still on the cards, but getting pushed further into the future by other demands on my time. It feels like something which would be extremely pertinent if produced now, but may well feel dated by the time I finally get around to it.
I did see that story and a BBC news feature about it. It amazed me how blithe both the artist and the judges were about awarding the prize to that piece....but I also wondered if they just did it as a publicity stunt, or just to try to appear "with it".Delete
No judge in an art competition should award anything to AI generated art (unless the competition is specfically for "best AI generated art"). Shouldn't that be obvious?
I think the point was that they didn't realise it was AI art, and they judged it to be better than all of the other entries.Delete
I think that AI art will scratch the commercial itch sufficiently and be so much more cost-effective for publisher that it will do to the fantasy art market what photography did to the Golden Age of Illustration in print magazines at the beginning of the 20th century (which itself was launched by the new lithographic printing technology).ReplyDelete
Truthfully though, the fantasy art market is bloated because of the previous tech break-through --- digital painting --- which massively lower the bar in a multitude of ways.
What AI art won't do is take away the urge for people to create. The die-hards will persist, even after the financial incentive for unimaginative and repetitive Magic: The Gathering art goes away.
The real question is will the world be richer for AI art? What good is a new technology if it just makes the majority of folks depressed?
That is the real question and thank you for putting it very succinctly and clearly!Delete
Personally, I doubt that any AI is going to create art at the level of professional human artists until it becomes sapient. And when we get to that point, we'll have much bigger ethical and legal questions to worry about.ReplyDelete
AI feels to me like the next industrial revolution. A massive increase in productivity for humankind leading to a massive increase in material goods and wealth for the human race, but with massive cultural shifts in how we perceive work.ReplyDelete
I'm sure AI will fire a bunch of artists in the future. There's a reason tech innovation is called "disruption". But it will also make great art accessible to everybody. Shouldn't people be able to access great art without a barrier based on how much money is in your bank account?
On the other hand, professional artists will probably be forced to shift how they perceive their work. They may not be able to charge as much. They may have to get a different job and pursue art purely as a hobby. So I can see the anxiety on the part of professional artists, who probably already struggle in the increasingly saturated digital art market. We saw how NFTs were a desperate attempt to assign very artificial value to art by artificially constraining its supply using the blockchain. I'm sure we will see more thrashing of that kind in the future.
I've been thinking about AI again, plus I revisited my post which prompted you to take it me seriously. I'm guessing that this post of yours may have been prompted (indirectly) by some conversation on Twitter regarding the ethics of using AI to do one's illustrations. Regardless, I think there is a big difference between images and writing which also feeds into the ethical debate.ReplyDelete
I was (rather immodestly) impressed, on returning to my own post, by the utter whackiness of the prompts I had to write in order to generate AI content. By comparison, the AI generated examples, while fascinating and playable, lacked a certain something. However, they also so contained new and very creative elements which I would not otherwise have thought of. In particular, I was gifted an evil wizard who was burying a village in sand dunes, and who possessed a wand and potions which gave control over the movement and other aspects of sand.
From this seed, I could easily ditch the actual text generated by the AI, and run with the wizard idea, writing my own version of it and adding my own spin, perhaps with further trips to the AI in order to flesh out further details. While it's possible to do some to-and-fro with an AI-generated imagery, perhaps finessing it in Photoshop or recreating elements in one's own style, this feels neither as easy nor productive as the textual alternative.
This use of AI produces something more like one half of a comedy writing duo, not there to do the entirety of the work, but instead to suggest and finesse ideas. I don't see any ethical concerns with this (unless perhaps you are a comedy writer seeking to kill off your writing partner) and indeed I plan to do it myself if ever I get around to my Nanodeities project.
I am certainly not the first one to have come up with this idea, and indeed a friend of mine (who is a copywriter for computer game RPGs) has developed software with exactly this aim in mind (which I have tried myself, but haven't yet amassed enough of my own writing [in the particular style and context of the stuff I wish to write] to train the model). It's called Laika: https://www.writewithlaika.com/
I spun this thought out into a blog post:Delete