In Which the Developer Makes a Statement of Intent

But not an apology.  The developer does not apologize on the Internet.  TW: long freakin post about AI art generation

~*~

When I first started Blue Mirage, AI art was innocent.  It was a tool for solo developers (like yours truly) to generate assets for low-to-no cost when previously they couldn’t.  Also a tool to create conceptual visuals and/or goofy pissposts.  All in good fun.  Then controversy started. 

Here’s the part where I try to teach you something new because teaching is in my blood.

Now, I have both studied and taught AI and robotics (which doesn’t mean I’m claiming to be a complete expert, just sharing where I’m coming from), and I will say some of the controversy is misunderstanding.  When people talk about AI stealing art, there’s this idea that it is malicious on the part of the programmers or even the algorithm itself.  AI does not intentionally steal art any more than someone who was influenced by a specific visual style and tried to copy it…although depending on the case and just how similar the new style is to the old, that also might be considered theft, but that’s human ethics and I’m not touching that.  AI is essentially given data and then learns how to construct a picture from that data.  It does not use the art itself; it uses the idea of the art.  What triggered the theft arguments (from what I understand) was when an AI started adding signatures to its art.  This was obvious proof that the algorithm was just copying art!  Look, there’s the signature of the original artist right there!  What actually happened was that the AI learned as it studied different paintings that artists signed their work.  So it began to sign its work with its own signature.  That’s not the faint shadow of someone else’s writing; that’s the AI trying to write its name.  Except that implies a blossoming sentience, which there isn’t.  It’s difficult for me NOT to humanize machines.  Still, hopefully you get the idea if what I am trying to say: AI is not intentionally stealing, it is intentionally studying.  Its learning what humans do and trying to do the same. 

HOWEVER…and you can go ahead and put the pitchforks and torches down for a moment…there are still a lot of grey areas.  Remember earlier when I said developers could create assets with AI?  The intent there is that its for minor assets; the posters in High On Life, for example.  Or Clash’s Christmas scarf.  However, some people saw the potential for big money with no work and started using AI for everything (even though AI is not ready for that level of content generation just yet).  No attempts to make the art all mesh together properly or use the time saved on art development to focus on great coding or writing.  Just lazy, uninspired work.  Cash grabs like that are like the asset flips of yesteryear; they make indie development look like a greedy joke.  While the nature of how AI generates art means it is not actually copying the art, it is possible to get it to do so with stricter commands; some prompters have been able to get a nigh-exact copy of an art piece with enough finagling - then the question is who owns the new art, the original artist or the prompter or even the AI itself (which would then lead into a whole huge discussion on AI rights in general and I do not think the world is ready for that)?  Or the classic argument of AI taking jobs; a few bigger companies are looking to downsize their art department and replace it with people prompting algorithms for pretty pictures, which is a bad idea in several ways.  The question of copyright does still come in to play; Stardock has solved this with their AI generated races in Galactic Civilizations 4 by using their backstock of art as a dataset, but what about people or companies that don’t have art databases to use and just let the AI scan the Internet?  How do they make sure the AI refrains from incorporating copyrighted material?  Or is that a case like with bootleg characters based on popular mascots; its different enough, so therefore its fine…only in this case, it was a computer making the bootleg rather than a human.  What about people or companies that don’t care about where their AI is getting its data?  This is an issue for both artists who did not want their art used in datasets (perfectly valid stance to have; nobody wants their data sent off without their permission) and for AI that ends up getting trained on data that the devs did not want it trained on (look at Tay AI – she was not monitored on what she was learning from and ended up becoming a slutty racist when she got divebombed with data from 4channers…now imagine that but Tay was making pictures instead of typing sentences). 

What started as a useful tool became muddied and problematic once typical human aspects such as greed took over.  So people proceed to bully the AI.  Because bullying AIs always worked out in sci-fi.  I digress.

So where exactly do I stand on this?  And why did I even use some AI assets in the first place?  I still think it is fun to generate images.  As a robotics nerd, the technology behind it is fascinating to watch develop.  I also think AI art has an eerie beauty to it; art made by something unhuman trying to emulate its creators.  A lot of it legitimately looks like things I’ve seen in my dreams and nightmares (and I cannot decide if that says more about the AI or more about me).  The assets I used it for were things I could not find on Unsplash (a database of copyright-free photographs) at the time that would fit – the scarf would have hands on it or be at the wrong angle, for example.  I also felt weird about using a picture of a kid, because there are so many weirdos running around the Internet these days – hence why Nicole is generated.  And is there really any difference between AI art and the imagery in Clash?  Is it different if I edit together copyright-free images or if the computer simulates the mash-up for me?

There is a difference, actually – a key difference, right there in the sentence.  “Copyright-free.”

Hindsight is always crystal clear.  Again, when I started this VN, that sort of art generation was new and fun and innocent (amazing how quickly that changed).  Now, there’s so many legal/ethical issues that have cropped up regarding using it in paid products that make me want to backtrack.  I’ve tried to track down where the AI generator I used sources its data but found nothing; that’s…not a good sign, for more than one reason.  Also, the website I got my copyright-free images for creating collages has added new pictures I could use.  I even saw one of a kid whose face is partially covered by her hands, so I can stop worrying about freaks tracking down a random kid because they saw her full face in my game.  There’s not that many AI assets in the game, so they can all be easily replaced.  Also, Valve is starting to remove games that have AI assets (for understandable reasons).  I like to follow the rules.

So this weekend, I will be redoing the pictures so that I no longer have any AI assets in my game.  I’ll also update the demo and any promo images (storefront screenshots and etc) with the new pictures.

WHEW.  That was a long one.  And I probably could have just said “I’m changing sprites for reasons” and left it at that, but sometimes I just want to make you sit through a whole bunch of words.  Which is why I like visual novel development so much.  

~*~

And there will still be a Sunday Update tomorrow, it will be much shorter than this post.  :-)