Artificial Intelligence

Hey everyone, I'm slipping in one last post before the year ends because it will push me past my previous year's count on the main blog.

I wanted to talk a little about artificial intelligence. Recently there have been some significant advancements in this technology, and the focus has been on visual art and script writing. This has led to a lot of controversy between people who enjoy using these tools and those that find them disturbing.

This isn't our first foray into artificial intelligence. When I was younger, Google's Deepmind was being developed to play chess. It eventually managed to beat the world's best human chess player, but after his loss the guy practiced a bunch, came back and beat Deepmind. So they kept developing the AI, had another game, and the human lost again. From there on out he wasn't able to bridge the gap and AI has been consistently better at chess than the greatest human players. This happened as well with the Japanese strategy game Go, which has simpler rules but is far more complex. Eventually AI was able to beat the world's greatest Go master.

They have developed AI to write classical music to the point that it can produce pieces that can fool musicians into thinking they are human-made. The limit here is that the AI can only create music by blending pieces in its dataset, so it can't predict changing music genres. Basically, it can create something passably good and original sounding from an established time period, but it's still not "inventive".

Our most recent developments aren't our first attempts to use AI for visual art or written script either. Even in the early stages of the Internet, we had things like Cleverbot and iGod, which basically had a bunch of pre-programmed responses to common questions and statements that it would present when it saw them, as well as some sassy vague lines that it could say if it didn't recognize what you were talking about. It was kind of  off-base more often than not but this is still sort of the premise of the current technology.

I also remember systems that would let you merge faces to show what a combination of traits shared between two people could produce, and systems that could predict what you might look like at various ages. Kind of similar to the AI art.

The difference is sophistication. Our newer tech is far more sophisticated than it used to be. Visual art and written word are both areas that AI has had incredible difficulty adapting. In fact, both of these fields are listed as belonging to the "humanities", as if they are what distinguishes humans from everything else.

I'll give you a little history on these developments. Things got big when a program called "Dal-E" was created, named as a combination of the artist Salvador Dali, and apparently Pixar's little trash collecting robot Wall-E. I think it would have been self-explanatory for them to leave the "E" as assumed to mean "Electronic". Wall-E isn't an artist.

Anyway, how it functioned was that you could put in some prompts, like "steampunk, vampire, Seinfeld" and it would search the Internet for images of those things, find shared traits between them, blend them together and produce thousands of unique images for you to sift through and decide which ones you like.

But you could take it further and request that it be in a specific genre, like cubism, or even a particular person's style, like Van Gogh. I saw someone request a picture of the New York skyline done in the style of Salvador Dali.

An interesting point about this tech is that it has as much bias as the human hivemind, because it's dataset is the collection of human thought online. So if you ask for a wedding, you'll see only images of white people getting married, or if you ask for a doctor, you'll only see images of male doctors because those are the prevailing human biases. But if you want something more nuanced you can just further specify the traits you're looking for, like "woman doctor". 

Dal-E wasn't available to the public, but a little bit later a clone of the program called Midjourney was developed and released for only a $10 monthly subscription fee. Now I see AI art plastered all over the front page of Reddit.

Honestly, it's still not perfect. Notoriously it still can't draw hands for some reason. They bend in all the wrong places, are disproportionate and not realistic at all. Also, while I can't verify that I haven't been duped yet, I will often think "That looks kind of like AI", check the comments and the OP will say they used an AI system. It has a certain "look" it hasn't been able to get rid of, even with its better pieces. But it will likely get better, the music bots started off "creative but bad" before getting sophisticated enough to create passably good pieces, while admittedly being still a little uninspired.

I'll get to my overall opinion on this, but let's first talk about our written word AI, which is called GPT.

This has been around for awhile and honestly, only tech people seem to care about it. It failed to create the social stir that Midjourney did, but scriptwriting has been an even larger hurdle for AI than visual art. Basically, you can speak to it conversationally and it will respond to you, and you can ask it to write you stories.

I know a guy with the app "Chat GPT". He asked it to write a story about Pikachu and Albert Einstein. The chatbot made something up about how Pikachu got separated from his friend Ash and ran into Einstein experimenting with electricity. Einstein became fascinated with Pikachu as an energy source, and he taught Pikachu that he could one day be a genius too.

So it's a crappy story, but it did a few "human-like" things that are kind of creepy. It screwed up about Einstein experimenting with electricity, because the man was known for other things, but it did figure that he was a genius and that humans associate historical geniuses with electricity. It then gave Einstein this trait because it knew that Pikachu has lightning powers and it could create a connection between the two characters this way. Again, it's not great, but all previous attempts to use AI for script writing has resulted in bizarre, disjointed stories, and who knows how far it can be developed?

So these are the new technologies. What do I think of them? I don't like them. I think they're creepy and weird. I'm not really asking for them to be dismantled, I don't really think that's an option at this point. But from a subjective, human perspective, I find these things gross and disturbing without much potential to fill a need or improve lives.

There is some controversy surrounding who "owns" art produced by an AI. Is it the person who put the prompts in, or the moderators of the AI? Also, some artists worry about people claiming their works by putting it through the AI but having it change almost nothing, then saying they "put it through a filter" and that the almost identical work was just it "taking inspiration".

AI artists, sometimes referring to themselves as "prompt engineers", claim the AI produced work is genuine art, as the program still requires human direction and discernment. They point to the invention of photography and say that classical artists back then said it would ruin the medium as it could produce more accurate images with less effort.

I guess there's a point there. I don't think high-end art is going anywhere anytime soon. Originally, the greatest physical advantage humans had was our long distance running. We weren't the fastest, but we could outlast any other species and we used this skill to stalk and hunt prey. Then we tamed horses and invented cars, which could outspeed and outlast our best runners. The skill of running became less important and less practiced, but we still celebrate and value marathon runners. Also, we wound up getting horse farmers and mechanics. New skills were developed because of the new technology, so society didn't simply become "less skilled", it became more varied.

I mentioned before that AI can beat the best chess and go players, but we still have tournaments and champions for both these games, because even if humans aren't the best anymore, people are still impressed to see a human perform at such a high level.

I do think that the field of commercial art could change significantly, because even if you still need a human to operate the AI, it requires less skill. But commercial art was the worst form of art anyway.

What I don't like about Midjourney is that it changes the medium from one that is visual to one that is descriptive. Also, creativity is about the process of creation, and this changes it to something closer to consumption.

Whenever technology makes something easier, it makes people less skilled at it. This is even true for memory aids, even at the level of photography. In Mali, the people out there didn't have regular access to photograph technology, so everyone had super strong memories. This is because if you forgot something, it was gone forever. Since we're able to instantly look at anything we want, there's no need to remember everything anymore, so we let our brains release more information. This sounds bad, but if we can access our memory aids, what's the point of putting so much effort into remembering everything. Is that skill even necessary?

So from that perspective, maybe it's fine if people become less creative and less skilled at traditional art, if they're able to manifest their creations more easily with new technology. If it gets sophisticated enough, we might wind up in a world with better creations but less creative people.

Or maybe we're capping out on how good this technology can get. In terms of food, highly processed items are more "high-tech" but they're also less appetizing and worse for the body than more organic food. It still has a function because it can be mass-produced and feed a higher quantity of people, but in terms of quality, artificial foods have been used and modified extensively and have never been able achieve as high a quality as their lower tech alternatives. Maybe AI art will be seen similarly, as more accessible but lower quality. 

But what I'm really wondering is, who benefits from the introduction of this new technology? I can see how photography may have been intimidating to realist artists who spent their lives perfecting their crafts only to be outdone by any commoner with a camera, but at least photography had a strong utility for recording information. With this AI art, it just seems like it gives corporations an excuse not to hire a graphic artist.

Also, this stuff might get misused. If you've got access to enough of someone's signature, you could plug them into an AI and get a bunch of approximations of it. The concept of a signature is that the subtle, unconscious habits of the writer can't be reproduced easily but there's still some variation, indicating it's not just a photocopy or whatever. You can get the subtle habits with slight variations through this technology.

The chat stuff might be even worse. You could use it to spread propaganda online. Create lots of dummy accounts and have it auto-generate "creative" dialogue to give the illusion that plenty of people are engaging and agreeing with your cause.

But anyway, for myself the damage has been done. Even if AI art hasn't quite gotten that organic look down, it's good enough that every time I see someone's work online, I have to ask myself "Is that from a person or a robot?" And as silly as that sounds, that has just made my life a little bit worse.

It's 15 minutes to the new year and I haven't had a chance to look over or edit this post. I want this to count for 2022, so I'm posting it now. Hopefully it hasn't been too meandering or nonsensical