DashMcCool

This was part of my thoughts I expressed in the “Now I’m a hypocrite” post, but that was already too long. This is related but different.

Lots of people are worried about AI taking over the world, often in a “robot uprising” scenario where Skynet nukes humans and unleashes killer bots on the survivors, a lá Terminator. That might be a possibility in the far future but I see it as highly unlikely in the near term.

What I see as a clear and present danger in the immediate future is the loss of tens of thousands of jobs in the arts. Programs like Midjourney and Dall-E (and whatever DeviantArt’s version is called) are coming for the painters and picture-makers. And doing it in the vilest way possible by stealing their own work and remixing it. Other programs like ChatGPT are stalking writers in the same way. None of these algorithms are very good yet, and you can usually tell when a picture or prose has been artificially generated, but they’re improving rapidly at a terrifying pace.

I’m writing this in March 2024; look at how good they are already and compare it to the state of the art just a year ago. How much better will these algorithms be a year from now? Five years, ten?

Years ago in the 1980s, when I was in college (I turned 59 earlier this month) some company came up with tech that could understand human speech. It was terrible and frequently hilarious, and early products using these voice recognition programs were seen as a joke. Back then I said online (I think it was on Usenet, so it might still exist) that this was going to kill jobs. Most people dismissed my concern, but when was the last time you called a business of any sort and a human answered? In my youth, if you called the doctor a person answered. Now you have to go through several menus to speak to anyone. Telephone operators were a thing. The last time I called Information (dial 411 in the US), which was some years ago, it was just a computer. There weren’t any human telephone operators to assist me.

Now voice recognition is ubiquitous: Apple’s Siri, Amazon’s Alexa, every car. It’s still only 80% accurate, but it’s been good enough for years to have caused the loss of hundreds of thousands of jobs.

We’ve already passed the point where AI composers can create convincing instrumental music. These programs aren’t yet in wide use but I’ve listened to the concocted classical music and ersatz film scores and I can’t tell a difference between stuff composed by computer and that crafted by humans. I’m no expert but I listen to film scores almost daily and even had a weekly post on Twitter about the subject for several years, so if they’re fooling me….

One of the people I follow on YouTube and Instagram is a guy who goes by the handle of There I Ruined It. He combines AI and his editing skills to create mashups and cover tunes that would never exist in real life. Several of the musicians have passed away, so there’s no way they could be the real deal. It is absolutely hilarious to hear the late great Johnny Cash sing “Barbie Girl” or Elvis Presley sing “Baby Got Back” (and just recently the still very much alive Simon & Garfunkel singing a mash-up of “Baby Got Back” and “My Humps” to the tune of “Sound of Silence”), but the implications of this tech are dire.

Several years ago Adobe showed off their program that can take a snippet of someone’s conversation and turn it into a full conversation generated by the computer. It was absolutely amazing and terribly bone-chilling. Recently we’ve been warned to never answer “yes” to any question posed to us by a stranger on the phone, not even if they ask us to verify basic information — such as, “Hi, am I speaking to So-and-so?” — because they can then use that in a simulacrum of your voice to trick the computerized operators for all sorts of nefarious deeds, up to stealing your identity.

You’ve probably seen several deepfakes by now, where they superimpose a computer generated face over someone else’s body and use the voice mimicry to make it sound like the person. Tom Cruise is a popular target of this, as is Scarlett Johanson, and even the ones from a few years ago are almost convincing. Actresses are especially vulnerable to this, as creepy-ass men are putting their images in porn, including deepfake videos of them having sex.

Often they can issue a statement that makes it clear they did no such thing and generally most people believe them. But what about non-famous women? “Revenge porn” has been a pervasive problem for years, where jilted men have posted nudes of their exes online, frequently ruining a woman’s reputation. Now imagine if they generate actual video using voice mimicry— how does a regular person recover from that?

The implications for the legal system are equally grim. Rich people now have yet another way to avoid the consequences of their criminal behavior. It’s easy to imagine a scenario where someone generates a deepfake video placing them miles away from the scene of the crime. “See your honor, I couldn’t have murdered my ex-wife in Houston when I was hundreds of miles away visiting the Grand Canyon.” It doesn’t even have to be an alibi, it just has to cause a judge or jury to have reasonable doubt. In the courtroom on the fly they could take a short 10-second video of the judge and use that to fabricate a video inserting them into a famous movie. It might look rough but the lawyer can argue, “You saw how we did this in just a couple minutes. Think how convincing it would be if we had a week to refine it.” How long will it be before video evidence is treated as suspect? Yes, there might eventually be a way to guarantee that the footage is authentic and untampered with, but in the meantime nobody is going to jail while this tech exists. The law and white-hat technology is always playing catch-up.

This type of digital generation was a key condition to resolve both the screenwriter and actor strikes last year. Wringing concessions from the studio that they wouldn’t use AI to write scripts or CGI to replace actors was a big deal, and it underscores the looming threat of this technology. Have you seen Unreal Engine 5? In less than a year it’s already improved so much that many of the photorealistic videos it produces look genuine. They use Unreal 5 to create real-time backgrounds in TV shows like The Mandalorian Using giant flatscreens in a stage they call The Volume. Go look it up, it’s cool.

As far back as 1999 with the release of The Matrix and Fight Club we saw CGI being used seamlessly in films, often invisibly. Fight Club director David Fincher even joked at the time, “Now if we could just get rid of the actors!” 25 years later his wish has come true. 10 years from now you can probably remake The Lord of the Rings on your laptop, starring all your friends. Or with James Dean playing all the roles. You’ll just plug in an auto-composer to generate a suitably epic film score.

All of which will be cool, sure, but the downsides of these AI tools will be grave indeed.

I’m not thrilled with AI art taking over everything, particularly as it’s clear that the people creating these algorithms are essentially stealing the work of real artists. As something of an autodidact I did a deep dive into the subject of the current state of AI in all its forms, and eventually decided to try it out for myself.

As is clearly evident from my Gallery, I was quickly hooked and started using the various tools available to create scenes and characters, turning me into a hypocrite. It’s insidious how addictive it is. The rendered results of the prompts are produced in seconds, which gives you a dopamine hit, and 99% of the time they come out wrong. Either slightly or massively. Some of it is real body-horror stuff, with melted faces, twisted bodies and extra limbs/fingers/toes.

But the wrongness of the resultant pictures simply gamifies the whole process as you try different keywords or put phrases in a different order in the pursuit of the image you’re trying to create. Turning the whole process into a sort of game also boosts the dopamine hit, and every now and again you end up with exactly what you want.

Several scientific studies have shown that intermittent rewards are the most effective at hooking people. (It doesn’t work as well on dogs; I’ve tried using intermittent rewards while training several of our dogs - we currently have five - and they really favor consistent rewards. But for humans? Fuggedaboutit.) Intermittent rewards can hook even the least susceptible, making us come back for more. This is the reason why gambling can destroy people’s lives, and it’s the basis for loot drop boxes and raid rewards in video games. Now And then you’ll get something cool or rare or epic, and that causes you to want to try again. Even if it’s a vanishingly small percentage of a chance from the random number generator, it makes people want to try, try again.

I’ve always hated raids in MMOs for that very reason: you participate in something that takes hours on the off chance that you’ll get that awesome hat or amazing stat booster, but the odds are one in a million (or worse!) that you’ll get anything, if I’m working toward a reward I want to get said reward. I guess I’m like a dog in that regard.

But damn if this pursuit of getting just the right image didn’t hook me in exactly that way. If it were one in a million attempts to convince the AI program to produce what I want then I’d likely walk away. But it’s more like 1 in 10, which is clearly enough bait to keep me on the hook. And every now and again one of these programs will hit it first time out. Bizarrely yet absolutely predictably, I sometimes get disappointed that I “won the game” so quickly. “Well that was easy,” I think, kind of wanting that intermittent reward. You’d think those moments would balance out all the times the AI utterly fails to give you what you want, but it doesn’t.

I’m disappointed in myself for enjoying the chase so much and I hate the fact it’s turned me into a hypocrite. But I’m equally disappointed that someone finally found a way to get me with intermittent rewards. After 59 years of being able to resist such things, turns out there’s a hole in my armor and I can no longer claim special “I’m not like other people” status. For nearly 30 years my wife has maintained that “You’re nobody’s target audience,” because I’ve proven for decades that I’m impervious to marketing and salesmanship. She’s literally one of the smartest people on the planet, a polyglot with a masters in psychology and a PhD in marketing who’s been CEO and COO of several companies, including a couple Fortune 500 joints, and I’ve always been that untameable unicorn that won’t fall for the wiles of professionals like her.

Yet here I am. Stupid AI, exploding my self-myth.

DeviantArt, please stop asking me to “welcome new artists”. First of all, none of them in the past 6 months have been artists. They’ve been AI code kiddies who are typing in prompts to AI art thief/generators which are then pooping out this generic junk. One of the keys is seeing 22 pages of new stuff that was all uploaded in a 2-hour period, all of it vaguely resembling some weird Photoshopped mash-up of other artists, some famous, most not. Second, I can;t welcome 37 people a day. WTH is that all about?

DeviantArt amusements:

1) I keep seeing a piece of art I think is cool, only to discover I added it to my favorites 13 years ago or something.

2) DA keeps recommending “Suggested Collections” and one of them is inevitably mine. 😆