Why I’m STILL Cynical About A.I. Despite Hearing A Few Good “Songs” Made By It…- A Ramble « PekoeBlaze

Before I begin today’s article, I want to point out that I am NOT “Pro-A.I.”, and this fact is integral to everything I’ll be talking about here. But, as cynical as I usually am about generative A.I. these days, there are some very uncommon moments – mostly regarding A.I. “music” – when I stumble across an example, by virtue of pure chance, of one these soulless machines actually belching out something good.

Last year, it was the song “Metal Fight For Freedom” by Feral Cry – a fake heavy metal song, created without a single human so much as touching a musical instrument. And, whilst I’ve certainly heard better metal songs, this one wasn’t terrible. I found myself listening to it multiple times, had parts of it stuck in my mind occasionally etc… It was good enough to become part of my nostalgia for last summer.

Still, this article was inspired by a video I found on Youtube in early June. I’m not linking it here, and it wasn’t completely A.I., but someone had used one of these programs to “re-imagine” five songs from the metal band System As A Down as 1970s hard rock songs.

Obviously, bands like Queen had been press-ganged into the “training data”, and all of the lyrics were written by Serj Tankian and/or Daron Malakian (making them infinitely better than anything that a LLM could generate) – so there was a human element to it. And… Wow! Even a cynic like me couldn’t help but be incredibly impressed by it. It was transcendent! Thankfully, this collection of songs wasn’t for sale because, well, I would have bought a copy. And listened to it non-stop for weeks.

And I want to talk about how stumbling across very uncommon examples of “Wow! This is actually GOOD!” things that involve A.I. doesn’t mean that you suddenly have to become “Pro-A.I.”. You can still be cynical about generative A.I. – because there is a lot to be cynical about – but also acknowledge that, once in a blue moon, it occasionally excretes something vaguely worth looking at.

The old metaphor of a room filled with chimpanzees with typewriters accidentally writing the works of Shakespeare if given enough time springs to mind here. A lot of people use A.I. these days, and millions of pieces of A.I. media are being generated every week. Statistically, by pure random chance, a tiny portion of these are actually going to be good.

The best metaphor is probably gambling. Yes, a tiny number of people have won large sums of money from this… but almost everyone who gambles end up losing money from it. “The house always wins” and all of that. You are much more likely, to the point of it almost being certain, to lose money from gambling than you are to win any. Gambling is a scam. And it’s the same sort of thing with generative A.I. – there might be a one in a million chance that it makes something everyone thinks is really good… but it’ll often just spew out mediocrity at best. Like gamblers, A.I. users “lose” much more often than than they “win”.

After all, unlike something made by a human, the people generating these things can only make a fraction of the number of creative decisions that a real artist, writer or musician can.

Whilst real human creativity takes practice and – if you’re a beginner – you will probably go through a phase of making stuff which seems “worse” than anything an A.I. spews out, you are the one in control of it. You are the one making the decisions. You get to express yourself, rather than crossing your fingers and hoping that the machine spews out something vaguely similar to what you wanted. Every word, every brushstroke, every line on the page, every note etc… is something you have chosen. It’s a satisfying game of skill, rather than an uncertain game of chance.

And, whilst you will need to practice, this means that your odds of eventually making something really good even vaguely consistently are much higher than that of the average A.I. user. Going back to the pieces of good A.I. “music” I mentioned earlier, they were so noteworthy to me because they were either completely or partially made by A.I. Because this level of quality is not common for A.I….. but it is for humans.

All of my favourite songs, the ones I’ve listened to for years and consider to be total perfection… all human. This is literally the norm for high-quality music. Humans make it much more often than A.I. does.

So, if you hear, read or see a very uncommon example of A.I. actually making something good, you don’t have to have an existential crisis about it. It’s just random chance, pure luck. Good pieces of A.I.-generated media are so shocking and noticeable because they are so incredibly rare. On the other hand, a skilled and well-practiced human artist, writer or musician can put out amazing stuff a lot more consistently. It might take them more time and require more effort, but real human creativity is more consistently good more often.

To re-use a metaphor and misquote an old phrase from the “Duke Nukem” games. “Always bet on humans“.

——————–

Anyway, I hope that this was interesting 🙂