"Teens are opening up to AI chatbots as a way to explore friendship. But sometimes, the AI’s advice can go too far."

This article is interesting as much for the comments as for the article itself. The comments are vacillating between (to paraphrase) "Yeah, I get it, talking to real humans is hard because real humans are shit," to "I don't see the harm in it, but yeah, I wouldn't take what an AI says at face value without outside verification," to "This is dangerous and will be the death of humanity as we know it."

(I'm kind of a mix between the first and second opinions, with only the slightest dash of the third one, myself. Mostly the second one.)

One comment is (direct quote):

---

We really just gonna pedal-to-the-metal speedrun making every overtly horrifying scenario from Black Mirror into reality, while simultaneously criticizing Black Mirror for being corny as shit and lame now, because all it does is tell us stories we go out of our way to make true within 5 years.

This shouldn't be a thing at all. It's monstrous, full-stop.

---

I haven't seen any of Black Mirror, so I don't know how "accurate" that statement is, at least in the context of this The Verge article, anyway. I have to assume they're talking about this? That's what came up when I searched Google for "Black Mirror artificial intelligence," anyway. If there's more than that that's actually relevant, I didn't come across it in the minute or two I devoted to searching. *shrug*

What I can say, though, is that Star Trek has dealt with this topic quite a lot, in its own way, and at least a couple decades before Black Mirror ever existed. And that's not even all of it (as of five years ago, at least, as there's been more since). Basically, all we need is for someone to invent hard light holograms now, and we'd practically be almost there already.

As for me, I've messed with Character.ai (as recently as a couple weeks ago), AI Dungeon (back when it was still cool, before it became total shit), and NovelAI (I havehad a long-ass on-going story thing I mess with fairly frequently and occasionally dabble in "one-shot" story things), and it's all interesting enough, sure. I don't mistake it for a friend or a psychologist, though, and all that I've tried have absolutely given horrible "advice" on occasion. I think it's mostly fine for this purpose, within reason, with moderation, but when people start confusing this stuff with reality, that's where the problems start.

"Thoughts are swarming my mind like a nest of cyber-rats."

I haven't tried Sudowrite, and I probably won't try Sudowrite, because it's rather a bit pricier than the alternatives like NovelAI or HoloAI (both of which I have tried) or whatever. I don't doubt that it's better, given that it's using GPT-3.5/4, rather than the less powerful (at least for now) models used by those others.

But ignoring those specifics, just generally speaking, my take on novels/novellas/whatever written by AI is this: as long as you are completely upfront and transparent about the thing you wrote with an AI being written by an AI, then I'm fine with it. You say, "Hey, look at this cool thing that this AI wrote for me," then I'll probably take a look at it and be, "Oh, that's neat."[1] I've certainly done that myself, after all. And honestly, I'd rather read a fair-to-middling story written by an AI or a human/AI collaboration than read piss-poor dreck written solely by a human.

However, if you show me a thing and then say, "Hey, look at this cool thing that I wrote," I don't care how amazingly good it might or might not be, if I find out that you actually didn't write it, but instead it was written mostly or entirely by an AI, then I'm going to be way more blasé about it, at the very least. More likely, though, I'll be actively antipathetic toward it. Or, rather, toward you. Just as I would be if you came up and said, "Hey, look at this thing I wrote," and it turned out that another not-you human was the one who wrote it.[2] And, let's be honest here, anything you get an AI to write for you, at least for now, is going to be the equivalent to something written by not-you humans, given that these AIs are trained on the writing of not-you humans. Even if you supply it with samples of your own writing, it's still going to be pulling from the writing of other not-you humans (unless, somehow, it's an AI that was trained solely on your own writing and nothing else).

TL;DR version: I think there's a place for AI-written stuff. And I think there's a place for human/AI collaborative efforts, as long as the human is honest about it. I don't, however, think that there's a place for humans passing off AI-written stuff as human-written stuff. Or, at least, I don't think there should be a place for it. And I'm not sure how I feel about humans directly profiting (whether it's monetarily or "just" reputational or whatever) off of the "work" of AI, regardless of whether the human is upfront about it or not, given that AIs are trained on the works of other, not-them, humans (willing/knowingly or, as is most likely the case, not so). If I had to say how I felt about it, I would say it feels a little hinky to me, at very best.

[1] - Disclaimer: I have not read The Electric Sea, and I'm not sure I'm going to. I'd guess that it's still worthy enough of an "oh, that's neat" and all, but I'm simply not all that interested in reading the whole thing myself, at least not at the moment. There it is, though, for anyone else (or even myself, later) who may want to give it a read.

[2] - Just take a look at my review of Reality Hunger: A Manifesto by David Shields if you want to see how I feel about the subject of plagiarism (or almost-plagiarism or plagiarism-in-everything-but-name). Essentially, I felt strongly enough about it that it's one of the few books on Goodreads for which I wrote an actual review as opposed to just leaving a star rating. Not to put too fine a point on it, I'm not a big fan of plagiarism, whether it's wrapped up in pretentious bullshit or not.

"AI chatbots like Bing and ChatGPT are entrancing users, but they’re just autocomplete systems trained on our own stories about superintelligent AI. That makes them software — not sentient."

I'm definitely in the "it would be pretty cool if they were there, but they're most definitely not there yet, and they probably won't be there for a very long time still, if ever" camp. Far too many people seem to think that they're "there," already.

If you've ever messed with things like NovelAI or HoloAI or old-school, pre-fuckup Dragon from AI Dungeon (i.e. OpenAI's GPT-3) or any of the others, I'm sure there were brief moments where it might have felt like the AI was "reading your mind" or "on the same wavelength" or whatever. However, for me at least, there were (and are) far, far too many more instances of the AI just responding with a complete non sequitur, nonsensical gibberish. Sure, it may still be proper (in most cases) English, and it may even be a sentence that would make perfect sense in a different context, but it just shows that the AI is merely trying to parrot back something from the vast database of text that has been poured into it, and those times it picked up a random bit that didn't fit at all. Maybe they're getting better at not picking bits that don't fit, but they're still just picking up bits and trying to fit them into the slot of the "conversation," rather than having, you know, an actual conversation. They're not coming up with their own words. It's kind of like a more generalized equivalent of, say, if someone communicated primarily (or only) via quotations from famous people or pop culture references or memes or whatever. I'm sure you've met a person who seems to have a quote or reference to fit just about any situation. That's what the AI is doing, except its "quotations from famous people" and "pop culture references" are just... plain-ass quotations and references, in general (without citations).

This is true not just of the text-gen AIs but, in a broader sense, also the art-gen AIs (sometimes you get a Lovecraftian horror when all you want is a simple human image, but even when it gives you exactly what you wanted, it's still just cribbing together something from all the human-made art it has been trained on), and the voice-gen AIs, and the music-gen AIs, and all the other AIs out there.

These things are just tools at best. I mostly just use them as toys or games, for the most part. They're pretty damn good at that, but not for much else, at least for me, and at least for now. I don't think for a minute that they're alive or truly aware. Maybe someday we'll get a Commander Data or an EMH or something similar, even if it's just a text-on-screen equivalent, but we're nowhere remotely close to that now. Yet. If ever.

Follow-up to this.

I also find it interesting that the NovelAI image generation thing definitely knows who some fictional characters are, but not others.

For instance, even if it decides to embellish or take liberties at times, it mostly at least kind of knows:

• Mr. Spock (the only input I gave the AI was, simply, "Mr. Spock"): #1, #2, #3.
• Captain Picard (again, prompt was only "Captain Picard"): #1, #2, #3
• Superman: #1, #2, #3
• Batman: #1, #2, #3 (though with the "Batman" prompt, it really seemed to want to make Batman into Batwoman or Batgirl, though, for instance:) #4, #5, #6
• Wolverine: #1, #2, #3

I'm going to stop there, at least with saving and uploading, because I just don't much feel like messing with it more, at least to that extent, for now. I did try a few more prompts that I won't bother uploading.

For "Captain Kirk" (the first prompt I tried after "Mr. Spock"), it just gave kind of generic looking blonde guys wearing vaguely Star Trek-ish uniforms, but that was as close as it got, so it at least sort of recognized him, but not really?

Others it didn't seem to know at all, though, like "Mr. Data" (just random girls, mostly) or "Dr. McCoy" (just random doctor looking people, both male and female).

It definitely knew "Spider-Man" (though it tended to sometimes give him extra limbs... you know, like a spider, so it wasn't exactly wrong, per se). And none of the several results I got even tried to feminize him the way it did with Batman.

It knew "Super Mario" enough to give me characters that looked like Princess Peach or Daisy or Rosalina, and sometimes they were even wearing a Mario-ish hat. It also gave a couple rather disturbing images of (what kind of looked like) Mario's head on a generic female body.

It did a better job (for the most part) with "Sonic the Hedgehog" than what the first movie initially did with him.

And it knew "Samus Aran," for sure, though a lot of the results for that particular prompt were of a rather NSFW-ish nature.

It also knew "Godzilla."

"Ghostbusters" was rather interesting because though it mostly just gave generic anime-looking people with vaguely Ghostbusters-ish clothing and (sometimes) equipment, rather than any specific characters from the movies (or cartoons), a couple of the results were actually of landmarks that looked quite like the GB firehouse or the Shandor building (complete with ghost vortex above it) rather than people, and there was even an image of a hearse (not Ecto-1, mind you, just a plain red hearse, but still, that's somewhat relevant to Ghostbusters), or even images of what looked vaguely like the ghost in the Ghostbusters logo. All of that with just a "Ghostbusters" prompt. I kind of wish now that I'd saved those (though it probably wouldn't be too difficult to get other similar images later, if I really wanted to).

It even kind of knew "Doctor Who," giving me guys who sort of looked like The Doctor (typically David Tennant or Matt Smith looking guys, and also a couple Jodie Whittaker looking women), but it also gave me a couple of Dalek-ish and Cybermen-ish aliens. And... an image of a person with what looked a bit like a TARDIS/police box for a head, which I just had to save.

And on that note, I think I'll stop right there, for real.

A post on John Scalzi's website, which is a follow-up to an earlier post on John Scalzi's website.

As for me, personally, my main issue with AI-generated art is kind of selfish, in that it has essentially all but ruined NovelAI. The other reasons that Mr. Scalzi gives are also good reasons, though.

When one reports a bug in a given piece of software when doing a specific thing and receives a response of "well, just don't do the specific thing," that tends to erode one's confidence in said software, you know? Makes one wonder why one is paying for said software, you know?

— Kane Magus (@KaneMagus) July 18, 2021

It's like that joke of telling a doctor that you feel a pain when you move in a certain way and the doctor replies with "well, don't move that way, then." Har har har. Except this isn't a joke and it's not too awfully funny.

— Kane Magus (@KaneMagus) July 18, 2021

(EDIT) So, I just unsubscribed from NovelAI for now. I might try it out again later, in a few months/years, when they've had more time to work the kinks worked out, maybe. However, this experience of essentially being told "you're doing it wrong" in response to what is clearly a bug, the specifics of which I won't bother to get into here given that they are fairly trivial, has left me less than enthused to continue on with messing around with it. Oh well. It was fun for the half a month or so that it lasted, I guess? (/EDIT)