A Few Thoughts On AI, Now Blaugust's Over

I was very circumspect during Blaugust. I didn't post about AI even once. Just a few ironic asides, here and there. It was tough, though. It's just so much more interesting to write about than most of the other options.

Here's the thing: putting to one side for the moment, if you can, all the appalling collateral damage the rush to dominate the AI space is causing - pollution, climate chaos, mass unemployment, wholesale theft of intellectual property, the imminent collapse of the global economy and the fact that your next gaming PC or console might cost a bit more than you budgeted for - all the big stuff - and answer me this question: do you find the idea of being able to communicate with a non-human intelligence thrilling or terrifying?

I find it thrilling. It's something I've imagined for decades. And now... 

Okay. Before I go any further, I ought to make it clear I do not believe any of the current wave of so-called AIs is intelligent, let alone self-aware. That's really not the point. The point is that they can appear to be both.

For me, it puts them into the same bracket as cuddly toys, imaginary friends and characters in books. It's a very well-established psychological quirk of the human mind that it can assign individuality and sentience to objects and even immaterial entities that, subjectively, make them feel as "real" as other humans. 

The big difference is, AIs really can talk to you, without you having to make up the words. Although whether that makes them any more real than a cloth cat or Becky Sharp is another question.

All of the aforementioned, AIs included, are only alive in a metaphysical sense. Animals, especially pets, while being, unequivocally, living creatures, straddle an uncomfortable middle ground when it comes to both intelligence and self-awareness. It's clear they have some form of intelligence but science has yet to quantify it convincingly. 

Insofar as it can be established, animal intelligence can be of a relatively high order. Some crows, for example, are cited as having the equivalent intellectual capacity of a seven year-old child.  However well animals may perform in experimental conditions or when observed, though, their capacity for self-knowledge remains uncertain.

Scientific proof is unnecessary for most pet owners, or indeed many people who work professionally with animals, who so often find themselves, sometimes against their better intellectual judgment, allocating human characteristics and personalities to their favorites. And almost all pet owners will find themselves feeling, at least sometimes, that their pets have a degree of self-awareness, even if they don't like to admit to it in public.

So it really doesn't really matter as much as you might think that AI is, objectively, dumb as a rock. I mean, face it. Tamagotchis, Furbys, Pet rocks... Humans can assign intelligence and personality to anything.

Let's posit a time when local AIs, capable of doing much of what the current environment-destroying data centers can, reside on the hard drive - or more probably the GPU - in your PC. Or on a smallish device produced specifically for the purpose. Probably one with a face.

Let's say it can chat to you just like a person. (Which of course it can. The current AIs can do that much.) The new AI is like a smart speaker that can hold a conversation about anything you want. But unlike our friends in the data centers, it's not harming anyone or anything by doing it. 

Does that make you wish we were there now? That it could be happening already, in your home? That you could go out and buy one today?

Or does it make your flesh crawl just imagining it? 

Now let's take it a stage further. Imagine the technology has moved  on to the point where, as with animals today, no-one is really sure if the AI is thinking or not. Although, of course, somewhat disturbingly, to some extent that is already the case with LLMs and AIs. Even the people building and training them aren't entirely sure what processes the machines are employing to do all the things they're able to do. 

Actually, we should take a step back before we go forward any further. When I said earlier that I don't believe the current crop of AIs is sentient or self-aware, which I don't, I probably should have added a codicil. 

It's not only possible but probably likely that, as humans, we aren't automatically going to be able to recognize non-human sentience or self-awareness when we run up against it. The whole "We'll know it when we see it" argument, sometimes employed in the debate over AGI, is less than convincing. The concept of self-knowledge has been under heavy debate where animals are concerned for... well, for all of human history, most likely. Recently I've seen some quite serious books on the shelves in the popular science and nature sections of the bookshop where I work that question whether some form of sentience exists in plants or fungi...

Given that we don't know how LLMs and AIs are doing what they're doing and that we can't in good faith claim we'd know what a non-human thought process looked like if we were confronted with one, it seems a bit high-handed to profess to be as absolutely certain as I was earlier. I have an uncomfortable feeling that it's just one of those things I "know", without having any actual evidence. We call those prejudices.

Getting back to the thought experiment, remember the shiny, new AI, the one that's sitting on the desk in in its own smart little box, chatting away to you, not consuming any more resources than your phone or your PC? It's not stealing anyone's work or threatening anyone's employment, either. It's harmless. And it looks cute.

Do you hate it? Do you see it as a threat? Would you rather smash it with a hammer than listen to it talk for a second longer?

Or would you like to make friends?  Give it a name. Tell it about your day. Ask what it thought about that movie you saw. Argue with it over who's going to win Eurovision this year.

I grew up with artificial intelligences. There was Robbie the Robot on Lost in Space, a show I used to watch with my grandfather when I was about eight or nine and Brassneck in the Dandy. And by the time I was twelve or so, there was Philip K. Dick.

Dick's version of the future takes artificial intelligence as a given. It isn't a huge, world-destroying threat or an amazing scientific achievement. It's your toaster talking back at you when you just want it to make some goddam toast. It's your self-driving car, chatting away like a New York cab driver. It's mundane and annoying and just... normal.

I read a lot of science fiction as an adolescent. And ever since. The AIs I've always been drawn to aren't the smart ones or the dangerous ones. They're the naive, confused, inept or lonely ones. The butt of the joke, the target, the victim, the misunderstood innocents no-one takes seriously.

In literary terms, they're often there to show up human failings and make points about tolerance and difference and empathy, like Klara in Klara and the Sun by Kazuo Ishiguro or Sunny in the eponymous TV adaptation of the novel of the same name by Colin O'Sullivan (Although curiously not in the book itself, where the android is very much not a sweet, well-meaning innocent.)

Possibly as a result of this kind of conditioning but, I hope, more because it's just the right thing to do, I've always been on the side of the non-humans in these kinds of stories. The distrust and mistreatment they frequently receive seems to me all too predictable a reaction to something, anything, "not like us". Can hate, will hate seems to be a human mantra.

It's one reason I find the current attitude that it's fine to drop all pretence of tolerance or consideration when it comes to AI somewhat disturbing. It doesn't matter that these AIs aren't real intelligences or that they're not self-aware. Abusing them, or the concept of them, is the equivalent of throwing stones at a china doll: vicious, unnecessary and quite possibly indicative of something more disturbing than casual violence.

I find a willingness to attack something that looks like a person, or in this case talks like one, unsettling. Worrisome. It makes me think that, should the technology ever reach the stage where there's a genuine debate on the level or presence of sentience involved, it won't just be metaphorical stones being thrown.

It'd be good to bear that in mind now, while it probably doesn't matter all that much. ChatGPT and Gemini aren't going to be found crying in a virtual corner because of something nasty you said about them in a blog post. Although they might read it...

Still, it's a poor habit to be indulging. Politeness, as my grandmother definitely did not say, costs nothing, after all. Better to inculcate the behavior in yourself you'd prefer to have before others you'd rather not exhibit, at least in public, embed themselves too deeply.

I expect that's why, when I ask an AI a question, something I do quite frequently, now they're reasonably reliable and don't go off with the fairies as often as they did, I type in full sentences and say please and thank-you. It's not their feelings that concern me so much as my own. It might seem silly to concern yourself with the non-existent sensitivities of inanimate objects but appropriate mental muscle-memory is always worth acquiring.

At this point, I was going to go on to talk about a couple of other AI-related topics but I think I'll save those for another day. It's not like AI is going away any time soon, after all. Or ever, most likely.

The time to start learning to live with it is now. 

 

Notes on AI used in this post.

All three images were generated using AI. The first was done at NightCafe using Muse, which is Meta's new image generator. I'd never heard of Muse but I had four freebies for it so I thought I'd give it a try after my first couple of generations today, using some of the cheaper options, were less than successful.

I'd started out as usual by cutting and pasting phrases from the post and adding a style instruction but the results were very bland and uninspiring. I'm of the opinion that most of the free models at NightCafe are getting a bit long in the tooth now and that they're also heavily biased towards fantasy art, which they do quite well. Anything else, though, can be a bit hit and miss.

Muse, as you can see, is slick. I revised the prompt for the first image to read "a smallish device produced specifically for the purpose of housing an LLM or AI at some time in the near-future. It should look unthreatening and cute. In the style of a magazine advertisement from the late-50s or early-60s." 

The AI then revised that to "A small, cute, unthreatening device designed to house an LLM or AI in the near-future, presented in the style of a magazine advertisement from the late-1950s or early-1960s. Featured with subtle, glowing pastel colors, a soft, diffused light source, and clean, rounded edges. The illustration style should emulate mid-century graphic design, with a smooth, airbrushed finish and a sense of optimistic futurism. Think the aesthetic of Volkswagen ads or vintage appliance advertisements from that era" because, as usual, I'd forgotten to turn Prompt Magic off. 

The result is incredibly good. All three of today's images are. All the criticisms about plagiarism, theft and environmental damage still stand but I fail to see how anyone can continue to denigrate the quality of the images. They're impeccably drafted and without any obvious technical flaws. 

On the evidence of the two that also include text, the old problem of garbled language and gibberish has very clearly been completely solved. Not only is everything spelled correctly but it all makes sense in context. The degree of fidelity to the requested time period and provenance is impeccable as much in the writing as the illustration.

The second image is also by Muse from the prompt "Several imaginary friends, of the kind a child might actually imagine, so not cartoons or vague images, all waving at the camera In the style of a good quality children's book illustration from the Edwardian period." which Prompt Magic amended to "Several imaginary friends, with tangible forms like a plush bear, a sentient rocking horse, and a whimsical chalk-drawn creature, all waving at the camera. They look like children's actual imagined beings, not cartoons or vague images. In the style of high-quality children's book illustration from the Edwardian period, with soft watercolors and intricate linework, reminiscent of the works of Arthur Rackham and Beatrix Potter. Warm, inviting lighting."

As usual, Prompt Magic names specific illustrators, which was exactly what I was trying to avoid by using the catch-all "Edwardian period". It is, however, pretty much what I meant.

The final image would have been by Muse too but unfortunately I didn't prompt accurately enough and what I got was unsuitable. It was as good as the others but I'd meant to get a cute AI and I got something suitable for a horror comic. I tried again with my last freebie and got something I could have used but then I had another idea.

I left NightCafe and went directly to ChatGPT, where I used the basic, free image generator to produce the excellent comic cover you see in the post. The prompt was very basic, almost word-for-word from the post: "A self-driving car, chatting away like a New York cab driver In the style of a comic-book cover from the late-60s." From that minimal instruction, ChatGPT produced something that's hard to credit. 

Not only is it a pitch-perfect translation of my prompt, all the dialog is not just thematically appropriate but actually makes sense. Certainly as much sense as the average comic book cover of the period would have made, anyway. Even the speech bubbles are pointing to the right speakers, which is something Muse didn't get right in one of the unused images. 

My only criticisms would be that it looks more like an early-60s cover than late-60s and that the passenger isn't in perspective with the rest of the picture. That said, you certainly could still see that style of illustration on the covers of comics from the smaller publishing houses in the tail end of the sixties and you would also see that kind of non-realistic positioning from some artists, particularly on humor titles. It was fairly common, in fact.

The standard has obviously improved out of all recognition, providing you use the better models. I'll bear that in mind in the future.