AI Slop
In this episode, we explore the phenomenon of AI slop - the flood of bland, over-polished, formulaic content generated by chatbots and their ilk - and ask how AI content has gone from cool, to naff, to actively despised.
The conversation ranges from AI-generated writing and artwork. to clip art, stock photography, PowerPoint transitions and social media, asking whether technologies become naff simply because everyone can use them, or because democratisation genuinely lowers quality. Along the way, we examine how status, craftsmanship and scarcity shape our perceptions of value, and whether AI is simply following a path that other technologies have trammelled before.
Links:
Transcript
Hello and welcome to the Cognitive Engineering Podcast brought to you by Aleph Insights and produced by me, Fraser McGruer. I'm here with Nick Hare and Peter Coghill of Aleph Insights. On this podcast, we look at a wide range of topics from an analytical viewpoint.
And today we're discussing AI Slop. Nick, what's on your mind?
Speaker B:I want you to cast your mind back, I would say something like five, five or six years. There was a website appeared. I think it had been. It was something like the. Well, it was an.
It was an academic institution in the US Somewhere, and it was a project that I think they'd done with the, you know, the Metropolitan Museum of Art or something. And you could put in a photo of yourself, and it would generate. Use AI to generate some artwork of you, and it would upward pop, you know, a.
A Gainsborough portrait or something of, you know, of your photo, suitably turned into, you know, Gainsborough eyes, but still recognizably your photo. And it was astounding to me and really fun. And, you know, it took about three or four minutes to generate, but it was. It was very cool.
And it was like, well, this. This is going to be really exciting.
Imagine when this gets, you know, even more effective and we can start saying things like, well, now please get me a photo of me, but as a, you know, as a Mongol warrior from the 10th century or something. But no, that hasn't happened. What's happened is AI content has become unbelievably naff in a very short space of time. And.
And the term AI slop, which is relatively recent, is. Is really a. Indicates the reaction that people have to it.
And I think, you know, there's a number of things where I think we can see this playing out.
One is the use of sort of AI generated drawings, which I think two years ago I used AI generated drawings in presentations because, you know, it wasn't widely understood that you could just. You could kind of say, well, I want something illustrating this.
And it would look really good and professional and, you know, whereas now I wouldn't think of doing that. The AI generated drawings just look really. They look smooth. They look like AI and there's that whole kind of AI Vibe coded app.
Look and feel now, which things, you know, where. Where someone will say on Reddit, oh, I built this app and you look at it, it's got the same sort of color scheme and everything.
It looks flashy, but it's a little bit chonky. AI writing, of course, which is the big battleground the Wikipedia has a brilliant.
Obviously they, they are constantly fighting people putting AI generated content onto Wikipedia, but they've got a really good list of AI writing tells, which I think I might just briefly run through because I think you'll recognize them. And it's a really useful little framework to have so undue emphasis on significance and legacy and broader trends. So here's an example.
was officially established in:But AI feels compelled to make everything sound more pat and more part of a bigger trend. An emphasis on notability and attribution.
So the example they give is the subject has been profiled in multiple high quality, independent and widely read outlets, including the Australian SBS news, etc. You know, it's like we're going to talk everything up. The superficiality of a lot of the analysis. So the example they give here is.
As of the April:Promotional language Nestled within the breathtaking region of Gonda in Ethiopia, Almata Rayakobo stands as a vibrant town with a rich cultural heritage and a significant place within the Amara region. The vague attributions that observers have cited and industry reports, that stuff people say. Yeah, I mean it's actually.
And the interesting thing is all of those are things which, you know, certainly as a civil servant you are really trained to not do. It's like all of this kind of bullshit language, it's like, no, stick to the facts, please. You just cannot get AI to stick to the facts.
You can say to it, please stick to the facts and the bloody thing will keep droning on like LinkedIn, like a LinkedIn post all the time.
Speaker C:I've used that list to inform, inspire my system prompt for chat GPT to say avoid talking, please don't do any of these things.
Speaker B:Do these things. Yeah, I mean, I think the interesting thing is the feeling I get is.
And I think what it teaches us perhaps is that, you know, writing badly like this is easier, therefore it's more plentiful. Like there's more bad writing out there. I mean human generated bad writing. So of course that's ended up occupying more of its training data.
Hey, presto, it's trying to emulate the. The most writing which is bad.
Speaker A:Are there More examples.
Speaker B:No, that's.
Speaker A:That's.
Speaker B:That's a good. Yeah, go on.
Speaker C:There are syntactical examples, like M dash,.
Speaker B:The famous EM dash.
Speaker A:What's the EM dash thing?
Speaker C:It's like the sort of elongated hyphen to sort of uses. You can more or less replace them with Oxford commas.
Speaker A:Yeah, yeah, yeah, you're right. Yeah, yeah, yeah. Oh, but Oxford commas is addicted to Oxford commas. And I keep telling you, I like them.
Speaker B:Yeah. I think they're better anyway.
Speaker A:For me, I just go, certainly when I'm going, I just have to spend all my time taking them out.
Speaker B:Yes.
Speaker A:Because it won't bloody take them out, even though I keep telling it to. Yeah.
Speaker B:Anyway, so it's addicted to it. So anyway.
Speaker A:No, no, so before you go, just a couple for me, which is. I mean, kind of what you're talking about is overwritten.
Speaker B:Yeah.
Speaker A:And adjectives all over the place. And also it's got. This is something. It talks about something being quietly. It's always saying quietly. In fact, that quietly did this.
That quiet, you know, this painting quietly introduces. It's like, what. But the other thing is agreeability, especially when it's wrong. So I. It reminds you of the.
That 90s comedy sketch show, the Fast show.
Speaker B:Yeah.
Speaker A:There was a bloke who was a pub bloke, right. And he'd be down the pub and, like, he'd start speaking to a stranger, talking about, I don't know, cameras.
Speaker B:Yeah.
Speaker A:And he would go, yeah, no, cameras. And he'd be agreeing with them.
Speaker B:Yeah.
Speaker A:And he'd go to the next person, or that same person would go, but of course, there's this. And it being opposite what he said, and go, yeah, no, that's right. Yeah.
Speaker B:Fair challenge. I know it'll say fair challenge. Yeah. The other day it said. Oh, when I'm. It said to me, when I'm using Excel, I.
You know, I always use this function because I think it's, you know, it's. And I said, what. What do you use Excel for? Chat GPT. And it. And it said, fair challenge. Yeah.
I suppose what I mean is, when I advise people about what's like, you are so full of shit.
Speaker A:Yeah, yeah, exactly. Anyway, carry on.
Speaker B:Yeah, so. So anyway, I think. And I think, you know, people are very alive to this.
I think, you know, and it has now become, well, certainly a sort of mark of distinction to be able to spot this and to avoid using those kinds of things, but actually, you know, seeing that transition from quite cool to unspeakably, naff gives us quite an interesting case study into, well, where does NAFTUS come from?
And I think that's what we want to sort of focus on is, you know, where does some new cool thing suddenly pivot into being naff and incredibly uncool. And so just thinking about past phenomena which are sort of similar kind of technological domain is you.
know, Word came out with its:Everyone was changing fonts every sentence because they thought it was cool. And now that's like the sign of the world's worst church newsletter is like 15 fonts clip art. Oh great.
I could just press a button and get a guy with his, with a light bulb. Let's put clipart in every presentation. All of the word art and 3D charts and things in Excel, you know, oh great, we can make this 3D and rotate it.
All of the flashy graphics transitions in the 90s, which used to be really expensive and you know, then got massively overused on the kinds of things that the day to day parodied, you know, and of course all the stock photos, the stock photos of the multicultural people point, smiling and looking at a whiteboard is all these things that, you know, one point were rare and hard to get hold of and genuinely look professional, suddenly get put into the hands of the common man and become really, really naff because they, they go everywhere.
And, and so, I mean, you know, I, I think just before we sort of drill into that, I mean I have found myself actively trying to produce things that look anti. Anti AI, that look non AI.
So that means kind of creating spreadsheets that are sort of a bit broader, brutalist, a bit austere looking, or writing very consciously descriptive or logically precise things to show that I'm not being, you know, using AI to do this or, or making diagrams that are kind of self consciously a bit sketchy looking and a bit rough. So it's interesting that I find myself actually trying to signal this is not AI. And in some ways those aren't bad things, those are actually good.
Like what we're trying to focus on is content and not surface. And you know, we're trying to show what's actually valuable rather than these superficial signals of value and stuff.
So you know, producing smooth prose is no longer a signal of value. And so, you know, what is it that is left.
Speaker A:So hold on. So just to Peter, I want to bring you in, but just before we do. So, what are we trying to discuss here? What's the discussion?
Speaker B:I think the question we want to answer is what makes things go naff? Because is it that the things.
That these things are actually low quality, or is it that it is merely a perception of low quality because of abundance? So did we used to think these things were valuable because they were scarce?
Or is it that the thing that has become AI slop is actually just a low quality simulacrum of what used to be actually quite valuable?
Speaker A:Okay, cool. So it sounds like we still will be discussing AI a fair amount. It sounds like. It's not just that. Okay, fine, Peter, what do you got?
Speaker C:So I think it's worth discussing a few general principles for the process of de. Coolification.
Speaker B:Warmth.
Speaker C:Again, we've seen before that. Well, it's not. It's not.
Speaker B:It's not. It's not the opposite.
Speaker C:No. Because hot is the same thing as cool.
Speaker B:Right. I got you.
Speaker C:A few general things from historical examples of diffusion of innovations and why they've become uncool.
Speaker A:Yeah, just. Sorry to interrupt. Just say, I think that's brilliant because I think that was a really good example of something AI wouldn't come up with.
Come up there. But. Sorry, Peter, I interrupted you. Go on.
Speaker C:And then perhaps discuss a little bit. But why AI is different. Because AI is different as it has different qualities as a technology. Anyway, so diffusion of innovations.
Rogers:Imagine an S curve sort of starts off slow, small number of people, then shoots up and then kind of plateaus as it either maxes out or it gets replaced by something else, something better. So it's. The rate of adoption follows an S curve. So you get. There are five kind of distinct categories.
You've got the innovators, the people who are actually building it. To start with, then you've got the early adopters, and then you've got the early and late majority. And then you've got the laggards, you've got the.
Speaker A:You got your grandparents and me.
Speaker C:So it always follows this sort of. This curve. And there's a chasm which lies somewhere between the kind of early adopters and the early majority, where value is lost.
Because certainly to the. To the innovators, the cool kids.
Right, because the innovators, to them, part of their identity is wrapped up in using something that very few people use. Yeah, right, exactly. Gatekeeping. But as soon as your parents get on Facebook, Facebook stops being cool. Right.
So the coolness factor lives somewhere with the innovators and somewhere with the early adopters.
Speaker B:I can add to that. It's going to go somewhere else.
Speaker A:Well, just before you go on helpfully, we did a podcast about this and Nick, you'll remember, it's like five years ago we did this. It was like, get me the something of Reginald Multiband. What is it? Forget the, you know.
Speaker B:Oh, the, the. The Hollywood life cycle. Yes. Of Ricardo Montalban. Yeah, get me, get me. How does it go? Get. Who is Ricardo Montalban? Yeah, get.
Get me Ricardo Montalban. Get me someone like Ricardo Montalban. Get me a young Ricardo Montalban. Who is Ricardo Montalban.
Speaker A:Yeah, we need to put that, a link to that in the show notes. But it feels like what we're talking about here. Peter, carry on. And then Nick come in.
Speaker C: ,:It wasn't yet massive consumer. And there was a. There was a sort of coolness to it, like an edgy newness to it.
Speaker B: they're late adopters. In the: . Who was creating chatbot in: Speaker C:You're an innovator. So another place, another way to look at this is different types of good. You have Veblen goods. You can think of these as luxuries.
There are things where the demand increases as the price increases because of their sort of exclusivity.
Speaker B:Yeah. The value is only like. Or mainly the scarcity.
Speaker C:Yes, exactly. And then you have positional goods where the value is. You get the value from the fact that few other people have it.
And as soon as you democratize any kind of innovation, more people have it, then these values kind of go to inflation.
Speaker A:Right?
Speaker C:Yeah, yeah, yeah.
Speaker B:I mean, yeah.
So just to give a few examples of like Vabel and goods would be things like a front row seat, a royal wedding, hereditary titles, rare postage stamps. They're not things that are particularly intrinsically valuable. What they are is valuable because there's so few of them, you know, that they.
If Everyone in Britain had a front row seat at the royal wedding. It wouldn't be worth very much, you know, and I don't mean not worth very much in the sense that you wouldn't be able to sell it.
I mean, it wouldn't feel valuable. So, yeah, that's your veblen goods.
And I guess, I suppose the question is though, is, well, some things become democratized and do not lose their cachet.
So for example, books, democracy, but anyways, well, I think that's arguable, but the books, anyway, a ten pound copy of Shakespeare is not now considered to be like naff because everyone can go and buy it. Books and reading are still cool things to do.
Recorded music, right, we can all listen to Bach, but people don't go, oh, not more of that bark slop just because he's on Spotify.
You know, indoor plumbing, spectacles, photography, these are all things that have been put very much in the hands of, you know, the common man and yet are still, you know, considered worthy.
Speaker C:I think the key thing though is that because AI does a lot of the skill for you, that it's become much more sloppy, it's become much more, it's lost its value. If to do, anyone could do photography, but to do it well takes skill and practice.
Anyone can own Shakespeare, but to be able to quote it or to be able to talk, converse about it in a intelligent sounding way, you have to be a proper, you have to be a proper boy. Yeah, so it's sort of like, it's not the owning of the thing itself, it's the loss of skill required to do. It's still. AI just generate you, whatever.
Yeah, yeah.
So I think it's kind of like this, this, this speaks to a lot of different cultural innovations as, as things become main, as things become popular, capitalism will, will, will get its teeth in them.
Speaker B:It's grubby mitts.
Speaker C:It's grubby mitts.
And we'll, we'll, we'll, we'll, we'll sort of seize hold of it, see it, you know, develop it into a product and it will lose any kind of, any kind of coolness. It's how it happens over and over again.
Any, any kind of, any kind of art aesthetic, any kind of music, anything that was cool, looks promising, will get productized and shittified to death. It's just natural. A classic sort of non AI example is Facebook.
% in:And it's purely because teens don't want to be using the same thing that their parents are using, but also because.
Speaker A:I think the nature of it as well.
Speaker B:Yeah, they just put on moan about their sciatica and. And share. Share, you know, hoax videos and moan about political things. Yeah, it's not. It's like they don't set up community.
Speaker C:Groups and moan about the potholes.
Speaker B:Yeah. They don't do the cool stuff which used to be, you know, putting ironic and funny posts up and, you know, all of that. So it's. Yeah, I. So.
And I think this, with what we're touching on here is this question of can you have. Is there a necessary trade off of quality? And I think this is.
This is the question we perhaps have to answer is like, have is democratization in this case of AI and the generation of AI content. Must that be associated with a fall in quality? And that's. That's. I think what. Where does the quality in here, I think, is part of that question.
It's like actually when we massively increase our ability to generate writing, what has happened? Is it that we've said, oh, writing, which used to be scarce, is now common, therefore I don't like it. Or is it that we have.
What we've managed to do in order to increase that volume is simply lower the quality that actually it turns out that what we have is a load of words, but no more insights.
That the AI is generating what used to be considered something interesting and might have contained words, insights, but all it has done is learn to imitate people who do that. And, and that AI writing really is low quality.
Speaker C:Right.
Speaker B:Because part of me thinks that that's almost certainly the case.
Speaker C:Well, I think it's. I think it's not necessarily the case. So there's a possibility of disagreement here. I don't think it's. I think you can.
If you just naively use AI to say, write me an essay about this, and then you. I think there's a. There's a recent survey that's.
That coined the phrase of AI shitting, where you just hand that in without any kind of any refinements or any real human input that is doing what Nick is saying. But if you use AI as a new tool in your writing, toolkit and you use AI to improve your ideas to begin with. You use it to assist you with editorial.
But actually you are still doing just as much creative input into the task as you would have done if you'd written the whole thing yourself. I think it can enhance the quality, absolutely.
Speaker A:Except most people don't.
Speaker B:Right. And I think I do want to. I mean, I think so.
So I think, and I just to skirt around this question before we dive right in with, you know, what is going on when we are policing, you know, policing standards, if you like, the fact that we are calling out AI slop and saying that's a load of crap and trying to stop it. Finding its way into things is the late, great David Hockney, who just passed away and his embrace of, you know, new technology, his use of.
So I went to see before he died, went to the new exhibition at the. At the Serpentine. And it's his. His iPad, you know, used. He used an iPad to draw.
And you can see that some of the things he's done, for example, with Blossom is he's created a little pattern of five dots and then copied and pasted that pattern of five dots all over a tree. And you, you know, in the. In lesser hands, this would be kind of lazy, but he's nailed it. It means that actually it gives a.
An impression which would not be the same as if he'd added every dot manually. It gives an impression of actually nature's patterns in a way. And so, you know, that kind of thing where you think, well, this is.
This is how you use modern technology and allow it to enhance the skill that you have, and that for those with eyes to see, that skill is still there. And, And I think that's what Peter's saying is that's how we should be using AI.
And I think, you know, all of the, really the very best artists you think of, people like David Bowie, you know, have. Have tried, really tried to move with the times. You know, they've embraced all this new technology and they've used it to do something interesting.
And I, and I. It just, I think of that quote which Brian Eno said, you know, which is actually the imperfections in the art form, which is what people.
You look back on, like things like record scratches or, you know, the kind of particular quality of film versus video or whatever it says, actually the kind of imperfections. And I think that is what we should be. You know, a good AI person will be.
Will be trying to do something with the Ainus of its content rather than just Copying and paste. Oh please join me. This cartoon and dropping it into a presentation.
Speaker A:Okay. But I don't know if this is the right juncture for this, but this makes me think of a couple of things. I believe I'm not derailing us. So two things.
One, something that really annoys me, but I've got a sort of a one hour documentary, right. I've shown a few times in different places. And one thing that really kind of annoys me is often people come up to me and say, oh, I loved that shot.
That shot was brilliant. Now if you're a filmmaker, like me and others, if you're a filmmaker, you will know straight away that's stock, right? Okay.
And I look at it and there's a reason why it's in that. I'm fine with it being in there and it's licensed, et cetera. But the reason I know it's stock is it looks perfect, it's glossy, it's in focus.
Every sweeping vistas.
Speaker B:Right.
Speaker A:Sometimes I need that, but I can't shoot that stuff. Right. And the reason. But someone has shot. And by the way, we're not talking about AI at this point, right? This is still just other people.
But that's not my kind of filmmaking.
Speaker B:Well, this, this reminds me of what my uncle, who is a very talented artist always used to say, like the, the kind of nast art criticism you could ever make of a painting. And it's what people would, you know, the kind of common man would come in and say, oh, it looks just like a photograph that was always his.
Like that's how you know someone knows nothing about art is the highest praise they can give is it looks just like a photograph.
Speaker A:Yeah. And so just to finish this bit off.
And so I always say to people, listen, if you're looking at film, if there's any bits that's really glossy and stuff, and that's not mine, but if there's anything a little bit out of focus and like bits of rough around the edges, that's my.
Speaker B:But that. But you can tell that there's more artistry in that than in a perfectly glossy.
Speaker A:Quite possibly. Which leads on to this briefly, the second point, which is about a year or so ago I was at this sort of.
I think it's a photographer's gallery in London running an event about commercial work and also about commercial trends. And one of the things they were saying is we're seeing a trend towards more kind of real and bit dirty, rough and not so perfect. Yeah, of course.
Music to my Ears. But that is definitely referring to AI, what they're talking about. I'm wondering if this derails or gives us. Where do we go from?
Speaker C:I mean that sort of speaks to the. The craft is in the struggle. You know, it's in the fact that it's hard to do things well, to make good, shoot good video well.
Speaker A:And we're seeing lots of perfect glossy stuff everywhere.
Speaker B:And I, I think we've been here before and I think that that movement towards the rough is very reminiscent of what happened at the end of the 80s when, you know, we had reached a sort of hype where, you know, like commercial graphics had got much more, much cheaper, much easier to produce. Very nice looking, very glossy. You know, all of the typesetting was automated.
You have beautiful, lots of color, much better to have much, you know, you could have color photography in magazines and all of that sort of stuff. And things reached this kind of high point of just everything being glossy and 80s and perfect looking.
Hey presto, you know, the 90s comes along and everything is suddenly grungy and you know, cool magazines suddenly started having messy looking typefaces and you know, badly typeset or inverted commas, badly typeset paragraphs that were slightly tilted or out of place, you know, and actually like that is that, that was the reaction against that perfection. I think that's absolutely right is what I expect to see. So I think, you know, thinking about this, this concept of naft.
So I think we've talked about some of the drivers of why people are reacting badly against AI, which is to do with what it's doing and the quality and those kinds of things. But I think there's an interesting social policing element to this.
Definitely ties into what Peter was talking about earlier about the, you know, this kind of S curve. But do you remember we did a podcast a few years ago about why it is that some very popular artists are nevertheless critically panned.
And the example I always think of is Jack Vetriano. You know, people love his work, but critics hate it. And I think there is a vitriol that these kinds of people attract.
J.K. rowling is another example where she's. Her writing style is scorned by proper writers, you know, but she's hugely popular and they're not unconnected. Right.
The more someone is able to, to be successful without necessarily having what is considered to be the requisite amount of skill, the more the insiders want to make sure that they are not. They're dragged down. We've got to pull these people down. They're cheating, they're conning people into thinking they have an amount of skill.
People who can't tell, people whose highest praise is. That looks like a photograph. They are making these people successful and that's bad. So let's drag them down.
And, and that reaction I see very much in my reaction to sort of AI slop. I think it's the sort of feeling that this stuff is being used everywhere. It's very, very popular. I want to put it in its place.
production, you know, in the:You know, suddenly there were all these new middle class people and they were able to go out and buy mass produced replicas of all of the very nice artisanal things that, you know, the arts and crafts movement were producing. But now everyone has one. And all of the intellectuals at the time were like, no, no. Oh, this is horror.
Look how naff these, these little trinkets are, you know. And that's pink NAF you're talking about there.
Speaker A:Yeah, yeah.
Speaker B:And I think so that's.
I think that's where that visceral reaction to naftna, that sense of I wouldn't, the wrong sort of people are doing this now comes from, you know, when enabled, when being able to use AI tools is a bit of a refined thing that requires skill and discernment, then, you know, it's fine. Only the people who know what they're doing are able to produce stuff now. Everyone's doing it.
The last thing I want to do is someone to think that they're getting away with it. Someone to think that they're actually getting status from using AI slop in their presentation.
And I feel this great urge to sort of scorn them, pull them down. So I think this comes, this is essentially about. And where that concept of, of NAFNAS comes from is unearned social status.
It's like you think that this thing you're doing gives you social status. I'm here to tell you that it doesn't. If anything, it lowers your social status as a way of policing a status hierarchy.
And there is a sociologist called Pierre Bourdieu who has.
He proposed this idea of distinction, which I'd never seen sort of articulated, but I think is very spot on, which is that a lot of the status associated with being in a high status group, so we think of Peter's sort of small cultural, you know, elites and so on, is not actually about having stuff or being able to do stuff that other people don't have so much as being able to value things that other people don't. So being able to say, oh, yeah, it sounds like noise to you, but to me, this is very beautiful. The Stockhausen piece is actually much more.
You're too stupid to recognize it. You know, this abstract art is more beautiful than, you know, that representational crap in the next room.
It's being able to discern beauty in things which are inaccessible. That I think is very much what is going on. It's like you think that this piece of AI written slop is perfectly acceptable prose.
I can tell that it's AI slope and. And that I am getting status from having that discernment and that. So that fits. You know, I think. I think it.
It to me, just feels very much over peace with all of that. It's the same reaction that you have to you. You've bought a Clarice Cliff, you know, reproduction. I can tell it's mass produced. My one is.
Is actually authentic and it's much better. And I'm gonna look down on you because you think. You think that gives you status. Well, it doesn't.
Speaker A:And I.
Speaker B:And I feel like that, you know, oh, you've written some AI slop. Brilliant. You know, that makes you look stupid.
Speaker C:It'll be much better if, you know, in North London, you wouldn't get away with it.
Speaker A:Yeah.
Speaker B:Be much better if you had a few spelling mistakes, actually. But you can't tell that because you don't have the requisite skill.
Speaker C:You wouldn't know where to put the spelling.
Speaker B:No, exactly. I put. I put my spelling mistakes in. In really subtle, clever ways which subvert expectations. But, you know. Yeah. So there we are. Anyway, that's.
I think so. I think that's, you know, that's it. That is. Deep down, I think that's driving a lot of the reaction. Yeah.
Speaker A:Peter, you got something for us?
Speaker C:So I think I'd like to return to the S curve because I think something interesting.
Speaker A:Sorry, attend to the S curve.
Speaker C:Yeah, back to the S curve. Because I think something interesting happens at the point where it starts to plateau off when it becomes so mainstream that everyone uses it.
I think a lot of innovations become. That persist, become part of the fabric of life. Right.
So we no longer worry about computers being cool or uncool just because everybody uses them and they're just so multipurpose and they're just there. Right. And same thing with smartphones. Used to one. One time, having a smartphone had a certain cache.
Everyone's got one, so it's not a problem anymore. So I think there's something interesting going on, goes on there.
As a sort of technology or a innovation matures to that point in the S curve, you stop worrying about the coolness thing. It becomes no longer an issue.
Now I think what, I think there's something interesting about AI which might make it different and it won't become part of the fabric. And this cool versus uncool debate might rage on forever. And that's because of the rate of acceleration, the rate of change of AI is accelerating.
Speaker A:Go on.
Speaker C:We'll soon be able to do ever more impressive looking, superficially impressive looking things with AI.
So there's, there'll be new dimensions of cool versus uncool opened up all the time because it, because it simply doesn't, it simply doesn't plateau like other technologies do.
Speaker A:Right.
Speaker B:I am not sure I agree.
Speaker A:Well, and this is the thing, and I think it's probably just, I think, you know, we're a long way from being the first to talk about this.
But, but this is the thing, I think that people wonder with AI, which is this question, and I don't know if we want to chat about it or not, but it's that, you know, will there come a time when we can't tell? Right. I think maybe that's what we're talking about.
Speaker B:I think that's a lot of AI optimists think we're closer to that because.
Speaker A:I was going to say, I think, you know, it's really unclear at the moment and my general feeling is I think we're kidding ourselves if we think that there will always be a difference and that somehow the things I've talked about, the imperfections and so on that we go. Yeah, but we've still got that.
Speaker B:You're saying the AI will learn how to do those.
Speaker A:Yeah, I'm pessimistic.
Speaker B:I acknowledge that. I don't know. And I certainly acknowledge it's possible that AI will continue. Okay.
What I will say is I think there's a good chance that the current approach to doing generative AI, particularly writing, may turn out to simply have a hard limit on the quality of what could come out for various reasons that we are potentially getting signs.
I mean, Peter says the rate of improvement is accelerating, but actually I think the bigger problem, certainly if I was AI companies, is that a lot of these problems have been talked about for years now and us don't seem to be closer to being fixed. Like, like persistently putting. Why can't we actually just stop it doing Those things, can't it read Wikipedia and learn not to do those things?
Apparently not. And, and I think, but I think to drill a bit deeper, can AI writing like you said earlier, Fraser, hot and cool mean the same thing?
I mean, that's it. It is the kind of thing where you think, well, can it actually come up with innovative concepts?
And I don't want to sound like one of these AI skeptics who says, oh, it's just fancy autocomplete. Right. I get there is a lot of hidden, far more hidden structure in language than we. I certainly thought there might be 15 years ago.
And I'm astounded by how good it is.
But I think there is a level of, kind of quality of thought of, I would say kind of precision, but also in the ability to innovate with, you know, in terms of concepts and to see connections between different concepts that I have not really seen an improvement in, in what comes out of generative language models in the last couple of years.
And so it's certainly possible to me that we are not going to be able, using this approach of, you know, of kind of large language model, transformer models. It might be that we are simply not, whatever that is, cannot be trained into a transformer.
Speaker C:I agree. It can't be true. I'm assuming that there will be continued innovation in new approaches. Transformers haven't been around very long.
Yeah, we're still riding a, we're riding a existential improvement on this one, this one approach. As soon as we have another approach, we'll have a new existential kind of uptick in capability. So it's like, it's, it's.
Yeah, it's predicated on continued finding new approaches. Whether or not generally current transformer approach can get us something that's strong enough to auto reinforce. Who knows? Doubt it. Doubt it can.
If it can though, that means that this esque, this is, this, this exponential improvement will really take off and culture will just simply not be able to keep up the cultural lag. We just won't be able to keep up. So it won't never become kind of truly mainstream. And therefore we're always finding new ways to make it cooler.
Uncool.
Speaker B:I mean, I think to go back to let's say, sort of mass produced things from the 30s.
Well, I think mass produced things still have a little, you know, I don't think we've been able to solve the problem that mass produced things are not quite as high quality as artisanally produced things. And we've had 100, 200 years to get that right. And when it's still the case that mass produced things are necessarily not quite as good.
And I'm not saying it's obviously a different kind of driver going on, but it's possible that we will get to a point where that plateau is not quite at, you know, at one point that that plateau leaves a little bit of a gap where we are continuing to fight over, over, well, you know, what's the difference between human written, artisanal prose versus mass produced prose? And we will be having the same cultural punch up about, about who is, who is in that gap, you know.
Speaker C:But I.
Speaker B:Absolutely. Right. I think it's a really interesting observation that you made that actually if, if that plateau hits the top, no one cares anymore.
We've got nothing to police. There's no distinction. We're all just everything is AI written.
Speaker C:And actually it's just point. You care about the utility of something rather than its sort of artisanalness.
I think we need to draw the distinction between Peter Topia, between a petopia cool, this sort of, this qualitative thing about coolness versus quality. Because when you buy a laptop, you don't care. I mean Apple people think they're cool because they've got apples, but everyone's got an apple.
But if you're, you care more about the thing that's going to last longer, not break, have a good warranty, all these sort of boring, tedious, parenty type things. And there's a, there's a very strong thing. If you could go for a lesser brand that makes a thousand units of a particular thing.
Or you could go for a Dell or Lenovo or a big boring corporate machine that make 100,000 of that same unit because you know that their manufacturing process is going to be so incredibly dialed because if there's a problem in it, it costs a lot of money to fix. Whereas the more artisanal people might be a bit more edgy. But it's just a laptop.
Speaker B:Yeah, I think, I think, I mean, I think if we're gonna wrap up.
Speaker A:Yeah.
Speaker B:I've got a question.
Maybe I think the interesting, well I just, just made me think, you know, the interesting question here is actually when something is scarce, it's hard to tell how much of what we value about it is the scarcity and how much what we value about is the quality.
And I think the interesting thing is when you suddenly make these things much, much more easy to produce, you then can find out, well, what is the actual driver of quality. We used to think it was. Well, is it nice? Grammatically correct sentences, but they're cheap, they're not hard to produce anymore. So it's like.
Oh, it turns out it's not really the grammar or the smoothness of the prose or readability. It's actually something to do with the content. It's. It's actually something to do with the really hard bit. The thing we really care about with.
Speaker C:Good writing, novel ideas.
Speaker B:Yeah.
To go back to Shakespeare, a sentence like light thickens and the crow makes wing to the rookie wood is not because it sounds nice, it's because actually on a deep level, there is some really powerful metaphor going on there. For example. And you know, I could give me 100 similar looking sentences but none of them would quite hit the spot.
So yeah, I think that's, that's something that's really. We haven't. Certainly not resolved, but we are finding out now, aren't we?
Speaker A:Well, and actually I've got the perfect question for us, I think. Okay. Which I think perhaps might, you know, answer some of this.
So I wanted to ask a question which is, can you think of something that you personally value not just for its utility, that is mass produced? And which I think is kind of maybe what we're talking about. And the answer to that might answer the wider question here potentially.
So I've got one to kick off.
Speaker B:Okay.
Speaker A:As you know, I'm a pretty, pretty creative kind of guy, right? Very disturbing.
Speaker B:Creatives are famous for telling people how creative they are. Good sign of being creative.
Speaker A:Yeah. No, not me. I'm different to the others though. Yeah.
Speaker B:I think it's important to tell and not show how creative they are.
Speaker A:Yeah, absolutely. Within the first minute or so of meeting someone. Yeah. So look, I've got a print, a Japanese print print on my wall in my bedroom.
And it's really artistic, right? Really creative. So being serious for a moment, saw this print at an auction. I thought, yeah, I like that. I bought it.
Didn't really know what it was, but I liked it. And it looked old, maybe 100 years old, something like that.
And just by chance, I was in York a couple of months ago and they had an exhibition about Japanese printing. And so I went and saw it and it talked about processes, why these prints were made, etc, etc.
And you know, one thing I knew, I think, is that they were always mass produced. So if we think of like Hokusai. Exactly, Hokusai.
And like, you know, all the famous, you know, well, the woodcuts, the Whole point of it was that you get it out there, right? Lots of it. So it's mass. Right. And so. And I. And one thing I hadn't realized was what's the, what's the Japanese art form for theater called? Is it?
No, no, but it's called. Is it. I want to say.
Speaker B:Oh, kabuki.
Speaker A:Yeah, there we go. Kabuki. And actually lots of these essentially were taken.
Lots of prints were taken from famous scenes that everyone would kind of recognize, you know. And so this got me thinking. And so rather appositely I went to ChatGPT and said, hey, what's this all about? And sure enough it's, you know, it was.
It was a scene from a famous. Famous scene from a famous kabuki play. And so. And I knew it'd be mass produced and sure enough it was okay.
However, for me it doesn't make any difference. I look at it, I still love it. I think it's really beautiful.
But is that because I'm not living in 19th century Japan and here I am, you know, we're at this place in this time. But all I know is that it's mass produced. But I love it. I think it's beautiful. You know, I don't know. Does this tell us anything?
Interested in talking more?
Speaker B:Well, I've got, I've got the answer and in fact I'm holding it right now. I think now I really be great if we could get sponsored by the Mitsubishi Pencil Company. I know.
Speaker A:I'm surprised they haven't because.
Speaker C:Because it's becoming a weekly mention.
Speaker B:Yeah, well, I mean, I mean, let's get ways. This is the best, quite simply the best pen ever made. And not this exact one, but the ones like it, the Uniball.
I, I mean I've been using them since the late 80s and I can't live without them. I.
And I just feel like anyone who claims that they'd prefer a different pen, particularly people who use things like Mont Blanc, they're fooling themselves or they don't. And this is my other thing is like I always think people who rave about something probably don't use it.
And I think people who rave about their Mont Blanc pen probably don't use pens. They don't. Exactly. Yeah. Yeah. And I feel the same. Yeah. So anyway, that, that's, you know, I mean, I knew what my answer was going to be.
But I mean, for a mass produced. Mass produced does not always have to mean crap because this is one of the best objects ever made.
Speaker A:So the, the nick And Fraser verdict so far is, you know, AI can be cool, it can be great. Is I think where this is leaning towards. Okay, Peter?
Speaker C:I mean I'm trying to think of an example that's not. I don't admire it purely because it's utility. Very difficult.
Speaker B:Brompton is. Brompton would be a good example of.
Speaker C:A. Yeah, I mean, yeah, I suppose the bicycles I tend to buy are not mass produced. They're fairly artisanal ones.
Speaker A:See this is the problem, you see, because as we know as. I don't know if you remember, but I'm a really creative person, right. And so I look at your Brompson.
Speaker C:You've not mentioned it.
Speaker A:No, no, I know. I look at it and I go.
Speaker B:Sentence to say at some point.
Speaker A:Yeah, I definitely, definitely could. I need to go home. Yeah.
Speaker C:His performance, he's deliberately being boorish. That's right, because it's part of it.
Speaker A:Whereas you know, for you as a utilitarian and engineer, it's always going to be that.
Speaker C:But yeah, yeah. So I mean I'm sort of. My bicycles are not my normal form, but I mean I. Utilitarian things which I sort of admire.
I just like the day to day objects of life.
Things like cups, crockery, cutlery, things which, you know, my grandparents, you'd give somebody a set of knives and forks and spoons as a wedding gift because that's pretty. That was high cost, sort of important thing in life to have.
Whereas now you can get a really decent quality set of knives and forks from John Lewis or somewhere. And they are very good for what they. For what they.
Speaker B:I bet Peter's got one spork that he uses for everything.
Speaker A:But it feels like, yeah, I feel like, yeah, the beauty is in the purity of the utilitarianism maybe.
Speaker C:Well, I think I really like a utilitarian aesthetic. So I think the stand up example for me are kind of space platforms, satellites and things.
Speaker A:Oh, go on.
Speaker C:I mean to me they were just in pure engineering. The engineer does not care what it looks like.
Speaker A:So when you say space platform. Sorry, what do you mean? I think I know what you mean.
Speaker C:A satellite space station or the space station. Imagine the International Space Station, right? It's a big sprawling mass of blobs all sort of strapped together.
Speaker B:Doesn't have to worry about being aerodynamic.
Speaker C:Or anything like that. When they built it, when they designed it, they did not care what it looked like. In fact, they actively traded out aesthetics because that.
And as soon as you want to make something look nice, that's more weight and therefore more cost. The cost driver in space is extreme. So you make everything as light as possible, which means you pare it right down to that. Does it have a function?
Yes. Is it more important than anything other? You know, you order things in terms of their importance of function. Does it have a function node?
Is it there to look like. Yes. Get rid of it. Get rid of it. So to me, there's a, there's a beauty in these weird, like, weird, blobby, stuck together things.
Speaker B:I think, I think, I think the important quote, the important quote here is Dolly Parton. It costs a lot of money to look this cheap. And I, I feel like that is a thought we should carry forward in this world of AI slop is, you know, we.
That actually things like the space station and to the untrained eye, God, that's a mess.
No, and, and you know, when we get the, when you get the spelling mistake, the strategic spelling mistake, when you get the slightly too austere spreadsheet, it takes a lot of effort, takes a lot more skill to do that than to press a button and generate some AI slop. So, yeah, there we are.
Speaker A:That's the last word on that. Right, we'll wrap up there.
You've been listening to the Cognitive Engineering Podcast brought to you by Aleph Insights and produced by me, Fraser McGruer. If you haven't already, please like and subscribe. We aim to release an episode every week or two.
If there are any topics you'd like us to cover, please email us at podcasts@aleph.com thanks as always for listening. Until next time. Goodbye.
