AI slop: Is the internet about to get even worse?

0:00This is The Guardian.

0:07Today, the flood of AI slop that could make your family WhatsApp even worse.

0:21The primitive forms of artificial intelligence we already have, have proved very useful. But I think the development of full artificial intelligence could spell the end of the human race. Back in 2014, Professor Stephen Hawking had a warning about artificial intelligence. Once humans develop artificial intelligence, it would take off on its own and redesign itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn't compete on what we superseded. Hawking died four years later, before the age of chat GPT, deep fakes and AI in every corner of modern life. What he might not have foreseen is its latest evolution.

1:13Hawking's still out front! Look at that speed down the main street! 327 kilometres an hour from a wheelchair! Pulling away from the Ferrari and the Mercedes. Hawking's likeness is now a frequent meme on absurd, often nonsensical AI-generated video. The kind of surreal content you might have noticed while scrolling through Instagram or sent by your dad on the family WhatsApp.

1:37Wait a second, there's a toddler and a pig out on the pitch. The internet is now awash with this kind of video, known as AI slot. I have a dream that I have a dream thinking about a dream. After my dream, I'd like to speak to the manager. What's wrong? No one's changed my diaper in the last 10 minutes. Tech giants like Meta and OpenAI are betting this is the future. But the risks from copyright and exploitation to deep fakes and disinformation are already clear. From The Guardian, I'm Lucy Hoff. Today in Focus, we were promised superintelligence. How did we end up with AI slop?

2:24Chris Stokel-Walker, you're a tech journalist and the author of books about YouTube, TikTok, AI and fake news, which makes you very much the perfect person to talk about what's going on with AI video at the moment. So let's start by talking about the new app from OpenAI, Sora 2. OpenAI, of course, being the parent company of ChatGPT. Talk to me about how this new app works. Yeah, it combines, I suppose, the best of AI image and video generation and the best of sort of TikTok. So it is downloadable for many people, although not for us in the UK from the Apple App Store. Well, I did not have this on my bingo card for the day, but OpenAI just dropped Sora 2.

3:12Sora 2 is here and it is so much fun. And I've honestly never been this conflicted about an AI update. You basically go on, download it, set up your account, and then can take a sort of picture of your face that allows you to appear in cameos. Here's the part that I'm conflicted about. Right here, you see where it says cameo? Which is OpenAI's big thing of trying to make this a social network as well. Here's me. I did the face scan, and I would say nine times out of ten, it looks insanely accurate. You can insert yourself into these AI-generated videos and have superhuman-style experiences that otherwise you might not be able to do.

3:53Obviously, this opens up a whole Pandora's box of sticky issues. You then can put in a text prompt to a simple CatGPT-style text box which says, I want to appear on a beach in Aruba, rich with money and surrounded by tortoises. and lo and behold you give a little while and it can produce a video not only in terms of the visuals but also the sounds accompanying that as well and then you can push that into the sort of feed of videos that is seen by other users here which is that social element as well as being able to export videos to your phone with a relatively easy croppable watermark. okay so let's pull up some of these videos i think and watch them together yeah hey sorry to bother you do you by any chance have any gpus i'm running short okay so this is the open ai founder sam altman through a ring doorbell surprisingly realistic evening you know why i pulled you over wait is that a giraffe sir ma 'am step out of the vehicle so we have a giraffe sticking its head out of a sunroof of a car and then yeah speeding off as if it's kind of like a clip from cops or something like that i'm being chased by a group of nuns right now i don't know why but they're fast look behind me this is absolutely not how sam altman being chased by a group of nuns through a kind of italian city of some description wow i'm on the amazon river with these elders and they just handed me a whole roasted snake it's a jake paul kind of like chicken but silkier all of this looks kind of alarmingly realistic hey quick question did you know you're actually artificial intelligence wait what no i'm i'm me you're a neural network Okay, so not only can you make videos of yourself, but also of your friends, which you share on what is a social network, Sora 2, a social platform that looks quite a lot like TikTok, but has a network of friends as you would on places like Facebook and Instagram.

5:49And that does mark new territory in the AI experience, doesn't it? That's very different to the experience on ChatGPT. Yes, precisely. And that, I think, is one of the things that is interesting about it is that OpenAI is trying to blend what it knows and has practiced very well since the release of ChatGPT in November 2022, which is the ability to create new content and is now trying to find a sort of purpose for why that content could exist. We were all a little bit skeptical of the idea of having an AI-generated feed and what that would feel like and whether you would lose touch with actual human connections.

6:27I think once we've started using this Cameo feature, it really does feel different. It feels like a new medium, like a new way of connecting with your friends in a way that I was even surprised by. Linking it up in a social sphere, I think, is a way of trying to create a little bit more stickiness with its users. So actually you have a reason to continue to revisit the app and start to use it more and more. It just does feel very, very different. And so I'm very, very happy with the way this is turning out. And it's not just OpenAI. Meta, the parent company of Facebook and Instagram, has also launched one of these apps too.

7:05It's called, I couldn't quite believe this vibes how is vibes different from sora 2 yeah so it's not all that different either it still has the same principle behind it which is that you can create ai generated videos from similar simple prompts i suppose it doesn't have that explicit you put yourself into the video aspect that sora does but actually you know they are in some ways one and the same and the thing that is I think particularly interesting about Vibes is the company that is behind it, Meta, parent company of Facebook, WhatsApp, Instagram, and Instagram Reels. And one of the things I think is notable about the kind of AI slop that we've seen through photo image generators is that often it has done best on Facebook in part because of the sharing functionalities, but also I suppose in part because of the maybe slightly less tech literate, slightly less media literate audience that is on there who are generally older, generally more easily hoodwinked by this sort of thing and also share this sort of content in groups in a way that it can spread incredibly quickly.

8:23So it's kind of a match made in heaven, I suppose, for engagement. It's quite risky, isn't it, for these two giants, OpenAI and Meta, to be releasing quite similar products at the same time. How would you say that the two products have been held up against each other? It does feel like Soria has been marked as the clear leader. Introducing Meta Vibes, the latest and greatest technology to not only waste your time, but to literally annihilate it with the adult equivalent of Cocoa Melon. Yeah, I think in part because of the audience that it is targeting, they are generally the tastemakers and actually anything to do with Meta is often seen as maybe just a little bit uncool.

9:03Vibes, the new platform for meta that lets people with no life create, share and watch short form AI videos. It sucks, right? I think that they do share a similar reaction in one way, which is there's been a general astonishment at actually quite how advanced these videos have become.

9:28Two years ago, the internet gave us. That AI-generated video of Will Smith eating spaghetti. And that was the first time I think that I remember that an AI video went really viral. Fresh pasta of belly. But it was really poor quality. It was very, very obviously fake. These videos that we just watched, some of them you can tell they're AI-generated, but others are much more realistic, particularly on Sora 2. how has this technology improved so rapidly in just that space of two years? We have generally seen a growth in the capability of AI systems in every single domain but video has been a slightly trickier problem because of the vast volumes of data that are required to actually a analyze a video and b to then produce it at the other end so we have seen through the AI boom, a huge demand for AI chips, those GPUs, which has been met by an increase in supply.

10:35So there is just more out there to be able to do. So actually, it's one thing to be able to produce some text, but that doesn't take a lot of computational power to produce a video very much does. And to produce a lifelike realistic video requires even more because you have to consider kind of internal physics, the things that we consider just innately as human beings, that an apple will fall straight down from a tree or that spaghetti will be slurped into Will Smith's mouth in a particular way. An AI system doesn't know that and so it kind of has to learn that through repetition. And why do open AI and meta think that this is the route to take, this is the strategy for their new social media apps?

11:19Is it going to work, do you think? There's no secret to the fact that a lot of these platforms are trying to create a sort of super app where actually they interact with large parts of our lives in many different ways. So OpenAI relatively recently rolled out a shopping interface for some users in the United States. You can imagine that coming to other countries worldwide. And if you add on to that, a kind of social media aspect where you are coming in, checking out what's going on in that space regularly, it cross-pollinates all of those other platforms. And I think it's also down somewhat to the money.

12:01These companies are investing vast amounts of money into generative AI. They've been told that this is going to change the world, change our economy, change our society. We're not quite there yet. And they're on the hook for that investment. So they want to maximize their chances, A, of being the winners in this race, and B, making sure that the race actually exists and we do use AI rather than just tinkering around with it as an occasional tool.

12:37Chris, what are some of the dangers of this technology being freely and easily accessible? I mean, as we've discussed, it is really hard to tell in some cases whether it's real or fake. Yeah. So if we go on, I suppose, a scale of the existential to the merely alarming, we'll start with the worst first. The existential thing is that we no longer know what we can trust. and I've been covering what we used to call deepfakes and still do somewhat but not very much anymore for almost a decade now and it used to be that when I was writing a story about deepfakes I could say very simply well just look at the earlobes the earlobes don't quite work or look at the blinking of the person they don't really blink very well or look at the fingers the fingers aren't rendered correctly on the videos.

13:33So they were easy tells of when there was AI fakery involved, and that was several generations of AI systems ago. Now, you can't really tell the difference, if I'm being completely honest. You could look at a video of Sam Altman appearing at a front door, and it might well be Sam Altman appearing at a front door. If it wasn't for the watermark, which is something that can be relatively easily removed, you would have no indication of where this content came from. So the existential risk, I suppose, is that we no longer are able to distinguish truth from fiction, real from fake, and all the consequences that come from that.

14:14I mean, we already live in a world of disinformation. This is likely to turbocharge that, right? I mean, how could that impact the news landscape, for example? Yeah, well, you can confect entire situations out of nothing in a way that is manna from heaven for places like Russia and China and North Korea and Iran. Russia used to have, and still does, an entire army unit devoted solely to disinformation. They would hire people to come into offices basically nine to five every single day, start posting comments on the websites, news websites and comment sections that were designed to try and shift opinion.

14:59Now, do they even need to do that? Could they create fake footage of race riots in one city in the UK and suddenly stoke this fear and uncertainty around us? The reality is you don't even need to be state-sponsored to do that anymore. You can, as an individual, do that. And if you have the right sort of message and the right sort of sharing of that message, then suddenly you can really alter a society. But what are some of the other risks? I mean, it strikes me that if you're using real people, whether living or dead, that is clearly an infringement. What does it mean for those people who are having their likenesses mocked up and used on Sora?

15:44Yeah, this is my merely alarming one that I was talking about rather than the existential risk, which is that without due care and attention, you could see images created very easily of people who have no idea that their images are being used in this way. Donald Trump, we know, is a big fan of AI-generated videos. Nobody likes Democrats anymore. We have no voters left because of all of our woke trans bulls**t. A video that he posted on his network Truth Social had the Democratic minority leader Chuck Schumer making a speech that he didn't make. And if we give all these illegal aliens free health care, we might be able to get them on our side so they can vote for us.

16:27They can't even speak English, so they won't realise we're just a bunch of woke pieces of shit, you know? Is America and the world ready for what this kind of thing could do in politics? It's a very good question. I mean, has the world been prepared for what's happened in American politics in the last few years is perhaps the more pertinent one. I think one is the consequence of the other. Certainly, we know that Trump is perhaps unique in his willingness to bend the truth when he is discussing things publicly. And by putting AI in his hands or in his team's hands, you have the ability to bend the truth even more convincingly to his base.

17:21And so we do have to, I think, be aware of that as a real risk. What I think will happen is less that people will be convinced by this stuff, but more that people will just be enormously skeptical of everything. So the fear here, I suppose, for me, it's more the poisoning, pernicious effect on the other side. If you have people pumping out fake stuff that we know is fake stuff, thanks to the power of AI, you become much less trusting of when you do see something, if it is real. It sort of formalises a belief that we've all had already, which is politicians have a tendency to lie. It's just that now we're not going to trust things that have previously been given the veneer of truth by being moving images.

18:19Coming up, if slop is our future, how do we make it safe?

18:39What kind of guardrails will Meta and OpenAI put on these products? Surely it would be very easy to copy these sorts of products and create them without those guardrails. Yeah. It's also, by the way, relatively easy to actually just bypass the guardrails entirely. We've seen this time and time again. These companies spend an awful lot of time and awful lot of effort building up these guardrails that they say will keep people on the straight and narrow. And then what happens is it's very easily broken and actually all chaos ends use. when Sora was initially released it had some restrictions on copyrighted content so if you are Taylor Swift and I saw Taylor Swift videos being shared in the early days of Sora yeah then you can choose to grant permission for users of Sora to create videos with your likeness now some enterprising celebrities influencers may want to do that because they think that it will get them some fame and actually people still found ways around it and actually you know in recent days sam altman has taken to his blog and talked about how they're thinking about how they deal with this sort of stuff in a much more intelligent way the approach that they want to take and it is an approach i think we should relatively firmly push back against is they want right holders to opt out of their work being used in this app by saying that actually it's not okay for you to use this, but you have to be told that it's not okay.

20:21That goes against more than a century of copyright law and several centuries of understanding of how copyright and the ownership of content would work in society. Is there anything that can be done to put the brakes on these tools? Regulation. Yeah. Telling people, no, you can't do this. Unfortunately, we don't seem to be in a situation where many politicians are up for that fight. In fact, quite the opposite. And that's because, rightly, they see vast amounts of money sloshing around this whole ecosystem. And they say, well, I want a part of that. AI isn't something sort of the future over the next hill.

21:05It's the present. It's already here. a chance to turbocharge growth, create the companies of the future and radically improve our public services. And the reality is that tech companies, they are very nimble, they are very quick, they cross borders and they can pick and choose who they want to support. If you're the UK government, if you're Sikir Starmer, do you want to throw up too many roadblocks or do you look at the investment that is coming into the country from AI companies and say, well actually that is too important for me to jeopardise. And right now you know this, the investors, the entrepreneurs, the researchers who will make those breakthroughs, they're looking around the world.

21:50They're choosing where to make their home. Trillions of pounds worth of investment at stake. At the minute, the Labour government seems to be taking that approach of we want to be on the side of tech companies. Now, it's how long that stance can hold and how willing they are to stand up while also still being the tech company's friends and getting that investment that nobody in their right mind would want to miss out on. Yeah, and I mean, conversely, OpenAI, Sora's parent company has seen its value, I mean, just rise astronomically in the last few months to$500 billion up from$157 billion just a year ago.

22:34But there is this talk of perhaps an overvaluation of AI companies, a sort of dot-com style bubble that could pop at any moment. I mean, these companies promised to revolutionise the world, but if what we're being given is this kind of video slop, is that any sign of where this is all heading? Yeah, it's notable that some of the people within the AI industry, particularly those in the leading lights of it, including Sam Ortman, are actually talking about the fact that they fear there is a bubble as well. We've started to see maybe some flashing lights on the dashboard from the likes of the Bank of England saying that they are also concerned about this.

23:18I think the arrival of AI social networks is not an indication of concern, but I think it's an indication of quite how much, despite those fears of a bubble, the companies involved believe it's vital to be in the dominant position if and when the bubble bursts. And one of the ways is by providing lots of products that people want to use and that people see value for in their day to day lives. Chris, thanks so much for your time. Thank you. That was tech journalist Chris Stoker-Walker. My huge thanks to him. This episode was presented by me, Lucy Hoff, and produced by Alex Atak. Sound design was by Brian McNamara and the executive producer was Sammy Kent.

24:08We'll be back tomorrow.

24:17This is The Guardian.