Are you sure these words have been written by a human? A few years ago, a Danish think tank predicted that 99 per cent of the internet will be generated by artificial intelligence generated by 2025. The EU was a little more conservative, finding that 90 per cent will be created by bots by 2026.
Luckily for you, The i Paper is a reputable newspaper, written and edited by real people. But in the time it’s taken me to research and write this, AI could have churned out hundreds of articles.
And that’s exactly what’s happening. There are already thousands of automated news websites according to analysts at misinformation experts NewsGuard. It warned that “the odds are better than 50-50 that a news website claiming to cover local news is fake”.
Earlier this year, hundreds of people turned out for a Halloween parade in Dublin which simply didn’t exist. According to The Independent, the event had been dreamed up by a AI-generated website based in Pakistan that was farming clicks for advertisement revenue.
Facebook has already succumbed to so-called “AI slop”, images of everything from Jesus in South Korean military garb to a grandmother in front of a knitted Lamborghini.
The motivation behind these wild fake images is simple: Facebook has a creator programme that pays users who generate engagement. And this stuff sometimes does get hundreds of thousands of likes, most probably from bots, but some will also be credulous users who really do believe they’re looking at a giant cockerel made of garlic.
Earlier this month, the former Greek finance minister Yanis Varoufakis fell for an AI-generated image of Bono and Bob Geldof supposedly protesting against Ireland’s support for the International Court of Justice’s genocide proceedings against Israel.
A few hours later, he tweeted “Friends tell me that this photo of Bono with an Israeli flag is fake. I hope it is. For his sake. [It would help if he made a statement]”.
I feel bad for Varoufakis. He grew up in the pre-internet age when almost all information was studiously researched and scrutinised, from encyclopedia fact checkers to radio news editors. Now his newsfeed has been flooded by uncanny images and texts designed to fool.
We’re all going to find ourselves fooled at some point, as the models improve. In some cases, the machines already know us better than we know ourselves.
An AI polling model created by two 19-year-old American high school dropouts managed to predict this summer’s New York Democratic primary to within 371 votes.
What’s fascinating is that their model didn’t involve speaking to any actual voters, it was all simply worked out on a server. The pair created “synthetic voters” using only demographics and predictive modelling. The result is hundreds of thousands of imagined people who in aggregate behave almost identically to real people.
Most AI companies are developing the next generation of their models, so-called “agentic AI”. These are going to allow their models to actually do things rather than merely being restricted to working as chatbots.
That means you’ll be able to tell your AI to book a haircut and the AI will read through your calendar to find the hairdresser you normally use, call the salon’s phone number, and use a text-to-speech voice to book you a slot when you have a spare hour. In a live demonstration of exactly this process at Google, the woman at the other end of the salon phone didn’t even realise she was talking to an AI.
Sure, some of this will be useful, but imagine what’ll happen when these bots are let loose on social media. Give the bot a vague goal like “generate likes” and it’ll start pumping out false stories.
I think we’ll look back at 2025 as the year the internet as we know it died. The solution is to seek out information from places that we know we can trust. Places like, um, the website you’re on right now.
A subscription to The i Paper might just save you or a family member from an embarrassing Varoufakis moment.
Gus Carter is The Spectator’s deputy features editor
