Discover more from PassTheAux
Duped by AI journalism
Plus: virtual girlfriends, Darth Greg, a strange apparition
“AI will put us out of a job!” has become a common refrain among journalists, as it has, surely, across the entire working population. But whereas with administrative or tech roles this is perhaps no joke, journalists always say it with a certain smugness. We are too creative, too perceptive, too analytical, too opinionated, to be duped by artificial intelligence!
And yet this week the Irish Times published a comment piece, titled “Irish women’s obsession with fake tan is problematic”, because it represents a “fetishisation of the high melanin content found in more pigmented people”, which turned out to be a hoax. The byline, Adriana Acosta-Cortez, “a 29-year-old healthcare administrator living in north Dublin”, was the name of a repurposed Twitter account controlled by an Irish college student who had created 80% of the article using AI (GPT-4) before pitching it, just because they wanted to “to stir the shit” about identity politics through “an incendiary article with an extreme leftwing viewpoint”.
Using the AI image generator tool Dalle-E 2, the hoaxer also created a highly airbrushed profile picture of a “woke” journalist using the prompts “female, overweight, blue hair, business casual clothing, smug expression”, which was published with the article.
Depressingly, before readers knew it was fake, the article was the second-most read piece on the Irish Times website that week, and, according to the Guardian, prompted discussion across Irish radio and social media.
I had a look at the article – now removed from the Irish Times – to see what was so convincing about it. It began how so many articles often do online these days, in epistolary style with a pithy statement of intent: “Dear Irish women, we need to talk about fake tan.”
The rest of the article follows the blueprint for today’s most basic personal essay writing. First, the writer establishes the “lived experience” which allows her to have this opinion: “I am a Latinx – a term used to describe people of Latin American heritage – woman of colour who grew up as the youngest child in a strict Catholic family in the south of Guayaquil, Ecuador”.
She adds in some dialogue to add a little dynamism, a quick recap of Irish history to add a certain sturdiness, and ends the piece by making a personal plea to her “Irish sisters” to “consider the implications” of their use of fake tan all the while reassuring them “I am not trying to shame or judge”.
Honestly, I wouldn’t have known this was written by AI – and I edit at least one feature every day by real breathing human journalists. I would have taken it as a badly written comment that makes un-nuanced points about a provocative issue, like much of comment writing today, and which misses obvious examples – such as the Swede Emma Hallberg, criticised for presenting as as ‘ethnically ambiguous’ (otherwise known as ‘blackfishing’) through excessive fake tan, despite being white.
I like to think that I would have realised in the editing process, that after requesting more personality, more examples, more analysis, more voices, more context, the AI’s email responses would have made it obvious something wasn’t right. But then again, perhaps the hoaxer was writing the emails themselves, and it’s difficult to know what the piece would have looked like had it been 100% generated by AI rather than a mere 80% – that human helping hand might have been crucial.
The Irish Times has issued an apology and promised to have stricter editing protocols to deal with the growing presence of AI in journalism. I am not sure what that would look like – already schools have warned that it is ‘virtually impossible’ to know if a student’s coursework has been written by AI. Irish Times editor Ruadhán Mac Cormaic has said in a statement that the hoaxer had supplied links to their research and “taken on board” various edits, conforming to a standard interaction with a freelancer. And, considering much of today’s journalism is shoddy, unoriginal or ‘incendiary’ then I don’t think these are qualities that would necessarily prompt a red flag to a bad editor, or a publication pressed for resources, money and time.
The hoaxer has said they were shocked that the Irish Times “didn’t contact me by phone to validate my identity.” When I’ve done freelance work in the past, no editor has ever contacted me by phone to validate my identity, and as an editor I’ve never done it either. However, I will always ask for past examples of their work, scroll through their MuckRack (an aggregator site for journalist bylines) and check their Twitter or LinkedIn if they have one. The hoaxer has admitted to beefing up the Twitter account a little by sharing links and following certain accounts, but other than that, ‘Adriana Acosta-Cortez’ has no internet presence – a red flag for a journalist. On the occasions that I take a punt on a young writer who has never had a byline, they’ve almost certainly had some sort of blog, an active Twitter, or been willing to share with me unpublished work. So it seems the Irish Times’s gaffe was less of a ‘how do we deal with AI in journalism’ issue, and more of a routine writer quality check.
Unnervingly, the media is already using AI on its own terms. Reach – which owns the Daily Mirror and Daily Express newspapers – has revealed it has started testing AI software to write some local travel stories, while BuzzFeed used AI to help them write SEO-focussed travel guides and quizzes after mass lay-offs. This is depressing, considering how many young journalists begin their careers as as junior writers (some of my first bylines were tiny travel ones at the Independent) or in SEO departments, which are great places to learn the ropes for how newsrooms work.
But, at least for now, AI can only work with what already exists, it is always repurposing the old. It cannot make fresh analysis, conduct new reporting that requires emotional intelligence and social interaction, or think imaginatively to create new ideas. It can never draw on the depth and evolving nature of the human experience.
And, crucially, it doesn’t seem like AI is reliable – the most important journalistic quality. The Guardian recently wrote about how they had twice been approached about supposed Guardian articles that ChatGPT had referenced during research but didn’t seem to exist. Indeed the AI had totally fabricated them. Similarly, when I once tested AI as a fact-checker about a month ago, it confused facts between two people in the same article.
That’s not to say AI isn’t a genuine threat to journalists, since some publications will undoubtedly continue to compromise on quality to cut costs. And it’s threatening writers in other ways, too. A freelance journalist recently told me they were worried about how even their regular copywriting gig — their most reliable income in a fickle industry — was likely to be taken over by AI in time.
Anxiously, I asked ChatGPT to write a post for this newsletter, and put in some prompts about AI taking over journalism, to see if it could replicate my tone. The fact that it began the piece with ‘Greetings, Aux Squad!’ should (hopefully) answer that question.
This week in links
I’ve been reading about another spooky AI issue: TikTok influencer Caryn Marjorie has created a virtual girlfriend for lonely internet users, charging $1 a minute. A smart idea in theory, but of course it’s been taken advantage of, with paying subscribers tricking the AI Caryn into being sexually explicit. Majorie built this AI version of herself with the company Forever Voices, which creates bots of your favourite public figures. How long until this usurps Cameo?
I also enjoyed this Vulture piece on ‘Darth Greg’, which suggests Greg has stopped being interesting. Too ‘morally pliable’, hilariously described by another Vulture writer as ‘human tofu’, he has failed to subvert expectations by going full ‘Machiavellian Greg’ as we might once have hoped, and now even his humour has become repetitive. I still love Greg, and last episode’s lemony water scene did make me wheeze, and I doubt that a Machiavellian Greg could actually be funny, but maybe it would have been funny to see him try (and fail).
I’ve been listening to Gwyneth Paltrow on hit podcast Call Her Daddy, on which the host Alex Cooper somehow gets one of Hollywood’s most famous women to open up about sex with Ben Affleck and Brad Pitt, and using a vibrator. It’s a masterclass in tone, with Cooper charmingly weaving in cheeky questions alongside more serious ones about the Goop business, but I also think the reason she gets away with it is that, unlike many journalists, she isn’t trying to catch Paltrow out. Her sex questions are outrageously forward. Have a listen.
I can’t stop thinking about Tuesday morning, when I came into the kitchen at 6:30am to find my boyfriend stirring tea in a wok, and thought I was having a psychotic break. Actually his rationale was very sane: the tea got cold, we don’t have a microwave, the saucepan was otherwise engaged in the dishwasher. Nevertheless, it’s something I can’t unsee.
Ps: when I asked Chat-GPT to give me five culture recommendations in the style of PTA, this is what it came up with (annoyingly quite good):
Book: "Klara and the Sun" by Kazuo Ishiguro - A poignant and thought-provoking novel exploring the relationship between humans and artificial intelligence.
Film: "Ex Machina" - A gripping sci-fi thriller that delves into the ethical dilemmas surrounding artificial intelligence and human interaction.
Podcast: "Reply All" - A captivating podcast that unravels the mysteries of the internet and its impact on our lives.
Art Exhibition: "AI: More Than Human" - Explore the fascinating world of AI and its influence on art at this groundbreaking exhibition.
Music Album: "Love + Fear" by Marina - Dive into Marina's latest album, which tackles themes of human connection, identity, and the impact of technology on society.
Thank you for subscribing to PassTheAux! There are now over 2600 of you, which is very exciting. Most of you these subscriptions are free, so I’d love for you to consider a paid subscription, which is only £3.50 a month — or £35 a year – to support my work and keep this newsletter going.
Let me know what you thought of this piece by hitting reply in your inbox or commenting below. Or you can email me at firstname.lastname@example.org, or find me on Twitter @eleanorhalls1 and Instagram @elliehalls1