In the winter of 1920, the man who invented Sherlock Holmes told the world that fairies were real. Arthur Conan Doyle had seen the photographs: two girls in a Yorkshire garden, ringed by tiny winged figures with fashionable bobbed hair.

He published them in The Strand Magazine and later wrote a book defending them. The fairies were paper cut-outs copied from a children’s annual and held up with hatpins. One of the girls, Elsie Wright, finally admitted it in 1983, when she was in her eighties. Doyle went to his grave believing.

He was not a fool. He was a man of his moment. Photography was still young enough to carry the authority of a witness, and “the camera cannot lie” was something people said without irony.

William Blake, writing a century before Doyle, put the danger in four lines: “This life’s dim windows of the soul / Distorts the heavens from pole to pole / And leads you to believe a lie / When you see with, not through, the eye.”

For our parents’ generation in Bangladesh, the trusted window was the television. They came of age when a moving picture with a voice was expensive to produce and almost impossible to fake. The shortcut they learned was simple and, for decades, correct: if it moves and speaks, it happened. A clip forwarded on WhatsApp today inherits that authority without having earned it. The habit of mind was reasonable for fifty years. It was made obsolete in about eighteen months.

Bangladesh learned the price of this before anyone had heard of generative AI. On the evening of 2 March 2013, a photoshopped image appeared on a Jamaat-aligned Facebook page showing the face of Delwar Hossain Sayedee, newly convicted of war crimes, on the surface of the moon.

To reach people who did not use the internet, the picture was announced through mosque loudspeakers in several districts. Various media cells’ newsrooms took calls from readers asking whether it was true. By the next morning, crowds in Bogra were marching on police stations. Human Rights Watch later documented at least six people killed in Bogra that day, among them a clothes seller named Manzila Begum, shot from behind as she ran. The image took minutes to make. The lie did not need to be good. It only needed to reach people who had never been given a reason to doubt a picture.

Now the pictures move and talk.

In the first two weeks of January this year, ahead of the general election, the fact-checking group Dismislab documented more than 800 AI-generated videos across Facebook, YouTube and TikTok: AI police officers endorsing parties, AI children praising candidates, AI opinion polls with results decided in advance, AI uncles dispensing homespun political advice.

Under one such video, a viewer wrote: “Thank you uncle. Everyone, come forward. Vote for Sheaf of Paddy.” The uncle was not a person. Sixty per cent of the AI content on Facebook carried no label. On YouTube, the figure was 94 per cent. On TikTok, not one of the 50 videos reviewed was labelled. And where Facebook did apply its “AI info” tag, it appeared only in the main app, not on Facebook Lite or the desktop site. Facebook Lite is what runs on the cheap phones in the hands of the very people least equipped to spot a fake.

Why do older people fall hardest? The lazy answer is that they are gullible. The evidence says something more precise. Psychologists have long observed that ageing weakens what they call source monitoring: the ability to remember not only what you saw but where you saw it. Older adults keep the claim and lose the label. Nadia Brashier and Daniel Schacter, in a 2020 review of ageing and misinformation, added that as thinking becomes more effortful, people lean harder on familiarity, treating a thing as truer simply because they have met it before. An algorithmic feed that serves the same AI-generated prayer, the same miracle cure, the same weeping politician five times a day is, in effect, a machine for manufacturing familiarity.

This vulnerability predates the technology. Studying the 2016 American election, Andrew Guess and his colleagues found that Facebook users over 65 shared links from fake news sites at nearly four times the rate of those under 30, even after accounting for politics and education. No other trait predicted sharing as strongly as age. Generative AI did not create the weakness. It built a cheaper, faster weapon and aimed it there.

And then there is pride. The Malaysian couple did not ask their children because they feared being laughed at. Anyone with parents on WhatsApp knows how that fear is made. We sigh. We send the fact-check with a dismissive emoji. We turn asking into an admission. The most effective defence available to any family costs nothing, and we make it expensive.

So what is to be done?

Platforms need to act first. A label that appears on one version of an app and not another is not a label; it is an alibi. Meta, Google and TikTok should be required to display AI disclosures in Bangla, in large type, on every version of their products, including Lite apps and feature-phone browsers. India’s new rules, in force since February, require synthetic content to be clearly and prominently labelled and, where possible, permanently marked in the file itself; an earlier draft went as far as demanding a tenth of the frame. Bangladesh’s election code prohibits AI content used “with dishonest intent” but says almost nothing about enforcement. The Election Commission and the BTRC should adopt a clear labelling standard, apply it all year round, and publish which platforms comply.

Second, meet older people where they are. The fact-checkers do superb work, but their audience is young and online. The people who most need Dismislab’s findings watch the nine o’clock news. Every television channel in the country should carry a short weekly segment showing the week’s most widely shared fakes and how they were caught. The same mosque loudspeakers that once announced a face on the moon can carry a Friday reminder to check before forwarding. Imams, union parishad members and village doctors hold the trust that platforms have squandered.

Third, families. We do not need to turn our parents into forensic analysts. The tells of synthetic video change every few months and will soon vanish altogether. What endures is a habit: pause, and ask. Ask a child, a neighbour, anyone. A video that asks you to act quickly, whether by sending money, forwarding it or setting off on a journey, is the one that most deserves a delay. Make it a household rule that nobody is mocked for asking. The couple in Perak did not need a better eye. They needed one person they felt safe calling before they got in the car.

Conan Doyle died believing in fairies because the photograph was new and he was not. Our parents grew old trusting the moving image because for half a century it earned that trust. It no longer deserves it, and nobody told them. Blake’s counsel still holds. Teach them, patiently and without contempt, to see through the eye and not with it. They spent years teaching us to look both ways before crossing the road. It is our turn.

---

This article is co-authored by S M Mashrur Arafin Ayon, a Research Associate at SIPG, North South University, and Mursalin Haque, an undergraduate student at North South University.