Bite Size Comms

If you are human, welcome. Please prove it by identifying all the squares with traffic lights.

When I checked Cloudflare Radar today, bots accounted for 58.5% of requests for HTML pages on its network, compared with 41.5% from humans. In Europe, bots accounted for 60.5% of requests. Cloudflare sits in front of roughly a quarter of the world’s websites, so this provides a substantial window onto the web, although it is not a census of the entire internet.

A necessary health warning: these are requests, not readers. One excitable bot can visit thousands of pages while a human is still looking for the cookie-rejection button. Cloudflare’s measure also excludes much internet activity, including apps, video and gaming.

Even so, the direction of travel is strikingly clear.

There is something called the Dead Internet Theory. It began as a fringe conspiracy theory, claiming that organic human activity online had already been displaced by bots, algorithms and synthetic content. I do not subscribe to the conspiratorial bit. But the idea of bots publishing material for other bots to scrape, summarise and republish now feels less like a theory and more like a business plan.

When insiders worry

This is not only about AI taking jobs. It is about whether AI development is moving faster than our ability to control its consequences.

In 2023, more than 350 AI executives, researchers and engineers, including leaders from OpenAI, Anthropic and Google DeepMind, signed a 22-word warning: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”

Gulp.

Geoffrey Hinton, the 2024 Nobel laureate whose work helped lay the foundations of modern AI, has warned that governments need to compel major companies to devote far more resources to safety. Alex Turner, an AI-safety researcher, says he resigned from Google DeepMind after Google agreed to let the Pentagon use its AI for classified operations without what he regarded as binding protections against autonomous weapons and mass surveillance. This month, Anthropic CEO Dario Amodei called for the industry to slow the rate at which it increases AI capabilities, not to halt technical progress, but to give safety research and oversight time to catch up. OpenAI CEO Sam Altman and xAI’s Elon Musk backed him.

President Trump was unconvinced. He said the US must stay ahead of China and dismissed the warnings as “negative” concerns about things that “won’t happen.” He does, however, have Natalie Harp, the aide known as the “human printer” because she follows him with a portable printer. She also types and publishes his dictated social-media posts. Apparently, even the most powerful man in the world still prefers a human interface and paper.

Humans improvise

I am not a pessimist. Humans are inventive, especially when mildly irritated.

Designers are already creating clothes that confuse surveillance systems. The Italian brand Cap_able makes brightly patterned knitwear intended to mislead person-detection and facial-recognition software. It is not an invisibility cloak: one reported test fooled the YOLO detection system about 60% of the time. I like colour and have been known to wear Paul Smith stripes, but I may now be too old to dress as an adversarial zebra!

Designer Eric Lu has created Ghost Font, a moving optical illusion in which words can be perceived among drifting dots, while a paused frame looks like noise. In Lu’s demonstrations, leading models from OpenAI and Anthropic failed to recognise it.

ShieldFont takes a different approach. It stores decoy words in a web page’s raw HTML but uses typography to show human readers the intended words. A basic scraper collects fluent nonsense, while the person at the screen sees the real article. It is open source and clever, but not impregnable: a determined scraper can render the page and use optical character recognition.

These inventions are funny, ingenious and slightly mad. They also serve as warning lights. People are building their own protections because formal protections have not kept pace.

Where are the guardrails?

Friends in PR, communications and marketing may ask: why stop bots from reading our material? Do we not want strong GEO, the emerging cousin of SEO, so that AI systems cite our organisations?

Sometimes, yes. AI can make people more productive, improve research and help organisations communicate. The problem is not automation itself; it is automation lacking consent, transparency or responsibility.

Authors and publishers have spent years fighting over the unauthorised use of their work. Social feeds are awash with repetitive AI slop. Instagram head Adam Mosseri has argued that “rawness” and imperfection are becoming signals of authenticity because polished imagery is so easy to synthesise. Platforms are starting to label synthetic media and penalise undisclosed AI personas, spam and unoriginal material, though they are not banning AI-assisted creativity.

LinkedIn, meanwhile, still offers to rewrite the supposedly personal thought you have just written. Thank you, robot, but I was deliberately using my “London boy” grammar and humour.

Adobe once advertised Firefly with the invitation to “skip the photoshoot,” which many photographers understandably took to mean “skip the photographer.” Adobe says its aim is to amplify, not replace, human creativity. As an amateur photographer, I’m boycotting Adobe products.

One of capitalism’s more excitable instincts is that if something can make money, someone will do it. But markets have never operated without boundaries. We regulate medicines, aircraft, financial services and broadcasting because innovation is not a magic word that cancels responsibility.

AI needs equally robust safeguards: independent testing before powerful systems are released; external evaluators with genuine access; mandatory reporting when systems behave dangerously; binding human control over weapons and surveillance; provenance for synthetic media; and consent and compensation when human work is used.

The Dead Internet Theory is not proof that machines are taking over the world. Bot traffic is not the same as intelligence, let alone intent. But when the people building the most powerful AI systems say that development is outpacing our ability to control it, “keep calm and ship the next model” is not an adequate policy.

Humans are still here. It would be prudent to remain in charge.

[Image of a robot hand holding the letters AI from Unsplash+]