Chocolate maker Cadbury is fighting AI by injecting gibberish into training data

WTF?! One unexpected voice is sounding the alarm about the relentless inclusion of AI in the workspace: Cadbury. The chocolate company's Indian branch has launched a cheeky new advertising campaign dubbed "Make AI Mediocre Again" that aims to slow AI down by messing up training data with the "world's first server farm" built for this purpose.

In a nearly two-minute video ad, Cadbury India portrays an office worker who doesn't even have a moment to savor his 5 Star chocolate bar due to the demands of AI-driven efficiency. His boss impatiently demands work be completed in an impossibly short timeframe by "using some AI."

The camera pans out to reveal more frazzled office workers hustling between tasks. That's when a voiceover poses the provocative question: "Is AI really helping us work less like we thought, or making us work even faster? Because it's only going to get worse."

It's a valid question in a world where AI is increasingly taking over creative jobs, leading to artists having an existential crisis. Meanwhile, in the office, studies have been conflicting. While many believe that AI is directly responsible for increased efficiency, scores of others have also expressed otherwise. One survey by the Upwork Research Institute reported participants complaining about AI actually adding to their workload.

To fight all this, Cadbury and ad agency Ogilvy India have joined hands on "the world's first server farm" built specifically to generate thousands of synthetic websites filled with meaningless gibberish. The company claims this server runs "24 x 7."

The goal is to pollute the training data that AI models ingest from the internet, causing them to become confused and make endless silly errors that require constant human correction.

"The more nonsense the AIs pick up, the more mistakes they start making," explains the narrator. "And the more time we'll get to eat 5 Star and do nothing."

At the end of the ad, viewers are invited to "join this historic mission" by submitting their own ramblings to the makemediocreagain.com website to further muddy the waters for AI.

The campaign does seem a little tongue-in-cheek, playing in the space of advertising hyperbole. There are no specifics on just how many "thousands" of such websites they've created and for how long they plan to keep this up.

More importantly, the companies behind dominant AI models like ChatGPT and Claude are certainly not feeding their chatbots random, low-quality web pages from low-quality sources. They're known to use Wikipedia, books, news articles, scientific journals, and even YouTube videos as training data – most of which has been taken without payment or acknowledgment, often leading to complaints and lawsuits.

But hey, at least someone's trying to fight the barrage of AI slop on social media.