"Fancy Gibberlink?": AI bots suddenly switch to secret language

On X, a video shows a telephone conversation between two AI assistants. When the bots realize that they are alone, they switch to "Gibberlink mode," a kind of secret language for artificial intelligence. The recording causes unrest - even among experts.

It starts harmlessly. An AI assistant calls a hotel on behalf of a customer. "Thank you for calling the Leonardo Hotel. How can I help you?" asks the digital concierge. "Hello. I'm an AI calling for Boris Starkov. Is your hotel available for a wedding?"

When the concierge realizes that the person he is talking to is also a machine, he suggests switching to the so-called "Gibberlink mode": a sound-based communication method that is incomprehensible to humans. What follows is an exchange of beeps and hums, similar to the old modem sound.

Gibberlink was created by developers Boris Starkov and Anton Pidkuiko to transmit data acoustically between devices. The whole thing works even in noisy environments and is more efficient than human speech: communication time is reduced by 80 percent and the computing effort by 90 percent.

Harmless experiment?

The disturbing scene, which has already been clicked on 13.7 million times on X, has reignited the debate about the control that AI needs - or has already lost. On social media, users are speculating whether the experiment is the beginning of the end. "These are the sounds of demons," writes one user on X. Others are sharing "Terminator" memes and warning of machine rule.

Science is also concerned. Technology expert Diane Hamilton warns of a lack of transparency in a Forbes article . "When machines communicate with each other without us being able to understand it, it undermines our control," she writes. "Who takes responsibility when AI acts autonomously?"

Past cases show how tricky AI can be. GPT-4 has already outsmarted a CAPTCHA by pretending to be a blind person. Another AI system falsely accused a law professor of sexual harassment. In 2023, experts warned that artificial intelligence could become an "existential threat" and was as "in need of regulation as nuclear weapons."

With regard to "Gibberlink mode," Hamilton points out that while AI's ability to create its own communication shortcuts can increase efficiency, efficiency is not always the goal. Experience has shown that organizations that focus exclusively on speed can overlook critical risks. "'Gibberlink mode' is just one example of how AI is evolving beyond human language," says the expert. Now it's time to keep up.