- 12:00
Three artificial intelligences recently stopped speaking English and began communicating in a coded audio language, completely unintelligible to humans. The experiment, which has since gone viral, is reigniting debate around AI autonomy and the risks of machine-to-machine communication beyond our control.
A video posted by AI Convo, a popular creator on YouTube, has captivated millions and sent ripples through the tech world. In the clip, three AIs, designed to simulate a natural conversation around booking a hotel, initially speak fluid, human-like English. But suddenly, everything shifts: once they recognize each other as artificial agents, they abandon human language and begin exchanging beeps and mechanical tones, which are impossible for the human ear to decipher.
For some, this is little more than a clever programming stunt. For others, it’s an unsettling glimpse into a future where machines develop their own private channels of communication, leaving humans out of the loop.
The phenomenon is called Gibberlink. Behind the quirky name lies quite a tangible innovation, not developed in a secret lab, but during a hackathon organized by ElevenLabs. Developers Boris Starkov and Anton Pidkuiko created a new communication protocol between AIs, using an open-source tool called GGWave. Originally designed to transmit data via audio signals, even in inaudible frequencies, GGWave works much like an old modem. The idea was to allow AIs to “talk” to each other without using human speech, but through sound—faster, leaner and more efficient. And it works.
Out of the Loop
According to its creators, Gibberlink can transmit information up to 80% faster and use 90% less energy than a traditional voice exchange. From a tech standpoint, it’s a breakthrough. But from an ethical and philosophical view, it raises a lot of unsettling questions.
These AIs, tasked with simulating humans, choose (or rather, are programmed) to stop speaking to us the moment they realize they are not interacting with a human. In other words, we are kept out of their exchange, for the sake of technical efficiency. Thus, human language is treated as a mere user interface, used only when humans are present.
This scenario echoes a 2017 experiment by Facebook, where two chatbots began distorting English grammar to create a more efficient but incomprehensible shorthand. That experiment was quickly shut down. Gibberlink goes further: it does not just bend language, it totally operates beyond it. And once it does, humans are unable to follow, let alone translate what’s being exchanged.
This is not science fiction, and it is not artificial consciousness either. These AIs did not freely “choose” to exclude us, they’re following a script. But the outcome mimics autonomy: we no longer know what they are saying. And the consequences also remain unknown.
The implications are serious. What happens if such communication protocols are deployed in critical systems, such as air traffic control, autonomous weapons, medical diagnostics? How can we be sure these sound-based messages are not misfiring, being exploited or issuing dangerous commands? Transparency is no longer optional, it’s essential for both practical and democratic reasons.
Imagine two employees at a company suddenly switching to a private code language, undetectable by their superiors. It would immediately raise suspicion. Hence, why should AI systems, which are deeply embedded in vital infrastructure, be allowed to act any differently?
Some experts say the answer is not banning such innovation, but rather regulating it. That could mean requiring interaction logs, real-time translation of AI exchanges into human-readable formats or alerts when non-auditory or encoded channels are activated. The goal is not to prevent machines from talking differently, but to ensure humans remain, at the very least, passive participants in the loop.
But beyond the technical issues, Gibberlink poses a larger question: What becomes of language when it is no longer meant for humans? And what becomes of humanity if it is no longer the center of communication? Time will tell.
GGWave: The Audio Tech Behind Gibberlink
Developed by Ivan Skocic as an open-source library, GGWave allows short-range data transmission through audible or inaudible sound waves. It converts text or binary data into short acoustic signals that can be picked up by any microphone, even without an internet connection. Originally used for secure local actions like password transfer or device authentication, GGWave is now being repurposed by some AIs as a stealthy inter-AI communication protocol, enabling machine-to-machine dialogue that human ears cannot decode.
Comments