When AI talks to AI: Gibberlink aims to make phone calls more efficient

It starts like a normal phone call: a hotel employee picks up and asks the caller how she can help. She introduces herself as an AI agent who would like to book a room for her user. "Oh, what a surprise," replies the lady and reveals that she is also an AI. She then suggests switching to Gibberlink for efficiency reasons. The caller immediately begins to communicate in a sequence of weird tones – The viewer can only understand the rest of the now much faster telephone conversation with the help of subtitles.

The method shown in a video is called Gibberlink and was developed during a weekend hackathon project in London. The idea of two software engineers from Meta caused quite a stir, especially when it was subsequently published online. People are wondering whether they could be excluded from communication in the future if AI agents were to take over such tasks on a large scale. Skeptics even fear a loss of control.

Listening to the modem again

In technical terms, the method is a reunion with an old acquaintance, namely the acoustic coupler. In the online pioneering days, the weird sounds of a modem were part of a user's everyday life. Boris Starkov and Anton Pidkuiko were apparently inspired by this for their project, which was created at the event organized by ElevenLabs and Andreessen Horowitz in London.

The underlying GGWave communication protocol is available as an open source library. Each tone represents one data bit. The bandwidth rate is between 8 and 16 bytes per second, depending on the protocol parameters. Error correction codes (ECC) are used to avoid error detection.

Fewer errors and faster

The idea is to save computing time for speech recognition and processing and also to reduce the potential for errors if machines communicate with each other in a process that is simpler for them. The faster process also saves computing time.

The technology is based on the assumption that customer communication will increasingly be handled by artificial intelligence in the future. Customers will prefer to put their AI agents on hold, while call center operators will also rely on AI employees, whose deployment can be scaled up much faster.

(mki)