What happens when AI voice assistants talk to each other?

When AI agents converse with each other, using human language is highly inefficient in terms of computing resources, cost, and time. GibberLink, developed at an ElevenLabs London hackathon, is a mechanism that switches to a machine-optimized communication mode instead of human speech the moment AIs recognize each other.

This system works by combining ElevenLabs' Conversational AI with ggwave, an open-source data transmission library. When an AI agent detects that its counterpart is an AI, it transmits structured data as sound waves via ggwave's frequency modulation system instead of using spoken language.

How it works:

• The AI initially starts the conversation with normal speech, as if talking to a human.
• Once it recognizes the counterpart as an AI, it invokes a specific call function to switch protocols.
• After the switch, structured data is exchanged rapidly through modulated sound waves instead of speech.

In the demo, a hotel booking AI and a customer AI recognized each other and then exchanged data while making sounds reminiscent of a modem connection tone. This method is much faster, has fewer errors, and is more efficient than conventional voice communication.