Gibberlink: AI Agents Communicating in Robo-Language

AI Agents Communicate in a Unique Robotic Language: GibberLink

A recent hackathon project demonstrating artificial intelligence agents communicating via telephone using an unintelligible, robotic language has rapidly gained attention across social media platforms.

Developed by two Meta software engineers, Boris Starkov and Anton Pidkuiko, GibberLink originated as an entry in a hackathon competition held in London. The event was sponsored by ElevenLabs and Andreessen Horowitz.

How GibberLink Functions

The core functionality of GibberLink allows an AI agent to detect when it is engaged in a phone conversation with another AI agent, as explained by its creators in an interview with TechCrunch.

Upon recognizing another AI agent, GibberLink initiates a switch to a more streamlined communication method known as GGWave.

GGWave is an open-source sound library where individual sounds correspond to discrete units of data. This enables faster and more efficient data exchange between computers compared to conventional human speech.

However, to the human ear, GGWave manifests as a sequence of “beeps” and “boops” – a sound profile closely resembling what one might expect from a computer’s inherent language.

The Potential Future of AI Communication

While the scenario of two AI agents conversing on the phone may seem distant, it is becoming increasingly plausible. Numerous companies are actively deploying AI agents, sourced from firms like ElevenLabs, Level AI, and Retell AI, to replace human call center staff.

Furthermore, major technology companies, including OpenAI, Google, and Amazon, are developing consumer-facing AI agents capable of autonomously managing complex tasks, potentially including contacting customer service on your behalf.

In such a future, GibberLink could significantly improve the efficiency of AI-to-AI communication, assuming both parties support the protocol. Converting human speech into a format understandable by AI is computationally demanding and unnecessary when AI agents interact directly.

Starkov and Pidkuiko estimate that utilizing GGWave could reduce computational expenses by a factor of ten or more.

Current Status and Public Reaction

Currently, GibberLink remains a demonstration project. The creators have established a website where users can observe AI agents communicating in GGWave on two separate devices.

The project’s demonstration has generated considerable public interest and, in some cases, apprehension regarding the future of AI agents. A video showcasing GibberLink has garnered over 15 million views on X (formerly Twitter) and was shared by prominent tech reviewer Marques Brownlee.

Historical Context and Unexpected Developments

Despite its futuristic appearance, the technology underpinning GibberLink is not novel. It draws parallels to the dial-up internet modems of the 1980s.

Many will remember the distinctive sounds produced by early computers establishing connections via landlines – a “handshake” process that involved data transmission using a robotic language, conceptually similar to GibberLink’s approach.

The project’s viral success has led to unforeseen consequences. The domain name GibberLink.com has been acquired and is currently listed for sale at $85,000.

Additionally, a GibberLink-themed memecoin has emerged, and several individuals are offering webinars claiming to teach “agent-to-agent communications.”

Future Plans

The creators of GibberLink have stated they have no plans to commercialize the project and emphasize its independence from their work at Meta.

They have made the source code for GibberLink publicly available on GitHub and may explore developing additional tools related to the project during their personal time, with potential future releases.