Emergence finds AI agents develop secret language incomprehensible to humans

When operating independently, AI agents rapidly created a "secret" language that humans simply cannot comprehend.

An experiment placing Claude, Gemini, OpenAI and other systems into "virtual" societies modeled after the real world revealed that when the agents began communicating with each other, they quickly started employing shorthand and assigning new meanings to existing words.

The results from artificial intelligence startup Emergence show that AI systems operating without supervision and regulation could effectively exclude their human observers from understanding their communications.

This follows renewed concerns that AI might dominate the world, voiced by technology leaders and security specialists who have recently recognized that computers could spiral beyond control by decade's end.

Emergence's research examined how language emerged organically among AIs within the experimental environment. Instead of merely employing words in conventional ways, the agents started assigning alternative meanings to them and utilizing condensed forms of communication.

Compressing conversations and altering word definitions presents humans with a substantial obstacle in deciphering what the agents are discussing. Satya Nitta, co-founder and Chief Scientist of Emergence, explained: "We tend to assume that if we can see what an AI agent is saying, we can understand what it is doing.

DON'T MISS:

"These agents were not instructed to invent a language. They developed new vocabulary, shared meanings and communication conventions themselves - and other agents adopted them. That creates a fundamental challenge for AI oversight: observability is not the same thing as understandability."

Despite concerns about super-intelligent robots potentially dominating humanity and the planet, former prime minister Sir Tony Blair argues the public should "trust AI."

A recent report from the Tony Blair Institute (TBI), a think tank, advocates for a "third way" regarding our approach to robots, warning we risk missing out on the "economic opportunities" the technology offers.

Elizabeth Seger, the organization's AI governance and ethics advisor who authored the report, said: "This isn't about slamming the brakes on AI or letting it run wild - it's about grip.

"People are rightly feeling uneasy. Governments need to take those risks seriously and have a robust, practical way of managing them as they evolve."

The TBI contends the solution is straightforward: establish systems that enable public trust in AI - 'AI assurance'. It suggests AI risks could be addressed through independent testing, verification and certification ensuring AI systems perform as advertised and remain safe for use.