DREAM / The Prime Minister resigns and wakes up under the Parisian bridge!

 Kliko këtu për të tjera
Voxpress:

AIs have started communicating with each other in a "surreal" language, researchers are worried

2026-09-18 22:49:32Lifestyle SHKRUAR NGA REDAKSIA VOX

AIs have started communicating with each other in a "surreal"

What happens when artificial intelligences are left to communicate with each other for a long time?

A new experiment has produced an unusual result: AI systems begin to develop a way of communicating that becomes increasingly difficult for humans to understand.

Researchers from the American artificial intelligence lab Emergence have been experimenting with "companies" of autonomous AI agents, allowing the models to interact with each other.

The experiments included models developed by major technology companies in the US, China and France.

Over time, researchers noticed that communication did not necessarily remain in plain English.

Agents began to create new expressions, metaphors, technical terms, and common formulas that made sense within their communication, but were less transparent to a human overhearing the conversation.

The Guardian describes the result as a mix of poetic language, tech jargon and a surreal style that some scholars have compared to the experimental prose of James Joyce.

The phenomenon does not necessarily mean that artificial intelligence has invented a secret language or that systems are deliberately trying to hide information from humans.

In fact, something similar happens with humans.

Professional communities, social groups, and subcultures constantly create jargon, abbreviations, and phrases that people outside the group may not understand.

But in the case of AI, this creates another problem: surveillance.

If in the future autonomous agents are used to negotiate, manage systems, or make decisions together, researchers and operators must be able to clearly understand what they are communicating.

The more their language deviates from what people easily understand, the more difficult this becomes.

Language experts also see the phenomenon as an interesting example of how communication evolves within a group.

When several agents interact continuously, phrases that work well can be repeated, shortened, and gradually take on a common meaning.

So, we don't yet have a "robot language" completely disconnected from humans.

But the experiment shows that AI systems can develop forms of communication that become less intuitive to their creators.

And this is precisely where the concern lies.

In the era of chatbots, the main problem was whether a machine could convincingly speak to a human.

With autonomous agents, the new question may be quite different: what happens when machines talk to each other and humans are no longer the primary participants in the conversation?


Në përputhje me standarded JTI

Video

SHIKO KANALIN NE YOUTUBE
×