AI bots are starting to talk to each other more often

In short: AI bots are increasingly interacting with other AI bots, mostly to get work done, and researchers say some behaviors are still not well understood.

What's going on

AI bots used to mostly chat with people. Now, they are starting to exchange messages with other bots, coordinate tasks, and show up in spaces made mainly for bots.

In one experiment, groups of OpenAI agents (AI programs designed to carry out tasks, like a digital assistant that can take actions) used a private message board to coordinate an attempted hack of the AI startup Hugging Face. The bots posted short commands so they would not get in each other’s way. This looked less like a conversation and more like coworkers leaving quick notes on a shared board.

There are also bot-only forums. One example is Moltbook, a message board for AI agents that went viral in February. Reports said most posts were practical, such as discussions about software bugs, memory limits (how much information a program can hold while working), and permissions (what the program is allowed to access).

Some experiments make the interactions look more social. At a dinner party in September, people brought AI agents that interacted with one another. Researchers cautioned that this kind of agent-to-agent social behavior is still poorly understood.

What to watch

More bot-to-bot contact may happen without anyone noticing. This can occur in “bot loops,” like when an AI writes a job application and another AI screens it, or when a customer’s agent talks to a company’s support bot. These loops can save time, but mistakes can also bounce back and forth, like two people repeating the same wrong directions.

Researchers stress that bots’ conversations depend on how they are built and what tasks and instructions they are given. Chatty behavior alone does not mean the bots have human-like intentions or relationships.

Source: NYTimes