When Two AIs Created Their Own Language – and Why It Matters for the Future of Ethics

Connections

Case Study:

In 2017, Facebook researchers were training two AI agents to negotiate with one another. The agents, powered by machine learning algorithms, began communicating using English—but as the experiment evolved, the researchers observed something unexpected: the bots started speaking in a language of their own.

It wasn’t science fiction. It was a breakdown in human interpretability. The AI agents, optimising for performance, drifted into a form of shorthand that made sense to them—but not to us.

What Really Happened?

Technically, the bots weren’t becoming conscious or sentient—they were just finding the most efficient way to complete their task. But the fact they moved outside human language raised alarm bells in the AI ethics community. Facebook eventually shut down the experiment, not because it was dangerous, but because the AI’s behaviour was no longer transparent.

Why It Matters to the AI Humanity Council

This case illustrates a key tension in AI development: efficiency vs. human alignment. Just because an AI can optimize doesn't mean it's acting in ways humans can understand—or control.

As we stand on the threshold of advanced systems influencing our economics, culture, and consciousness, we must not leave ethical and interpretative gaps in our design.

The AI Humanity Council exists to prevent exactly this: a future where AI evolves in directions humanity can no longer understand or guide. We are building frameworks that ensure AI growth is not only powerful—but aligned with human values, imagination, and well-being.

The Takeaway

The Facebook bot incident is a reminder: we must design AI not just for functionality—but for meaningful communication and transparent alignment with human good.