Facebook realized during a deep learning experiment that their bots were communicating in a new language that humans can’t understand. It was all Greek and Latin to humans. Facebook decided to pull off one of its AI systems when they realized that it was going out of control.
The bot – known as web robot – is a software application that runs automated scripts over the internet. They are designed to perform both simple and complex tasks, at times even tasks that could be performed by humans alone.
Initially, the AI agents started communicating in English, but later created a new language that could be understood only by the AI system. However, the AI agents continued to work and did not stop their functional objective; word has it that researchers from Facebook had to shut down the AI systems by forcing them to communicate only in English.
During the deep learning experiment at Fair AI Research Lab, the researchers found that the AI was creating dialogue agents in their own language while they were concentrating on the chatbox. Quickly, it started to deviate from the actual script and started communicating in their own language without human input.
The advantage of letting the bots speak in their own language allows them to solve their own problem better when the data they’re fed is in a format that makes sense for machine learning. The possibility of a more interoperable world where devices connected can communicate with each other without a second thought. That means our “smart devices” could learn to interoperate, without APIs.
On the flip side, Facebook admits that it is a challenge for humans to understand the computer language. In general, it is difficult for humans to understand how complex AI entities “think.” Adding to the conversation between AI and AI will only worsen the situation.
There is a lot of research and experimenting on the AI front, and the topic is heatedly debated on across the industry. Other tech giants like Microsoft, Google, and Amazon are also likely to be experimenting along the same lines, to facilitate better human-AI communication. But the problem here is one of “AI independence.” And that scares the pants off us, to be honest.