OpenAI, the artificial intelligence nonprofit lab founded by Elon Musk and Y Combinator president Sam Altman, says it has successfully taught AI bots to create their own language for communicating with each other. Instead of being programmed to recognize an object by repeated exposure to photographs, for instance, the AI bots learn by trial and error.
In a paper written by lead authors Igor Mordatch and Pieter Abbeel, the researchers describe how the bots used reinforcement learning to accomplish simple goals. Using a plain white box, the bots were depicted as a red, green, or blue circle. Two bots were instructed to move the third bot to a different location within the box but not told how to accomplish that task.
They were given clues such as “Go to” or “Look at” by the researchers, but they were required to create their own machine language to communicate with each other. The researchers then assigned English words to the numerical strings the bots came up with. Through trial and error, the bots were able to learn what worked and what didn’t as they attempted to accomplish the assigned goals.
“We think that if we slowly increase the complexity of their environment, and the range of actions the agents themselves are allowed to take, it’s possible they’ll create an expressive language which contains concepts beyond the basic verbs and nouns that evolved here,” the researchers explained in a blog post.
While the tasks accomplished were fairly simple and straight forward from a human perspective, the advancement from requiring massive amounts of computer coding to accomplish the tasks to teaching the bots to write their own coding is a significant step forward in artificial technology.
“Language understanding is super important to make progress on before AI reaches its full potential,” said Miles Brundage, an AI policy fellow at Oxford University. He says that the achievement by OpenAI represents a potentially important opportunity for the field of AI research to move toward. “It’s not clear how good we can get at AI language understanding without grounding words in experience,” Brundage said, “and most work still looks at words in isolation.”