OpenAI Researchers Develop AI Bots that Talk to Each Other
Researchers at OpenAI – a non-profit organization founded by Elon Musk and president of Y Combinator Sam Altman – announced that they have developed AI bots that can talk to each other.
OpenAI researchers (Pieter Abbeel, Igor Mordatch, Ryan Lowe, Jon Gauthie and Jack Clark) released their initial findings on how these AI bots learn to communicate on Arxiv. In a blog post entitled “Learning to Communicate,” the researchers at OpenAI researchers wrote that while the method of feeding an AI with extremely large amounts of textual data, produced a number of inventions and innovations, “it has drawbacks relating to the representational quality of the language that is learned.”
Watching two Amazon Echos chat has become somewhat of a thing and there are even live streams of the conversations, which while occasionally sounding profound, are more often than not, amount of meaningless big-data driven chatter.
The OpenAI researchers said that there is no evidence that feeding an AI with large amounts of textual data, the AI will have “a deep understanding of how that language is attached to the real world.”
The researchers said that the AI bots that they have developed invented a simple language which is “grounded” and “compositional.” Grounded is defined by the researchers as a “means that words in a language are tied to something directly experienced by a speaker in their environment.” An example for grounded is when an AI bot forms an association between the word “tree” and images or experiences of trees.
“Compositional,” on the other hand, is defined by the researchers as a “means that speakers can assemble multiple words into a sentence to represent a specific idea.” An example of compositional is when an AI bot gets another AI bot to a specific location.
Moving forward, the OpenAI researchers said they plan to increase the complexity of the AI bots’ environment and their range of actions in order for these bots to create an expressive language that goes beyond the basic verbs and nouns. As the complexity of the bots’ language increases, the researchers aim that this language can be interpreted by humans.