A group of scientists have created AI that can talk to each other. The new model is capable of learning and carrying out tasks that are given to it based solely on written instructions and then passing it on to other AI, the scientists say. A study of the findings and AI’s development is featured in the journal Nature.
To test just how effective it could be, the scientists had one AI learn how to carry out the task based on its written instructions. It then allowed the AI to communicate with a “sister” AI, which was then able to perform the tasks asked of it without needing any prior training or experience. It’s a great bit of progress for AI networks, and truly showcases the potential of using natural language processing (NLP) in AI.
The entire idea behind NLP is to recreate human language within computers, allowing machines to more easily understand and reproduce written text or speech more naturally. It’s a good idea, especially with so many AI advancements coming forward, but we’re still a long way off from it being a truly viable thing. That’s why AI that can talk to each other is such a huge leap forward.
By creating AI that can communicate with other AI and pass along information, scientists have taken a lot of the initial legwork out of creating AI models. Typically, you need to train each individual model on the tasks that you want them to complete. But, with a system like this AI network that can communicate with other AI within the network, you remove the need to complete all that training beforehand.
Thanks to the work done on the model used for the network, the scientists say that they were able to feed entire written sentences to the model, and then it was able to understand them and respond more naturally. It’s a huge leap forward for AI models as a whole, and it will allow AI to work more directly with humans.
Of course, AI being able to talk to each other won’t necessarily lead to an AI agent that can reason just as well as a human. We’re still a long way from that. But, with the right prodding and development, this kind of network could open a lot of new doors for AI development. It will also, undoubtedly fuel concerns over AI overthrowing humanity, too.
