New AI Models Can Talk to and Teach Each Other With Limited Human Intervention

What to know

  • Scientists have developed AI models that can talk to each other and pass on skills.
  • The transfer of skills between the two AI models took place with limited human input.
  • It’s the first time that AI models have used human-like communication skills for transfer of knowledge.

AI systems are evolving at a rapid pace. A new paper published in the journal Nature highlights the development of an AI network that can not only learn and carry out tasks relying only on written instructions, but also teach another ‘sister’ AI that has had no such instructions or experience to perform the same task. 

The scientists behind the development of the AI network mentioned that the AI used human-like communication skills, or natural language processing, to teach its sister AI skills it had no prior training on. Starting with a pre-trained model called “S-Bert” and then connecting it to a smaller AI network, the transfer of skills was achieved with minimal human intervention.  

This is the first time that two different AIs have communicated with each other purely linguistically, which is an exciting development in the AI sector.