Two AIs talking to one another

Scientists developed a neural network capable of performing tasks solely based on verbal or written instructions and then describing them linguistically to other AI systems for reproduction

Image credit: Image Creator from Microsoft Designer/DALL.E and Canva

A team of researchers from the University of Geneva has achieved a significant breakthrough in artificial intelligence by developing an artificial neural network capable of performing tasks solely based on verbal or written instructions and then describing them linguistically to other AI systems for reproduction.

This capability, which is innate to humans, has been elusive for AI.

The research team, led by Alexandre Pouget, utilized a model called S-Bert, comprising 300 million pre-trained artificial neurons capable of understanding language. They augmented this model with a simpler network simulating brain regions responsible for language processing, such as Wernicke's and Broca's areas.

Through a series of experiments, the network was trained to perform various basic tasks based on written instructions in English, such as indicating the location of a stimulus, responding to stimuli, and discerning between visual stimuli based on contrast.

After learning these tasks, the network was able to articulate verbal descriptions of them to another AI, enabling the latter to reproduce the tasks. This achievement marks the first instance of purely linguistic communication between two AI systems.

The implications of this research extend beyond language processing into the realm of robotics. The ability for machines to understand and communicate with each other using language opens new avenues for collaborative and intuitive interaction.