A team of researchers from the Université de Genève has made significant strides in the artificial intelligence (AI) field, modeling an artificial neural network capable of not only learning and performing tasks, but providing a linguistic description of them to another AI. A feat previously thought to be the sole purview of human intelligence, this development not only marks a milestone in machine learning, but has profound implications for future advancements, especially in robotics.
Humans have a unique ability to perform new tasks based on verbal or written instructions alone and can also describe it to others to reproduce the task. Until now, AI systems required numerous trials and error reinforcements to learn new tasks—and even then, they lacked the ability to convey their knowledge to their counterparts.
In an effort to recreate this human cognitive prowess, researchers have turned their attention to the world of "Natural Language Processing," a sub-domain of AI that deals with machines understanding and responding to vocal or textual data. This domain utilizes artificial neural networks, inspired by our biological neurons and their method of communicating through electrical signals. While not fully understood, advancements in this field promise to bring us closer to achieving this cognitive triumph.
Despite AI’s conversational agents' ability to integrate linguistic information to generate text or visuals, these AI tools have yet to translate verbal or written instructions into sensorimotor actions, let alone explain it to another AI for replication. This is where a team led by a full professor at the Department of Basic Neurosciences, Université de Genève, Alexandre Pouget, breaks new ground.
The team devised an artificial neural model capable of the cognitive prowess previously mentioned, albeit with prior training. They began with an existing AI model, S-Bert, pre-conditioned to understand language, and coupled it with a simple network of a few thousand neurons. This artificial neural network underwent training to simulate the roles of two key areas in our brain responsible for perceiving, interpreting, and producing language.
After training, the researchers used written instructions in English to command the AI. The AI system was able to execute several tasks like indicating stimulus locations, responding to stimuli direction, and differentiating brightness contrasts between visual stimuli. It then was able to instruct a second AI to reproduce these tasks, marking the first time AI systems have been able to communicate in a purely linguistic manner.
While the present model is novel, the future potential for AI interactions is enormous. This discovery contributes to the broader understanding of the relationship between language and behavior, opening new doors for advancement in the robotics industry. Given the exponential leaps in the power of AI, the team's neural model serves as a promising starting point for developing more complex systems that could be used in humanoid robots that can understand humans and each other.
Disclaimer: The above article was written with the assistance of AI. The original sources can be found on ScienceDaily.
