"New Artificial Intelligence Inspired by Human Brain Function"

2020-02-02

In recent years, artificial intelligence (AI) has driven the development of high-performance machine learning technologies. However, these technologies are typically applied on a task-by-task basis, which means that an intelligent agent trained for one task will perform poorly on other tasks, even if they are very similar. To overcome this problem, researchers at the University of Liège (ULiège) have developed a new algorithm based on biological mechanisms, called neural modulation. This algorithm can create intelligent agents that can perform tasks they have not encountered during training. This novel and remarkable result will be published in this week's issue of PLOS ONE. Despite significant advancements in the field of AI in recent years, we are still far from human-level intelligence. Indeed, while current AI technologies allow computer agents to perform certain tasks better, they often disappoint when faced with conditions that differ even slightly from those they have experienced. Humans are able to adapt to new situations very effectively by acquiring and practicing new skills. For example, a child who learns to walk in the living room will quickly learn to walk in the garden as well. In this case, learning to walk is related to synaptic plasticity, which changes the connections between neurons. The rapid adaptation of the walking skill learned in the living room to walking in the garden is related to neural modulation. Neural modulation modifies the input and output characteristics of neurons through chemical neuromodulators. Synaptic plasticity is the foundation of all the latest advancements in artificial intelligence. However, until now, there has been no scientific work proposing a method to incorporate neural modulation mechanisms into artificial neural networks. This rather exceptional result, described in this week's issue of PLOS ONE, is the fruitful collaboration between neuroscientists and AI researchers at the University of Liège. The researchers developed an intelligent algorithm: two PhD students, Nicolas Vecoven and Antoine Wehenkel, and two professors, Damien Ernst (an AI expert) and Guillaume Drion (a neuroscientist). These ULiège researchers have developed a completely original artificial neural network architecture that introduces interactions between two subnetworks. The first subnetwork considers all contextual information related to the task at hand and modularizes the second subnetwork in a brain-like manner using chemical neuromodulators. Due to neural modulation, the second subnetwork determines the actions to be performed by the intelligent agent, allowing it to quickly adapt to the current task. This enables the agent to effectively solve new tasks. This innovative architecture has been successfully tested on navigation problems that require adaptation. Specifically, the trained agent is able to move towards a target while avoiding obstacles, even in situations where its movement is greatly disrupted by changes in wind direction. Professor Damien Ernst says, "The novelty of this research lies in the application of cognitive mechanisms, first discovered in the field of neuroscience, in a multitask environment for the first time."

RECENT AI NEWS

RECENT AI TOOLS