NEWS  AI  December 5, 2023  Reading time: 2 Minute(s)

Max (RS editor)

DeepMind research team at Google, has introduced an AI system showcasing remarkable social learning capabilities. This groundbreaking development, presents an unconventional approach where an AI application learns new skills in virtual environments by replicating the actions of an implanted "expert". Departing from traditional data-driven methods, this innovation seeks to emulate the human learning process for more efficient knowledge acquisition.

Conventional AI models, such as ChatGPT, rely heavily on vast datasets, but the industry recognizes the inefficiencies inherent in this approach. Researchers are actively exploring alternatives, with a significant focus on mimicking human learning processes, which often require fewer examples.

DeepMind's strategy involves mirroring human learning in a virtual setting called GoalCycle3D. In this simulated environment, AI agents traverse uneven terrain, navigating obstacles and spheres. Unlike conventional AI models, these agents receive learning modules without prior information about the virtual world. Their learning process hinges on reinforcement learning, gaining rewards as they repeatedly navigate through similar virtual scenarios.

To accelerate the learning curve, researchers introduced expert agents already familiar with optimal routes through the virtual world. Non-expert agents quickly grasp and emulate the expert's behavior, resulting in an expedited learning process. Observations reveal that agents trained with an expert not only learn faster but also exhibit improved navigation skills in subsequent trials, even in the expert's absence—an evident display of social learning.

The success of this social learning experiment opens new avenues for AI development. By integrating social learning mechanisms, AI systems could potentially acquire skills more efficiently and dynamically adapt. The study underscores the importance of memory modules in retaining learned behaviors, enabling agents to independently apply acquired skills.

DeepMind's achievement in crafting an AI system adept at social learning in virtual environments signifies a crucial moment in AI research. This breakthrough not only offers insights into more effective AI learning methods but also envisions a future where machines can learn and adapt through social interactions, mirroring human cognitive processes. As AI continues to evolve, the integration of social learning mechanisms may pave the way for more versatile and adaptive artificial intelligence systems.

 COVER IMAGE BY RAWPIXEL.COM ON FREEPIK / REVIEW SPACE | GOOGLE