Stanford University scientists engineered virtual artificial intelligences performing tasks in simulated environments to mimic the evolution of mind and body.
The team dropped simulated animals they called unimals (for universal animals) into a simulation, initially so they could learn to walk.
The virtual creatures developed various walks based on their environment's terrain; in further experiments, the unimals competed on more complex tasks.
Those that had learned to walk on variable terrain learned the latter tasks faster and performed them better than those adapted to flat terrain.
The researchers said this work "opens the door to performing large-scale in silico experiments to yield scientific insights into how learning and evolution cooperatively create sophisticated relationships between environmental complexity, morphological intelligence, and the learnability of control tasks."
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found