New AI algorithms compress billions of years of evolution into a few seconds.
This artificial intelligence (AI) program is not only fast, it can also run on a personal computer and design entirely new structures from scratch. The researchers say the AI-driven design algorithms they have developed can bypass "traffic jams" on the road to evolution without relying on the ideas of human designers. Simply tell the AI "we want a robot that can cross land," then press a button, and in the blink of an eye it generates a blueprint of a robot that looks very different from any animal that has walked the Earth. This process is called "immediate evolution" by the team.
The newly designed robots are small, soft and oddly shaped, and are currently made of inorganic materials. But researchers say it represents the first step in a new era of AI design tools that, like animals, can be directly applied to the real world.
The entire design process, from the invisible block with zero motion to the complete walking robot, took just 26 seconds on the laptop.
Even more surprising, the AI alone came up with the same walking scheme as nature: legs. But unlike the absolutely symmetrical design of nature, AI takes a different approach. The resulting robot has three legs, fins on its back and a flat face that is riddled with holes.
While the evolution of the legs makes sense, the holes are a strange addition. The AI punched holes in seemingly random places on the robot's body. The researchers hypothesized that the porosity would reduce weight and increase flexibility, allowing the robot to walk with its legs bent. When the researchers removed the holes, the robot could no longer walk.
According to the researchers, "When humans design robots, we tend to design them to look like familiar objects. But AI can create new possibilities and new paths forward that humans have never considered. It helps us think and imagine differently, which may help solve some of the most difficult problems we face today."
(Science and Technology Daily, Sciencedaily)
Author: Science Jun
Editor: Dina Zhao
Review: Gong Zimo
