Can Artificial Intelligence evolve like living beings?

Charles Darwin, in his theory of evolution, argued that all living species evolved over time from one or a few common ancestors. Thus, animals have an embodied intelligence: the adaptation of their morphology to their environment has enabled them to perform complex tasks. Researchers from Stanford University wondered if an AI could evolve in the same way as a living being and their research allowed them to create one that transforms according to the complexity of their environment. The results of their study titled: “Embodied Intelligence via Learning and Evolution” have been published in the journal Nature Communications.

The objective of this study is to elucidate certain principles governing the relationships between environmental complexity, evolved morphology and the learnability of intelligent control”. Fei-Fei Li, research team member and co-director of the Standford Institute for Human-Centered AI (HAI) says:

“We are often so focused on the fact that intelligence is a function of the human brain and neurons in particular. Viewing intelligence as something that is physically embodied is a different paradigm. »

An in silico playground

For this study, the researchers created a computer-simulated playground where arthropod-like agents called “unimals” (for universal animals) will learn and be subjected to mutations and natural selection. then studied how having virtual bodies affected the evolution of the intelligence of unimals.

To successfully scale the creation of embodied agents across 3 axes of complexity: environmental, morphological, and control, the researchers designed a deep evolutionary reinforcement learning (DERL) algorithm.

To understand the evolution of embodied intelligence, the team varied not only the animal’s body shapes, but also their training environments and the tasks they performed. Study co-author Surya Ganguli, associate professor of applied physics at the School of Humanities and Sciences and associate director at HAI, said:

“And all of these variables were much more complex than in previous work, which allowed us to examine many more scientific questions than before. »

To retain unimal diversity and reduce the computational cost of these simulations, the researchers opted for a tournament-style Darwinian evolutionary scheme that allowed them to ensure that each unimal morphology had the chance to succeed and be transmitted. to the next generation.

Each simulation started with 576 unique unimals and with the same neural architectures and learning algorithms. When learning, each of them moved either on flat ground or on more difficult terrain with stair steps, smooth hills or blocky ridges. He then entered a tournament against three other animals trained under the same conditions as him.

The winner was chosen to produce a single offspring that underwent a single mutation involving limb or joint changes before facing the same tasks as its parents. All of the unimals (including the winners) competed in multiple tournaments, only aging as new offspring emerged. The researchers stopped the simulation when they obtained 4,000 different morphologies. The surviving animals had then gone through, on average, 10 generations of evolution, and the successful morphologies were surprisingly diverse, including bipeds, tripeds, and quadrupeds with and without arms.

Study results

The researchers selected the 10 best-performing animals from each environment and trained them on eight new tasks: navigating around obstacles, handling a ball, pushing. a box on a slope… They thus observed that the animals which had moved in variable ground obtained better results than those which evolved/moved on flat ground but that they were exceeded by those which had manipulated a box in variable ground. On the other hand, at this stage, they were learning twice as fast as their first ancestor.

These results could be of interest to the robotics industry for the creation of multi-tasking robots.

Sources of the article:

We want to say thanks to the author of this article for this amazing web content

Can Artificial Intelligence evolve like living beings?


Discover our social media profiles as well as other pages related to it.https://www.ai-magazine.com/related-pages/