An AI made humans look foolish in a game of skill – something it did had never been seen before

An AI made humans look foolish in a game of

A robot controlled by artificial intelligence has broken the human record in the maze game.

Artificial intelligence is exploding in every field. A mythical fear is that one day it will surpass or even replace man. This is what we can fear with the latest feat of “Cyberrunner”, a robot which beat the human record in a game of skill. The game in question, the maze, is a marble game combining coordination, strategy and precision. The principle is simple: guide a ball through a maze littered with traps. The board, often made of wood, is filled with obstacles and holes. To reach the finish, you have to use small wheels to tilt the game board.

The latter was motorized to be connected to the robot controlled by artificial intelligence. CyberRunner was thus equipped with two mechanical arms for the cranks, a camera to analyze the route and a computer as the brain for the strategic part. Like a beginner, CyberRunner had the right to six hours of training. Obviously, the first attempts were failures. After this time had elapsed, the robot started a real game. And the result is astonishing: CyberRunner only took 14.48 seconds to take the ball to the finish.

He was faced with three humans who had the same duration of training, but who, once the time had elapsed, still had a lot of difficulty moving their marble forward. CyberRunner even beat the performance record that was set for this game in 2022 by Lars Göran Danielsson. According to Les Numériques, it was 15.41 seconds. The artificial intelligence therefore gave the game champion a 6% lead. The robot even found shortcuts that the researchers had not seen. However, these could sometimes amount to a certain form of cheating: the developers then decided to prohibit this practice from the AI ​​to put it on the same footing as humans.

CyberRunner has a great capacity for learning from experience. A camera analyzes each movement which is then absorbed by the algorithm. According to the researchers whose comments were reported by JVTech, the robot “recognizes which strategies and behaviors are the most promising”. The AI ​​algorithm allows it to quickly learn from its mistakes and thus gradually improve its tests.

If artificial intelligence had already won in strategy games, this is the first time that victory has occurred in a game of skill with a physical part. The images in this part are truly impressive.

lnte6