A scientist from Peking University recently published a preprint of a scientific article detailing
What is the essence
Most studies in the genre“pursuit-avoidance” in AI and game theory are concerned with teaching machines to explore space. Since most AI training involves a system that rewards the machine for achieving a goal, developers often use gamification as an incentive for learning.
In other words, you can't just stick a robot in a room and say “do this”. You must give him goals and a reason to achieve them. That's whyresearchers are developing AI that by its nature seeks to reward.
The traditional intelligence training environment challengesAn AI agent is tasked with manipulating digital models to explore space until it completes its goals or finds a reward. It's reminiscent of Pac Man: the AI must move around the environment until it eats all the reward pellets.
History of the issue
Ever since DeepMind's AI systemsmastered chess and go, SCII became the primary training environment for competitive AI. It is a game in which players, AI, or combinations of players and AI naturally oppose each other.
But more importantly, DeepMind and othersresearch organizations have already done the hard work of turning the game's source code into an AI playground with several minigames that allow developers to focus on their work.
Researcher Xun Huang, the aforementioned scientistfrom Peking University, set out to study the “pursuit-evasion paradigm” for training AI models. But I discovered that the SCII model has some limiting limitations: in the built-in version of the game “pursuit-evasion” Control of pursuers can only be entrusted to AI.
The basic scheme includes three pursuingcharacter (represented by the soldiers from the game) and 25 evader characters (represented by the aliens from the game). There is also a mode that uses “fog of war” to darken the map, making it difficult for the pursuer to detect and destroy the evader, but according to research, this is a 1V1 mode.
Funny but basic behavior 25Dodgers' strategy is to remain stationary wherever they appear and then attack their pursuers on the spot. Since the pursuers are much stronger than the evaders, this results in the expected destruction of each evader immediately upon detection.
Prospects
Huang's article describes the paradigm in detailAI training in the SCII environment, which focuses on teaching AI to evade pursuers. In their version, the AI tries to hide in the “fog of war” to avoid capture and death.
This is a fascinating study usingvideo games that could have huge implications for the real world. The world's most advanced military organizations use video games to train people. And AI developers use these learning environments to prepare AI brains for life inside a real robot.
Purely theoretically, Huang's work seemsexciting. But just imagine a Boston Dynamics robot, endowed with the ability not just to run and jump around the site, but to purposefully evade pursuit by a special forces squad.
Source: arxiv, deepmind, thenextweb
Illustrations: goodfon
</ p>