15:11 GMT13 July 2020
Listen Live
    Viral
    Get short URL
    113
    Subscribe

    The OpenAI research company has presented an extravagant new study of machine learning based on sumo wrestling.

    RoboSumo, one of the latest Open AI experiments in machine learning, involves a pair of ‘robots’ dropped into a virtual arena without even the knowledge necessary to walk, and forced to learn the tricks of sumo wrestling purely by trial and error.

    The video posted on YouTube shows how the bots initially clash without employing any tactics or strategy, but after a number of bouts their movements start to resemble those of human wrestlers, as they learn to dodge and attack.

    According to the Wired, OpenAI researchers created RoboSumo because the competition apparently generated extra complexity which "could allow faster progress than just giving reinforcement learning software more complex problems to solve alone."

    "When you interact with other agents you have to adapt; if you don’t you’ll lose," Maruan Al-Shedivat, one of the RoboSumo creators, said.

    Related:

    Know Your Heroes: PornHub's New AI to Watch Videos, Guess Actors and Genres
    'Summoning the Demon,' 'Tower of Babel': Debate Over AI God Heating Up
    'Computers Have No Feelings': Can AI Replace Humans?
    Tags:
    machine learning, sumo, Artificial Intelligence (AI), experiment, demonstration
    Community standardsDiscussion