[ad_1]
In sci-fi movies where an evil robot dominates the world, it is usually always defeated by humans because of the "creativity" we have. But researchers at DeepMind, Alphabet's AI division (owner of Google), proved that creativity was irrelevant when the machine was able to predict and adapt to all its tactics.
Since 2017, when AI AlphaStar became a master of the game. The researchers worked on a much more complex task: to make the program master the game StarCraft II. The game is considered a "big challenge" for machine learning because it requires multiple tasks to be performed at the same time and in real time. Indeed, although the objective of the game is to annihilate the opponent, it is not possible to focus solely on this, and it is worth paying attention to several other minor objectives (card exploitation, improvement of the base and strength of the unit, etc.). that the ultimate goal is reached. In addition, a match may take a few minutes or hours until the end and it is impossible to calculate how long it will last before starting to play it.
But despite the difficulties, the researchers managed to reach their goal. And on Thursday (24), a demonstration left the entire StarCraft – and artificial intelligence – fan community, when AlphaStar confronted two professional players and "swept the ground with their faces". The AI not only defeated both players (TLO and Grzergorz "Mana" Komincz), but won by the wash, winning every 10 games played. and Facebook and you always guarantee the lowest price for your purchases of technological products.
The matches had originally been held in December, but they were only shown on the 24th as part of a show on the AlphaStar AI progress. And even though the AI was already starting the game and could see the whole map, she had a pretty "elongated" ability, her action time being equal to the average of a human player – which meant that she was performing actions as if it were a human.
The demonstration, which lasted about two hours and even had storytellers and commentators, was seen about 34,000 times live on transmissions via YouTube and Twitch. According to DeepMind, the trick used to teach artificial intelligence was to create different versions of Alpha Star, which used different tactics to approach the game and compete with each other. This allowed the AI to quickly learn to adapt and use new tactics other than those programmed by the researchers, then to compete with these "smarter" AIs, making them made it even more adaptable and more easily discoverable. new tactics. According to the researchers, this type of training – which lasted for about three months – has allowed AI to evolve in the equivalent of 200 years of human training.
But that does not mean that the human race is lost: during a friendly match played during the show, after studying for a while the tactics of artificial intelligence, she was eventually defeated by one of the professionals of StarCraft . The full event, including the 10 games in which AI is winning out, can be viewed by YouTube (video above).
Source: 9to5Google
[ad_2]
Source link
