Just past the May Day holiday, for the king’s glory of some players did not leave any good memories. According to the official news of King’s Glory, in the 5.1 to 5.4 period, players into the game can be with the aforept AI against each other, a time of mourning, whether professional players, ordinary players or game anchors, have been AI abused individual skinless.
Renewing The Glory of the King to play with the AI (Photo: King Glory)
the super-strong AI’s epiphany
The name “amazing” means “excellent insight”, developed by Tencent’s AILab Co-King Glory team in December 2017, its 1v1 version first appeared in the 2018 KPL Fall Finals, and in August in Kuala Lumpur, the King’s Glory “World Champions Cup” semi-final series 5V5 against the game, “The Same year” in Shanghai, “The World View” It also opened up the experience test against the amateur 1V1, with a total of 2,100 matches over four days, resulting in a 99.8 per cent win.
Epiphany AI (Photo: AILab)
A paper published jointly by AILab and Tianmei Studios points out that real-time strategic MOBA games such as “Glory of kings” are different from traditional board games and Atari games, which require more difficult and complex operations and are more complex in their playing environment. Take “King’s Glory” as an example, a game may involve 10,600 game possibilities and 10,18000 possible game operations, and this is only the basis, AI also need to launch IN the MOBA game GANK, defense, induction of opponents, tonic and other complex actions, but also may involve complex skills.
AI Team VS Human Team (Photo: King Glory)
Tencent’s researchers systematically encoded image features and game status information to digitally represent different units and enemy targets in the game, and the framework ran on a total of 600,000 processors and 1064 graphics cards, including NVIDIA Tesla P40s and NVIDIAV100s, and processed 16,000 non-hidden unit attributes and game information. Training a hero requires 48 graphics cards and 18,000 processor cores, and the system will continue to train at 80,000 samples per second. The current level of training in the system is equivalent to the 500 years of experience that humans have accumulated.
The training mechanism of The AI (Photo: AILab)
According to the researchers, “disillusionment” has been “out of the division” after 80 hours of systematic training, but after 30 hours of training, “disillusioned” already has the ability to beat the top 1% of players. Full-trained AI will be able to react in 0.1 seconds, which is already the equivalent of a top amateur’s response.
How strong can the game AI be? ==
For many years, game AI has only one purpose in the eyes of players: to teach hands. For game developers, writing programs can achieve the general AI effect, these AI performance in the game is always regular, once familiar with the game, you can find the cracks, so as to beat it, by the way, the modern game AI’s ancestor is “bean eater.”
Bean eaters (Photo Source: Google)
And as the idea of such a game AI first appeared in the eyes of mankind, in fact, from the Alpha Dog beat Li Shishi…
The deep learning AI developed by the Google team was not really the only Alpha Dog at the time, when they also did another AI called “Alpha Star”, and in early 19 years and the world-renowned interstellar player once, the result and Go the same, the human resistance, 1-10 defeat.
AlphaStar (Photo: PCMag)
By the end of the year, Alpha Star’s latest transcript was that it surpassed 99.8 percent of human players and reached the patriarchal level in the Gods, Humans and Bugs. You know, StarCraft as one of the most challenging real-time strategy games, the game not only needs to coordinate short- and long-term goals, but also to deal with unexpected situations, the difficulty is not generally great. And Alpha Star in its APM, the field of vision with human players in line with the case, still achieve the full control of the gods, humans, insects, but also unlocked many maps.
AlphaStar’s performance (Photo: DeepMind)
And Xishanju also joined the development of the game AI, they for their ownPUBG game “Sea of Glory” developed a 3D survival class AI: Orion star alpha; Through intensive learning and training from scratch, you can have complex 3D environment awareness, material search/use, combat, team work and other capabilities, that is, Orion alpha;
Orion Alpha (Photo: Sea of Glory)
Not so long ago, a start-up game studio from Canada opened up a new model: EnemyAI, which allows enemies in the game to automatically track player paths and actively avoid environmental barriers such as corners and rocks.
EnemyAI (Photo: GitHub)
Maybe one day, we really can’t tell if our opponent/teammate in the game is real or AI.