The hottest in-depth analysis of interstellar 2 it

In depth analysis: Interplanetary 2 AI conquers humans? It will take time

at 2 a.m. Beijing time on January 25, the AI alpha star jointly researched by Blizzard and Google deepmind team was officially unveiled live. According to the live broadcast arrangement, alpha star and two "human" professional players conducted a five match demonstration. In addition to the duel that was not demonstrated in the live broadcast, human beings won only one of the 11 games of human vs Alpha artificial intelligence

the war ended with a near complete victory of artificial intelligence. As a result, it seems that humans will soon be unable to compete with AI in StarCraft II, a very difficult game

from the perspective of professional interstellar technology and tactics, this article will help you understand the significance of this victory through a more popular explanation. In order to use the experimental machine to measure the deformation measured by the extensometer and convert it into a large voltage installation

(special thanks to kevinliu for his guidance of AI professional content in this article)

1 Does this victory really mean that AI in StarCraft II can defeat top humans

although mankind did not send the professional player with the strongest strength and the best condition to fight this competition, it is obvious that many friends hope to refer to whether artificial intelligence has the strength to defeat human professional players by referring to whether artificial intelligence has the opportunity to defeat all mankind in the near future

so answering this question actually requires a necessary condition. That is to see how to treat the artificial intelligence that challenges human beings. In the view of most professionals, if AI defeats human beings, it should be through the same operation as human beings or more advanced tactical understanding of games than human beings, which is enough to convince people

can the two players who lost to AI in this competition be convincing

I'm afraid it's difficult

TLO, the first to fight, used the race he was not good at in the game, and his technical and tactical strength can only reach the level of ordinary amateur players, even amateur masters are slightly reluctant. Then naturally, there is no reference to the significance of defeating professional players. It can only be said that this alpha star, which competes with TLO, has a slightly better technical and tactical level than some amateur players in PVP (Protoss vs Protoss) confrontation

mana, the second to fight, is a serious professional player who is best at race. He also won the runner up of Blizzard World Championships at his peak. But mana still lost to AI

is this victory of referential significance

objectively speaking, there are still objections

the reason why most people disagree is very obvious. This alpha star, which defeated mana, has an operating ability beyond human physiological limits

after all, human beings hope to compete with AI under a relatively equal condition like go, so as to let human beings know whether there are still defects in the accumulation of cognition of this game playing method

therefore, in the debate on how to ensure equal conditions, whether it is necessary to simulate forced errors like human beings and whether it is necessary to have a response threshold like human beings have always been incentive disputes. However, there is little debate about whether people have the ability to exercise beyond the limit of human victory. After all, in the mainstream cognition of AI against human beings, people hope that AI opponents are also virtual people

looking back on the five wars between mana and alpha star, taking a game broadcast live as an example, the three-line hunting and containment operation shown by alpha star is almost impossible for humans to complete. To complete this operation, we must maintain a very fast hand speed, accurate click, accurate screen movement and distance judgment within that time of more than 1 minute. More importantly, we must put an end to forced errors that human beings can't overcome

generally speaking, if you are also leading the troops to fight, my troop formation and war strategy are similar to yours, but your troops are tireless. Our army is tired and needs to rest, and the enemy can still carry a knife to fight forward, which is a bit shameless

in addition, the additives added in the current alpha can be reprocessed repeatedly without seriously damaging its original function. R there are still great limitations in technology. Judging from the competition on that day, what deepmind can show you at present can only be limited to the only map, the only racial confrontation and the only balanced environment. Compared with the human professional players who need to deal with at least a different map every season, three different racial confrontations, and often modified balanced versions, just like playing football, alpha star only shoots penalties compared with you, how can it be regarded as defeating mankind

but from a purely technical and tactical point of view, this AI war has still given mankind some inspiration. The farmers' distribution of the main mine and some ideas of operating to supplement farmers have indeed given some core star players new ideas

although putting aside the simple score results can not convince us that AI can challenge or even defeat professional players, it is still worth looking forward to from the perspective of its potential and learning level

2. What are the praises of artificial intelligence for interstellar professionals

in the first question, it seems that we have been singing down the alpha star developed by deepmind team. But strictly speaking, deepmind's first public announcement of alpha star is still beyond expectations for interstellar professionals, which is very commendable

compared with some interstellar scripting programs named AI, alpha star is more like AI we imagine. It will flexibly use tactics and make game instructions that are more like thinking

alpha star has been very strong. It has even learned to judge which target to attack first, how to use the opponent's tactics, and the basic ability to judge the opponent's play and use the counterattack. To complete these more abstract logical abilities in our view, we need more powerful algorithm capabilities

more complex algorithms bring more humanized advanced logical reasoning ability, which is also one of the reasons why deepmind chose to use StarCraft II to overcome the bottleneck of artificial intelligence technology. For interstellar players, more powerful logic capabilities may naturally produce more new tactics, new operations and new play systems that human beings have not thought of, which is very worthy of expectation

3. What will the future "StarCraft II" AI vs human war be like

you should know that alpha star, who participated in this competition, has only carried out self-training and learning for a week, but the amount of learning has been equivalent to 200 years of normal people. Like alpha go, AI has a very terrible computing power. But after all, AI at the current level of technology can't fully grasp and comprehend the game "StarCraft 2" like go with fewer variables (this is described in detail in the shorthand of StarCraft AI alpha star competition written by the author before). Therefore, even with 200 years of learning, Today's alpha star's overall understanding of the game is still not as strong as human beings who have only a few years of learning

judging from the current technical level, there may be a big bottleneck in the technical improvement of alpha star. If there is a great technological progress in the algorithm in the future, alpha star may have a great learning progress, so it is really possible to defeat the strongest professional player in mankind or inspire a new "StarCraft 2" playing method for mankind

at that time, alpha star will naturally not have to limit the dead map and the types of confrontation, nor rely on the operation ability beyond the limits of human beings. It will really tell you how to play StarCraft 2, and will tell human beings like alpha go: your landmarks are the strongest, in our view, they are just simple computers; The best way you have recognized is to make a mess

4. You will definitely confuse them: artificial intelligence and scripts in StarCraft 2

in fact, as early as before alpha star publicly played StarCraft 2 today, many star players will mistakenly think that alpha star is similar to Hummer 2000

Hummer 2000 is a program script of StarCraft II, which can show mankind the most perfect game operation in human ideal. For example, the most well-known riflemen killed light and poison without damage, and 2000 dogs charged the tank array with the lowest loss

this is similar to the operation beyond human physiological limits we mentioned earlier. This kind of limit operation is realized by writing the conditional action instructions in the program in advance, such as triggering the gunman to move in the opposite direction of the poison explosion movement when the distance between the poison explosion and the gunman is less than a certain threshold, and so on

however, if you use a script such as Hummer 2000 to compete with humans, there is no practical significance, because human knowledge and understanding are always limited. With limited knowledge, write a large number of conditional action instructions, and finally there will always be unpredictable situations for humans, and the script bot will be in trouble, so it will be easily defeated by humans. Alpha star is a learning type. Through a lot of game learning, it optimizes its decisions every time, so as to improve its game level. Such AI is capable of making logical derivation, situation analysis and even psychological and tactical interaction with opponents. After all, StarCraft II is a real-time strategy game

as for the difference between script and artificial intelligence, I believe many friends can understand from the above. Script is just a dead program, and it can't compete with your CFRP market in 2014, which mainly focuses on the utilization of aerospace and wind power generation. In a real sense, intelligent real-time strategy games; Artificial intelligence will judge your playing style and output the most valuable strategy through the judgment of the situation. In the eyes of mankind, it uses changeable playing styles to challenge you. In fact, strictly speaking, AI, the top technology represented by alpha star at present, is also a more advanced and complex script. All its choices that have become the first choice for lightweight design of many structural components are based on more advanced and complex derivation, learning and fitting than scripts written by humans. It's just the final result. From the perspective of human perception, AI has the ability to think independently

5. What is the significance of deepmind's research on alpha star

at the end of the live broadcast that day, deepmind mentioned that alphago's victory over humans through go was a milestone in AI. But the reason why human beings are human is that they can learn from different results. Deepmind hopes to find a self-learning algorithm that can be the same as human beings. This long-term plan is of great significance to mankind in the future

by challenging this game with changes, information inequality and other complex conditions in every frame, let AI achieve new technological breakthroughs, and finally hope that it can be used in human society and benefit human life