2017 AI in the game to win the human race, 2060 all the human task _ai

Click to have a surprise

In the 2017, Ai swept the game. Over the past 12 months, from old Weiqi to Texas poker, Ai has crossed a series of new thresholds, eventually defeating human players in a variety of different games.

Go

In the late the 1990s, a machine finally defeated the chess Grandmaster. In the face of the challenge, computer scientists have shifted their attention to the ancient Chinese game of Weiqi. Go looks simple and easy to play, but it's very complicated.

In October 2015, Alphago defeated 樊麾 and became the first computer go program to beat chess players on the 19-way chess board, writing a history, and related results were published in Nature in January 2016.

2016 countdown to the third day, a user named "Master" of the Go users, a continuous in the go platform for slaughter, "kill" including Coger, Park Tinghuan, Yuki Kana, Chen Yaoye, Shin, Changhao, Coulee and other road to master, and finally ended up with 60 winning streak. Coger even lost three, during the hospital due to acute gastroenteritis ... By January 4, the official confirmation that "Master" was actually the latest version of DeepMind AI Alphago.

May 27, 2017, Wu Town go Summit, Coger total score of 0:3 defeated by Alphago. In the competition, the Coger in the situation unfavorable long time leaves, comes back after tears sprinkle the scene. After the game Coger once choked to say: it is too perfect I am painful, see no hope of victory.

In October 2017, DeepMind's latest edition of Alphago paper introduced the most recent version Alphago Zero: Using pure reinforcement learning, integrating value networks and strategy networks into one architecture, defeating the previous version of Alphago after 3 days of training.

December 2017, DeepMind launched the Alphazero. Can master a variety of games within a few hours. After eight hours of self training, the system not only defeated the previous version of Alphago Zero, but also defeated the most powerful AI of chess and chess.

December 11, 2017, Alphago's first teacher, European go champion 樊麾 on Weibo revealed: "Alphago Teaching Tools a total of more than 20,000 changes, more than 370,000 steps of chess composition, through the Alphago perspective, analysis and suggest go to start a lot of dismount." Then, the first person in the world of Weiqi Coger later forwarded this tweet and commented: "Learn Weiqi again." ”

Alphago has retired, but technology endures.

Texas Poker

In addition to go, in unrestricted Texas poker, human top professional players are also defeated by artificial intelligence.

The success was first deepstack by computer scientists at the University of Alberta, with help from the Czech University of Charles and the Czech Polytechnic University in Prague. In the four weeks beginning of November last year, Deepstack defeated 10 of the 11 professional players, statistically, with a great advantage, playing 3000 hands with each opponent.

One months later, the Libratus developed by Carnegie Mellon University has again won the race against humanity. Libratus defeated four more outstanding professional players (Expert poker players), the overall settlement of more than 120,000 hands. The computer finally won about 1.8 million dollars in chips.

Electric competition Game

On August 12 this year, Ellon Masc's OpenAI, a company designed to study general artificial intelligence solutions, was trained in an AI algorithm that took part in the 1V1 competition in the international of the famous Electronic games DOTA2 International tournament. And overwhelmingly defeated Dendi, the top electronic athlete.

Compared to the complete information static game game (that is, both sides can get all the information of the situation, the chess is asynchronous), the electronic athletics game has the incomplete information attribute (both sides only grasps the partial information in the view), moreover the athletics process is the real-time dynamic. Its rules and characteristics are far more complex than Weiqi, so the evaluation and decision-making ability of AI is significantly higher. In addition, compared to the highly abstract game of Weiqi (only drop one action), the electronic athletics game is more similar to the human real world confrontation/war simplified simulation, involving a large number of resource deployment, troop composition selection, expansion strategy, attack and defense confrontation complex but realistic game, If the electronic athletics confrontation to achieve landmark milestones, it represents the artificial intelligence to the general and a big step forward.

Ai conquers humans in almost every game and what the next step is.

If in almost every game, AI can beat us. So what's next.

A researcher at the University of Falmouth recently revealed a machine learning algorithm that says he can play the game from the beginning. This artificial intelligence system, known as Angelina, is improving every day and can now use datasets to make games, which can be obtained from Wikimedia Commons (Wikimedia Commons).

What this means.

Perhaps the most important and frightening development of the 2017 years is the great progress in strengthening the learning system. These programs can effectively teach yourself how to master new skills. Alphazero is an excellent example.

A large survey of more than 350 AI researchers has shown that it will not be long before the AI finishes abusing humans. The survey predicts that artificial intelligence will perform better in 10 than humans. By 2049, it will be able to write a best-selling novel, and by 2053 it will perform better in surgery than humans. In fact, the survey concludes that by 2060, there is a 50% chance that AI will be able to do what humans can and do better.

2017 is undoubtedly an important milestone for artificial intelligence to defeat humanity in an increasingly complex game, although it may seem like a trivial achievement, but its impact is enormous. Many of these AI development companies are rapidly turning their eyes to the real world.

Demis Hassabis, co-founder and CEO of DeepMind, said: "In the end, we want to use algorithmic breakthroughs to help solve various pressing real-world problems." If similar technologies can be applied to other structural problems, such as protein folding, reducing energy consumption or finding revolutionary new materials, these breakthroughs may have a positive impact on our lives. ”

Click to have a surprise