Artificial intelligence must not be able to create everything from "zero". Artificial intelligence's self-learning ability is also based on foundation and cannot appear out of thin air. Yin Ye also believes that mutual fighting algorithms that do not rely on databases cannot cope with calculations with unclear rules. “For example, in the medical and health industry, data is still king, and the algorithm will be continuously revised based on the accumulation of data, moving from artificial intelligence (AI) to real Intelligence (RI).”
On October 19, the British magazine Nature published a paper reporting that a new version of the “AlphaGo” computer program can start from a blank state without any Under the conditions of human input, one can teach oneself without a teacher and become self-taught.
The creator named it "Zero". Taoism says that Tao gives birth to one, which gives birth to two, two gives birth to three, and three gives birth to all things. This is a Go skill that can surpass the top human players from scratch. It sounds subversive and makes many people afraid.
In addition, the world-renowned Go player Ke Jie, who was defeated by AlphaGo, said on social platforms shortly after: "A pure and purely self-learning AlphaGo is the strongest... In terms of AlphaGo's self-improvement... …Human beings are too redundant (please contact the context and pay attention to the context).”
The concept of “human beings are redundant” is taken out of context. An algorithm can learn superb chess skills by itself, and artificial intelligence can “zero.” "The creation of all things is interpreted in this way, but is this the reality? A reporter from Science and Technology Daily conducted an exclusive interview with industry scholars to hear what they had to say.
Just listening: Zero also needs a database
“Zero can generate its own data. The data of the next second will compete with the data of the previous second. If it wins, it will be generated again. In this cycle, winning strategies are gradually 'evolved'," said Gao Hanzhao, a member of the expert group of the China Chief Data Officer Alliance and founder of Hansi Technology.
So, without the support of a database, how does Zero generate data and how does it know what kind of data is generated? Is there really "intelligence"?
Yin Ye, CEO of BGI, does not agree with the concept of "self-taught". He compared AlphaGo Zero with AlphaGo in terms of learning sources. "The latter was input into 30 million chess games in human history and learned the algorithm corresponding to the rules of Go step by step. Only by standing on the 'shoulders' of AlphaGo and inheriting the latter's algorithm with clear rules can Zero understand Go. Follow these rules and fight each other."
In other words, Zero learns the rules from its predecessors. "It only does not need the previous human game data, which does not mean that it does not need a database. " Gao Hanzhao said.
Xun Endong, a professor at Beijing Language and Culture University, interpreted Zero’s ability to make “higher moves” as two reasons. One is that “under the same rules, the placement strategy captured by the machine is different from that captured by humans. "Different"; the second is "Its calculation speed is faster than humans, so it can perform more in-depth and even exhaustive calculations, and fully knows the next steps."
In other words, after finding the right path, Still running fast. However, there are prerequisites for such a situation - "The rules are simple! The outcome is clear! The strategies can be exhausted!" Xun Endong said.
By analyzing Zero's learning process, you can understand why these types of problems are necessary for them to win.
According to the rules, it continuously generates new data, and then the old and new data compete against each other, ultimately producing a victory or defeat. In other words, under fixed rules and constant verification of "win or lose", Zero has improved its chess skills, and when all these strategies are verified, it becomes invincible.