Chapter 364 The fun has just begun
With this level of artificial intelligence, bugs are not uncommon.
Over the years, AlphaGo only had an obvious bug when playing against chess players from South Korea.
After that, the AlphaGo team optimized the algorithm and the bug was fixed.
"Is it because of the upgrade again?"
“No, we just upgraded the computing power. The core algorithm has not changed.”
"If that's the case, why did this happen?"
"When AlphaGo encounters a move that it rarely sees, it will get stuck in a search dilemma. And if AlphaGo feels that its chances of winning are already very low, the program will easily crash."
After hearing this explanation, the Google leader was also very puzzled.
"Hasn't our AI played countless games? How come it still encountered a rare move? Could it be that they have already caught our weakness and deliberately made a rare wrong move?"
"No, this is not a wrong move, but a very rare good move."
At this time, a staff member said.
This staff member is a software engineer and also an amateur 6th-dan chess player. He plays a very important role in the entire AlphaGo development team.
"This move is rarely seen in games between professional chess players, because normally, humans think it is a move that puts them in a passive position, and the complexity of its changes is simply beyond the processing power of the human brain. However, for an artificial intelligence with extremely strong computing power, this may actually be a brilliant move."
As this staff member said, with the emergence of this bug, AlphaGo has been unable to recover and has made a series of stupid moves.
This is also the difference between artificial intelligence and humans.
When the artificial intelligence determines that its chance of winning is very low, it will start to make random moves, because no matter where it makes the moves, it will have no impact on the final outcome.
Seeing that AlphaGo's moves were becoming increasingly outrageous, the AlphaGo team had no choice but to quickly resign and admit defeat.
If this continues, AlphaGo might fill in all of its eyes, and eventually the entire board will become black's territory.
Seeing such a scene, Nie Jie, the commentator in the CCTV live broadcast room, was undoubtedly the happiest person.
"It looks like AlphaGo has suddenly encountered a serious bug like before. I don't really understand the logic behind these AIs, but I guess a big reason is that our Island AI made a move that AlphaGo didn't expect. This shows that Island AI has much more experience and computing power than AlphaGo. It's just a pity that this game ended too quickly."
Nie Jie couldn't help but feel secretly delighted. This upgraded version of AlphaGo couldn't even last as long as himself and the island's artificial intelligence.
If we round it off, Nie Jie's current strength is already above AlphaGo.
Although he thought so in his heart, Nie Jie also knew that he would definitely not be able to defeat either of these two artificial intelligences.
But no matter what, as long as China's artificial intelligence can defeat the United States' artificial intelligence, it will be a refreshing thing.
"As a professional software engineer, let me explain the possible reasons why AlphaGo has bugs."
At this time, another commentator suddenly said.
The commentator is called Chen Yuchang. He is not only a professional software engineer, but also participates in the research of Penguin Go artificial intelligence.
" Because all current Go AIs are based on deep learning systems to improve their Go skills, rather than using exhaustive methods. This means that every AI may have bugs. To be more precise, in this type of algorithm model, this kind of bug is called overfitting."
"Teacher Chen, can you explain what overfitting means?" Nie Jie asked.
While cooperating with Chen Yuchang, Nie Jie was indeed interested in this issue.
"The so-called overfitting can be understood as decision-making errors caused by unbalanced AI data. I just said that the improvement of AI's chess skills comes from its own deep learning. For example, when AI plays chess with itself, it accumulates 10,000 games. Some chess changes are accumulated more, and some are accumulated less. For the duels with fewer chess changes, the overall winning rate is less useful as a reference because there is too much randomness..."
After listening to Chen Yuchang's explanation, Nie Jie basically understood the reason.
"Can we understand it this way? For example, in a certain variation, the AI only accumulated ten games, won three games and lost seven games, and thought that the winning rate of this variation was only 30%. But in fact, if the accumulated data was enough, its actual winning rate might be much higher than this data?"
"Yes, that's what I mean. Simply put, as long as you keep taking changes that the AI has never seen before, it will easily lead to bugs."
"I see. If I had known about this bug earlier, I would have used more unusual and weird moves when playing chess with AlphaGo. For example, I might have made a Tianyuan move in the first step, and maybe I could have stolen a game and won it." Nie Jie joked.
At this time, in the island quantum laboratory, the atmosphere was also peaceful.
As everyone expected, the Island AI won the first game.
Even though the black pieces were not in an advantageous position, the Haidao artificial intelligence still gained a crushing advantage.
However, Xu You did not expect that the upgraded version of AlphaGo would be unable to withstand so quickly.
"It's difficult to test the true strength of the island's artificial intelligence with a chess game like this." Xu You said helplessly.
As the creator of the island's artificial intelligence, Xu You understood why AlphaGo suddenly collapsed just now.
It must be the island's artificial intelligence that accidentally took a move that AlphaGo rarely makes, directly triggering a hidden bug.
This kind of bug cannot be completely avoided. Only by relying on a large number of duels can the frequency of bug occurrence be reduced.
Compared with supercomputers, quantum computers can calculate much faster, so they can accumulate a huge number of games in a short period of time.
It can be said that the probability of a bug appearing in the island's artificial intelligence is extremely low.
"If it weren't for the fact that this is a match broadcast live on the entire network, I would have wanted to let Haidao AI give up the game directly." Xu Youxin said.
But after only one game of three-game chess was completed, Xu You still couldn't let his guard down.
Xu You decided that if he could easily win the second game, he could give AlphaGo a little more leeway in the last game to test the upper limit of the island's artificial intelligence.
