In 2007, Cao Yang asked their research institute three topics.

The first is to make an artificial intelligence Go software, hoping that Liu Guangran and the others can become nine-dan masters in the Go world within 5 years.

The second is to do bridge AI, allowing computers to simulate human strategic behavior. The acceptance standard is that the real person participating in the game cannot tell whether the opponent is a real person or an AI.

The third topic is an extension of these two. To do a chat AI, as long as most humans can’t tell whether it is a real person.

The reason why these three topics is to mention is that Cao Yang knows that although these topics themselves are difficult, they can be done and can be applied in practice.

After all, this is what Google did in his previous life.

On the other hand, after setting goals for Liu Guangran and others and having specific goals, things should be smoother. They will not waste time because the artificial intelligence technology is too broad, or because it is too broad to study everything or aimlessly research.

Also, the main purpose of proposing the project is to have an acceptance standard, otherwise how can Liu Guangran and the others know where to do it? Cao Yang cannot clearly know the current level of artificial intelligence, and they themselves have no foundation.

This is definitely not possible.

So the three topics played a compass for the research, giving them a lamp in the vast sky.

Time flies now, and more than two years have passed in a flash. Half of the time limit Cao Yang gave them at the beginning, so it is normal for Cao Yang to care about the research results.

Among these topics, the bridge problem is the easiest and easiest to produce results.

Then there is Go. Although there are certain differences between Go and Bridge in AI algorithms, the essence is similar, because they are all based on massive databases, and then the process of obtaining the optimal solution through algorithm research and simulation.

Go changes more, so it is more complicated.

AI chat needs to be more complicated and is a comprehensive topic, because you not only need to base your deep learning algorithms based on the above two topics, but also have to conduct in-depth research on Chinese character language and conduct a certain analysis of human thinking logic.

There is also feedback and processing of live chat information.

Otherwise, the answer you give will easily make people realize that this is not a real person.

Computers cannot make human thinking. Language is logical for them, but for humans, language itself contains too much information. In addition to solving problems, it is also the transmission of emotions.

"It's such a boss."

Liu Guangran stroked and searched for relevant documents in his mobile phone.

I had prepared relevant content before the meeting today, so it was more convenient to tell.

"Tell you good news, we have overcome the bridge's topic." Liu Guangran nodded seriously and seriously.

"Wow, is it true? So fast?" Cao Yang was a little overjoyed.

This answer really exceeded his expectations.

Because according to Cao Yang's idea, it would be quite good to be able to produce results within 5 years, and I didn't expect that the bridge problem would be able to produce results in such a short time.

"It's done." Liu Guangran touched his bald head and said embarrassedly, "That is to say, it meets the standards you set in the beginning that real people cannot tell whether the other person is a real person or an AI. We found a total of 1,464 players to test it.

By last month, the recognition rate was almost zero.

Many people think that their opponents are real people online, and there are even many people who directly criticize others because they play cards too badly."

"Uh..." Cao Yang stared at him in confusion, not very understanding what the shyness on Liu Guangran's face was, "So what?"

"But in high-end cards, that is, in terms of optimal solutions, there are still some shortcomings," Liu Guangran smiled awkwardly, "But this is also very consistent with the characteristics of real people. No one will never make mistakes. It is precisely because of some small flaws that our bridge AI seems more like real people."

"But it's just not very much like a bridge expert, right?"

Cao Yang rolled his eyes.

Finally I understand what Liu Guangran means.

The current bridge AI can achieve real-person strategies, but cannot achieve the optimal strategy. In other words, in the eyes of players, this is a real-person player with [limited IQ].

Sometimes I make mistakes, and the card counting is not always as accurate as ever, and there are certain shortcomings in cooperation, but I am always a good real person.

There are times when you play well, and there are times when you play badly.

This is probably true human nature...

Cao Yang felt a little uncomfortable.

If bridges are made into this ghost, then Go is just...?

"The progress in Go is quite smooth," Liu Guangran said with a smile, "The Bomi Go Master version 4.18 that we are developing now can already win professional five-stage chess player."

"Hey?! Really fake?" Cao Yang was a little embarrassed to believe what Liu Guangran said.

"But some strategies may need to be modified." Liu Guangran said, "The algorithm we used to learn and play chess by AI by itself will have some minor flaws. After several improvements, it seems to be a good result. I think that if we change it a few more times, we should be able to play a professional 9th ​​chess player."

Although there is a certain gap between the fifth and ninth stages of career, the gap is no longer obvious.

Because in the Go world, it is not impossible to win the Ninth dan at the level of the fifth dan.

It’s just that some Jiudan, who have been famous for a long time, are more experienced in experience and handling, and some talented players can maintain a greater chance of winning.

So if the current [Bomi Go Master] can play five stages, it is indeed not far from winning nine stages.

"What's the reason?" Cao Yang suddenly felt a little funny. "The bridge should have fewer changes than Go. There are only so many cards and combinations. Why is Go progressing more smoothly?"

Liu Guangran felt a little ashamed.

I originally thought that the boss raised these topics, so I knew better, but I didn’t expect that he would explain the reason in the end.

"The number of bridges has actually not changed a lot. Players can choose 8 decks, 12 decks, or 16 decks to play, and then there are much more people playing bridge than Go. It is a 4-player strategy game.

This complexity is higher than that of Go 1v1.

If we do Go, we just need to consider the front and back. Bridge needs to consider the relationship between opponents and partners, which are all different.”

Oh, Cao Yang finally understood after hearing this.

Therefore, it is the easiest for AI to do 1v1 strategy. The complexity of chess is far lower than that of Go, so it was quickly broken through. The reason why Google chose to do Go is mainly because it is relatively easier, more ornamental and effective.

It can maximize the topic and increase the company's reputation.

"No wonder……"

Cao Yang touched his chin and seemed to be too young and had limited understanding of AI.

In fact, at this level, although the complexity of Go has increased, it is mainly reflected in the strategy after each move. The later the space for change is, the easier it is to calculate.

And they are all based on simulating massive data and deep learning to find the optimal solution. Therefore, after the Go research, it will be more effective for the company to conduct accurate analysis and delivery of big data.

"This is a good thing," Cao Yang couldn't help but smile knowingly, "Then when will we win the Nine-Dan player?"

"This depends on the situation." Liu Guangran said, "After all, a game of Go takes a long time, and we have to make an appointment with the players in advance before we can do the test. The test cycle is once a month.

According to my personal judgment, there should be no problem at this time next year."

Wow.

When Cao Yang heard this, he felt very excited.

If Liu Guangran's judgment is not wrong, then when Bomi's Go master comes out next year, it will definitely cause a sensation in the world.

I feel extremely excited when I think about it. In this world, the first company that can defeat Lee Sedol is actually Bomi.

And it will definitely be Bomi.

Cao Yang is increasingly looking forward to the arrival of early that day.

...

"The remaining topic is progressing relatively slowly now." Liu Guangran said embarrassedly, "AI's understanding of language, even if it is given to it in very standard sentences that conform to Chinese grammar, will still cause errors when it analyzes it.

It is possible that if you say some more complicated sentences to her, the AI's response will be wrong.

This topic is really a bit difficult for us at present, and it probably takes a lot of time." Liu Guangran said.

"Yeah, it's okay, keep doing it."

This was also as Cao Yang expected. If it was really easy to simulate a real person chat, he would have come out in his previous life long ago.

However, according to the news, it seems that in 2021, a foreign professor can already let AI simulate her dialogue with students and assign homework.

It’s not impossible to do it, but this development does require a process.

Cao Yang is still optimistic about his team.

Bomi Company is not short of money, just invest slowly.
Chapter completed!