Jinxing’s Interview
https://medium.com/@ameyapimpley98/jinx ... d12f67277a

"This is an interview conducted by me with Jinxing, who is the author of the Go program Golaxy. Golaxy is considered to be one of the strongest available Go engines in the world and has won many championship titles such as UEC and Fujian to name a few. Below are some of his thoughts regarding the current state of computer Go, his engine Golaxy and a bit about himself.

1. Tell me something about yourself, and highlight your strengths and interests.

I am a computer engineer. After graduation, I have been engaged in artificial intelligence research and development for many years, as well as system development and cloud computing.

2. At what age did you start playing Go?

I began learning the game Go at about six or seven years old. But until I started developing Golaxy, personally, I am not a strong Go player.

3. What was your biggest motivation for creating a strong program like Golaxy?

I choose to participate in Go because it is a significant project. It is the intersection of culture, competition, entertainment, and technology. Go has been passed down for 4000 years and is an excellent representative of the traditional culture of the Chinese nation. It is also the best way of entertainment for self-cultivation. Nowadays, Go is also the best experimental field of artificial intelligence. AI teams worldwide use Go to improve artificial intelligence technology.

4. What are your views on Go tournaments having unfair hardware?

It can never be fair. First of all, we need to know what the competition is? The most robust Go performance or the best algorithm? The current tournaments do not limit the hardware, so it is the most vigorous Go-playing strength competition. Someone may argue that comparing the best algorithm is more meaningful. However, even if the hardware for the competition is unified, the hardware for training cannot be unified. Whether the hardware is limited or not, fairness cannot be achieved. The so-called fairness has no practical significance.

From another perspective, combining much hardware with playing a game of Go is a great challenge. From years of experience in system development, it is pretty tricky. If you want to win the game, it does not depend on how much hardware you have but on how much hardware you can make it work together.

5. Speak something about the other Golaxy developers. How did you guys come together?

Many colleagues in our team, engineers, and Go enthusiasts have Go feelings. We come together for the common goal, hoping to promote the development of Go through artificial intelligence so that more people can use the excellent Go artificial intelligence software like Golaxy.

6. When you first created Golaxy, what did you expect from it, were you sure it would be the no. 1 engine in the world?

When I decided to develop the Golaxy engine, my goal was to make it the best engine. If the goal is not great, it is not worth achieving.

7. Do you support KataGo clones? Also, do you feel they should be banned from Go tournaments?

Frankly speaking, KataGo clones makes Go AI tournaments dull. Anyone can download Katago and choose another name to participate in the tournament. If you change only one line of code or none, you can say that you have made the so-called optimization based on open-source software. The Go AI tournaments in recent years are no longer grand gatherings for researchers and developers. Such clone programs should not be allowed to participate in AI tournaments. However, the organizers are also unable to judge the innovation and originality of AI programs, which indeed caused such an embarrassing situation at present.

8. How has AlphaGo paper from Deepmind affected computer Go in general and Golaxy in particular?

AlphaGo is a milestone in the development of computer Go. All Go programs that surpass the top level of humankind today refer to AlphaGo’s papers. I have read these two papers repeatedly and learned a lot from them. The papers also inspired me with new ideas, many of which have been reflected in Golaxy. For example, the AlphaGo model has two outputs, policy and value. I believe that since 2 outputs are good, why not 4. So the model of Golaxy has four outputs: policy, value, score, and area. As a result, Golaxy can judge not only the winning rate but also the winning score.

9. In Chess and Shogi engines use NNUE to evaluate positions and that has greatly improved their playing strength compared to classical evaluation methods. Do you think NNUE can work with Go engines or is the current approach the best?

In recent years, many artificial intelligence innovations have been driven by deep learning. Deep learning has a fantastic ability to model complex problems, which is especially suitable for chess and Go artificial intelligence. Nevertheless, this method matured nearly 20 years later than the alpha-beta pruning algorithm used by DeepBlue. Since the advent of AlphaGo, chess AI researchers have also tried to use deep learning methods but once encountered some difficulties. NNUE is a successful attempt to make better use of deep learning technology in chess AI. However, for Go, MCTS is still the most suitable method.

10. Do you think open source Go can beat closed source in the future? Also what is lacking in open source which closed source has?

The choice of open source and closed source is related to the type of project. For projects like Go AI, success mainly depends on a few core contributors rather than the joint efforts of many people. Therefore, we cannot improve the AI level by open-source it. Open source is an excellent idea to provide a convenient way for the public to download and study for free. Let us wait and see who will be more robust in the future, open-source or closed-source.

11. How many more years do you plan to develop Golaxy?

The goal of Golaxy project is not only to develop the AI with the most robust Go performance but also to develop the best AI program that Go fans in everyday life can use. It provides rich functions such as training opponents of all levels, game review, game analysis, live data broadcast of the competition, puzzle-solving, etc. Golaxy is already on this road. However, it may take years to achieve our ideals.

12. After you are done with Go do you plan to move to other board games such as chess?

There is no plan to develop chess AI for the time being. Our team will still focus on Go AI and products for a long time. However, the possibility of developing chess or other games exists in the future.

13. What advice would you like to give to new developers who want to develop a Go engine and are just planning to start?

Go AI engine, which combines algorithm and system, science and engineering, is a perfect project for computer science students to practice and improve professional skills. I hope you can think more, innovate boldly, and not stick to the previous methods. Here, I wish you breakthroughs and success."