Here is an interview with Alcides Schulz where he discusses the development of the engine, the computer chess sector, machine learning, NN engines, and TCEC in general
We want to welcome the Brazilian chess engine to the Top Chess Engine Championship! Can you introduce us to Tucano and yourself?
I live in New Jersey with my wife, my son and the dog. I moved from Brazil to United States in 2001. Currently I work as software developer in a private equity investments firm. Besides game programming, I spend my free time mostly with my family, I like watching sports, mostly soccer, and reading about information technology and game programming.
I work as software developer and have programming as a hobby, my interest is mainly in game programming. I wrote some games before, like dominoes and cards. Then started researching about chess programming, that led to the development of Tucano.
I started working on chess programming around 2009 and released the first Tucano version in 2012. Whenever I can I do small improvements here and there and can have at least one release every year. Current version is 7.00 and was released in the end of 2017. In this version I added support to multi-thread search (SMP). The main feature I am planning for next release is the end game tablebases.
Tucano is the first engine that was confirmed for S12 div 4, yet it is a newcomer. What are your expectations?
Chess programming is a very competitive field. Even if it is division 4, the engines from last season were very strong. I imagine that will be the same in this season, so I hope Tucano can have a good participation and maybe inspire other Brazilian developers to get into chess programming.
4. How ambitious are you with Tucano, to what level do you want to bring it in the near future?
I have a couple of defined goals regarding Tucano development. In terms of ELO, hope to break the 3000 line someday. I want to improve the code readability and structure.
In terms of research I would like to test some ideas like machine learning, a chess engine can generate a lot of data (millions of positions), so I wonder if somehow it can benefit from machine learning techniques. In general, I want to continue improving Tucano, progress would be slow, but steady. Chess programming is very dynamic and there’s always something to learn.
You mentioned machine learning. As NN engines come into the picture, can you share your opinion of the match Alpha Zero – Stockfish?
Alpha Zero caught everybody by surprise, but I think it left with the sensation that we could see more. I think the chess community would benefit if we can see more games from Alpha Zero, and also learn from their approach to chess programming. Yes, I am not sure this can happen, but I would like to see a match between latest Stockfish and Alpha Zero in a tournament type time control condition, such as Top Chess Engine Championship. In my opinion, the time per move control, as they played, may have limited Stockfish a little bit. Engines benefit from the time management, where they can spend more or less time depending on the move. Of course, the match itself could be very interesting for chess players that will see good games and programmers that can be inspired by new ideas.
The last Superfinal of TCEC was for Stockfish heading to a definitive victory. Did you follow it?
Yes, I’ve been a TCEC follower since season 1, and most of the time I have the TCEC page open in my computer so I can follow the games, and read the comments. There is always a good discussion going on the chat. It is a little surprising the margin with which Stockfish won, especially the difference with the other top engines, but I guess this was expected to happen sometime. Stockfish has a great development framework, the Fishtest, with lot of resources, that provides continuous improvement. But as I said this is a very competitive field, with a lot of talented people. Eventually someone will come up with some breakthrough and things will change again.