The latter part of 2017 saw yet more work on Spades, which seems to offer endless opportunities for making improvements. An issue we have with Spades, not shared by games such as Chess, Reversi and Go, is the very large range of rule options. In the case of our Spades there are over 2000 combinations of possible rule sets. Reassuringly MCTS (Monte Carlo Tree Search) has that attractive attribute that it naturally adapts, so that an added rule will automatically cause the search to find different moves that accommodate the changed limitation. This is all fine but is no guarantee that it will play well with all rule combinations and, given how much performance testing needs to be done, the AI may indeed "cope" with some rule combination, but still not play well. In consequence our bulk testing has been based on the default rules and we have neglected block testing of other rule combinations. In doing so we have found that the Nil passing rule is not well handled, so we are back to working on this.
In among all the work, we again had our annual pre-Xmas AI Factory party with our usual Android cake, celebrating our 142 million downloads. It is a good chance to re-meet up with colleagues in the industry and all our friends. It is also a good time to take stock, prized away from our desks and with a glass of wine thrust into our hands. AI Factory still looks pretty healthy.
Our Chess Free still seems to hold ascendancy among the Android Chess apps, with over 70 million downloads and a fairly firm first place in the rankings. This owes much to being the first polished Chess app on Android, but also owes to the fact that it is a well-crafted app well suited for teaching and coaching people to play the game. It is aesthetically pleasing and pleasant to use. The game engine is also very unlike regular chess apps in that it uses our bespoke "Interest Search", which was core to the success of our Shogi program. This delivers human-like play and human-like mistakes. However one is never safe and a limitation in our app is its lack of support for on-line play, and yet up and coming rivals have this. We have tested and released this for our Checkers, so we can discover issues, and are now beta testing on-line play for Chess as well.
Part of the regular distraction is to pass on our experience to up-and-coming students, so we contributed a talk to the IGGI students at Queen Mary, London University, followed by a panel discussion. In preparing this it was a revelation just how much we had done as a company in our 15 years in the market. We tend to work looking forwards and past work fades in the memory, but we have had so many projects on so many platforms, and yet there are still only 3 of us. Preparing this is however also helpful in focussing your mind on what you are doing by forcing you to explain it to someone else. The IGGI group are full of very keen and sharp minds, so this is a great opportunity to bounce ideas with them.
In 2015 we published a chapter: Interest Search: A Faster Minimax in Game AI Pro2, Edited by Steve Rabin CRC Press: ISBN: 978-1-4822=5479-2. This describes in detail our bespoke search method "Interest Search" which proved highly successful on our Shogi app and is core to our Chess program. You can now download this excellent book for free using the link above.
Another side project has been to create a new Backgammon program based on MCTS. This has been stop-start as other priorities emerge, but already the program is stronger than the one currently in release, even though this as yet has no search added.
In conclusion another good year. The political mayhem of Brexit has one silver lining for us, in that the value of the pound is so low that our dollar income is inflated. It is compensation and does make us stronger.
