With the world engulfed by the pandemic, all businesses have been re-shaped to accommodate this new reality. For us at AI Factory the general setup was not radically different as we all worked from home offices anyway, but of course our domestic lives, with home schooling, has been turned upside down.
In terms of this periodical we have, for the time being, switched to an annual publication instead of biannual. Of course we have less "events" to comment on as events like GDC San Francisco has been missing 2 years. There is an on-line version moved to the Summer, but with a 7 hour time difference it just does not work for UK attendees!
However the delayed Develop Brighton has been shifted from its July slot to October and we have elected to book for that. It's a long shot given the current turmoil and maybe with a low chance that it will actually run as planned. If it does happen, we do not want to miss it.
In our confinement and detachment from any direct contact with the industry we have been beavering away on our apps. A significant body of this has been shuffling our ad network usage and at last moving further into header bidding. This has been on the cards for a few years but the take-up has been slow, so switching to this has been late. However this is now taking off and promises a more efficient and lower maintenance environment from where to carry ad traffic. We can gradually dispense with waterfalls and thresholds and instead allow the the system to free run. It has seemed curious that it has taken so long to become well established as essentially the same system has been hooked into Google search for years and proven that it was the way to go.
On the AI side the early part of this year continued with working on Spades. This has been such a long project that the tool environment around the engine has burgeoned into a highly capable system. Much of this is about ease of block testing and exposure of AI thinking. Happily most of this is generic and open to be shared by all our engines.
Our Chess has been getting a substantial makeover as well. We have been putting further enhancements to net play to provide a more personal play experience. Ideally we would support open chats between players, but this is a legal minefield and would need us to police it against abuse. Instead we have worked hard to allow players to communicate with pre-defined chat messages. This may sound very limited but with careful crafting it actually provides a significantly helpful and attractive facility, that at the same time avoids issues of abuse. In purely practical terms it is valuable for on-line play when someone temporarily leaves a game. It is so helpful if the player can ping that that has happened or for the opponent to ping a query why they have not played. Even simple pleasantries are nice as it makes up for otherwise missing engagement. It reveals that there really is a person at the other end, not just an AI bot.
Another significant new branch has been our work on Backgammon. We have had the same unchanged engine for years and we were well aware that its play strength should be better. The app Backgammon Free though has throughout its life been the # 1 ranked backgammon on the Google Play store, so our users deserve a better program. As new AI tech has emerged it is clear that neural net based solutions offer the most likely option to deliver the strongest Backgammon. However that is still new territory for us so we adopted a more traditional approach based on linear evaluation. The main article describes this interesting work in detail.