Our two new releases Tic Tac Toe Universe Free and Gomoku Free offer a shift in presentation style while still keeping the same UI structure. This was a partially outsourced venture with Tony Warriner. It is not until you start outsourcing that you appreciate just how tight the common UI definition actually is. Any deviation from exactly how keys and menus already work is surprisingly disruptive and makes a product uncomfortable for our loyal users to play. The extent that an implied template existed was only exposed when the project was actually tried!
On our own side this offered a chance to use MCTS (Monte-Carlo Tree Search)
for a perfect information game. Gomoku and Tic-tac-Toe are natural minimax
candidates but we wanted to see how adapting our existing ISMCTS, which was
designed for imperfect information games, might fare with perfect information.
We would like to have reported that this was a simple breeze, but life is
often not like that, so new generic hooks were added to make this work well.
These fed back into our generic structure, which each time gets a little stronger
and better to use. We were worried about the strength, but a play test for
both sides against the 5 top most popular Gomoku programs showed we won with
a clean sweep of 10 games to 0.
We attended NEMOG recently, decided right in the last 24 hours, but good to re-connect with the research teams. One presentation included a section based on work with our Spades engine (Dr Sam Devlin in picture below). The work they did has yet to be fully utilised, but the coming work on Spades will use this and the end product should be an AI player that is even closer to human style.
Game AI Pro2, published by CRC Press and edited by Steve Rabin, is now on release. This includes the first ever publishing of our bespoke search method "Interest Search", which we have used for some 18 years and finally felt it should enter the public domain. This method massively increases the speed of minimax, allowing substantial strength improvements through much deeper searches. This method is core to the tournament success of the Japanese Shogi program "Shotest". It is also core to our Chess/Chess Free that uses our Treebeard engine.
Another new track for us has been Amazon Underground, which is a bold new venture by Amazon to draw in users. This provides a selected set of Android paid apps, but with all in-app and retail purchases completely free. Once you have downloaded the app there is nothing to pay, nor any ads. This provides the user with a substantial block of selected paid apps for free. The developers get paid for how often the apps are played. Apart from the payment per unit time, this also provides extra brand exposure as people will discover our apps in this limited subset of the Amazon app store, so that knowledge of the brand will spill over into the Google Play store. On direct income alone there is still potential for significant value as selected app subsets have proven to have immensely high value with the Skype-like app Kakoatalk, which replaces Skype and Facebook for many users. Within this the top apps have even managed to earn millions of dollars per day, simply because they are a member of this limited subset. The idea of pay-for-usage is also used by KDDI, Softbank and DoCoMo in Japan, with great success, so the principle is already proven.
During this year we have had no fewer than 7 updates to the Euchre Free AI. We have connected with helpful beta testers and through this have been able to make some substantial further improvements in AI. Part of this evolution has been the addition of new extra tools to allow the developer to test the app's decisions. In particular to block test the bidding system on individual bids, which was mostly missing before. This intelligently randomises the actual deal using inferred data and then trialling a large number of such deals through to the end of play to measure the value of each possible bid (see below). Note that these playouts include opponent bids so that the value of a bid includes the value of allowing opponents to bid if you pass. This includes the value of not bidding on a strong hand if it causes your opponent to bid with unwarranted confidence. This is a curious "human-like" deviant play that our Euchre is capable of exhibiting. At the time of publication, our existing product Euchre Free is by far the top ranked Euchre app in Google Play and also the top rated.
A disruptive part of our unavoidable agenda this year has also been handling rogue adverts. If everyone played ball and submitted ads that obeyed the rules then there would be no problem, but when rogue ads do get through (which is often) it can be very time-consuming searching for them and getting them blocked. A common example is the ad that tries to tell you that your phone might be infected and offering a download to fix it. These ads are immoral, upsetting for our users and actually financially cost us a great deal to process. Another related annoying track are the plethora of new security apps out there that can easily spuriously report that one of your apps is malicious, even if the app has no permissions other than that used to save games. These flaky apps are annoying and damaging as users reasonably assume the reports are correct so write bad reviews and uninstall the app. We guess it is possible that such apps gain "extra credibility" if they appear to spot more malware than other apps, so making spurious extra detections makes them look more "thorough".
Net though it has (again) been a very good year, with our current total downloads now totalling 109,137,601.
