In the previous newsletter we reported a rise of downloads from 6 to 40 million in 12 months. This time we have shifted from 40 to 60 million in 6 months, so we are clearly making good progress. Of course the depth of competition is getting tougher so we need to keep our products competitive and relevant. To address that we have spent much time further re-optimising our apps for tablet use and employing our own in-house control system for pop-ups and for on-line control for which ad mediator we were using. This threw up interesting problems to solve and borrowed ideas used in controlling Ethernet contention. The issue is that on-line control offers the possibility of network saturation so we needed our apps to anticipate this and have strategies to avoid problems. This was driven by a downloaded script but the app needs to be able to fall back on the previous script if access is not immediate, and to know to back off trying next time if it previously failed. Having observed the behaviours of other apps with on-line components it was painfully obvious that an app could easily stall, waiting for some network connection. This is common and unacceptable: the user should never get to see this. If some network access stalls or is slow the app needs to be able to quickly move on as if nothing had happened.
We have also now started using Google Analytics in earnest. This is a refreshing step forward. It is so much more comfortable to be working with exact knowledge rather than basing decisions on intelligent guesswork. This has allowed us to gain a better understanding of user preferences in artwork, which was a good step for us to take.
Another dominating thread has been even further work on the card game Spades. The last issue came with the announcement of a new version and detailed how this had been developed and how much stronger it was. That was not the end of the line though as clearly we still needed to make more progress. This has been furthered by close work with the academic research project UCT for games and Beyond, which has not only been a great sounding board for discussion and ideas but has actually come close to creating the holy grail of a general system for utilising MCTS (Monte Carlo Tree Search) across any generic game. Our latest Spades, just released, contains a significant body of code created by them. This has been blended and manipulated to work with the existing Spades evaluation to create a genuinely generic framework that can be shared by all future coming card games and then on to other games too.
The above work has taken AI Factory deeper into a domain where your AI mindset really needs to change. So much AI work deals with precise prescriptive methods where solutions have an exact value. Nowhere is this more so than in the whole Minimax genre of analysis where a chosen move can be precisely explained by a concrete move variation, so the reason for a move choice can be exactly explained.
With MCTS and imperfect information games, solutions tend to be statistical, offering no clear solution path and even using variations which are not even legal to impact the move choice. This may sound bizarre and improbable, and indeed it is. However intuitively difficult this may be to understand, it is likely that human intellect depends more on such methods than they do on the exact prescriptive methods such as minimax.
Again we attended GDC 2013 San Francisco, and logically this should have been reported in the Summer 2013 edition, but lateness has moved it to this issue. Attending these events is about re-connecting with clients. Some of these, which we rarely see, are in the UK, but it makes better sense to meet up with them thousands of miles away in big gatherings such as GDC. With some 28,000 attendees and over 500 presentations, GDC is an important event and a good opportunity to track where the industry is going. We need to keep up and not find we are making ourselves irrelevant. We need to keep re-inventing our plans in this industry that changes so quickly.
Our main focus this year was to seek out effective retention mechanisms. It is clear that so many apps address this very well and have achieved great success. We need to learn from their example.
We have a gradually increasingly capable daughter here who is now 2 years and 4 months, who is getting alarmingly good at games. It is interesting and instructive to see what she likes to play. Being highly single minded she refuses any attempt to show her games or how to play them. Any attempt to do this and she takes her tablet elsewhere. Given this it is interesting to see what she likes. After switching and changing she currently has a strong favourite in Angry Birds, which is entirely in tune with popular choice. She was never shown this game but is now really very good and get irritated if she does not clock up the full 3 stars. Her second choice is Toy Tanks and, reassuringly, her third is our own Sticky Blocks, where again she clocks up some perfect solutions from time to time. Watching her play these games also helps us understand what is intuitively obvious to understand in game controls, as she cannot yet read, so depends utterly on how the game is presented to not only know how to play it, but what the game is about at all.
Interesting times.
