Another year has passed and AI Factory is still in good shape. We are a small company of just 3 people but with a disproportionally large footprint in the gaming industry. This is made possible because we run a well organised ship with well-crafted generic game architecture, making support and development very easy. We get to particularly appreciate the effectiveness of this when we have to work on the very small fraction of our games that do not fully conform (for legacy reasons) such as Backgammon. These slightly ill-fitting games demand disproportionally more effort to support.
Looking forward though, given our burgeoning large number of users, and small engineering base, we are faced with either expanding our company or contracting new games development. In practice we have focussed on the latter, improving the quality of the existing line-up of games rather than embark on something new. This is worthwhile as the quality of apps increases over time so users demand and expect more. Improving the quality of our user experience is a satisfying and worthwhile goal.
However we are still active members in the games industry and need the stimulation of exposure to the evolving culture. In this respect the last 3 years have been somewhat barren as we have been a little starved of access to conferences in this time, because of Covid.
We have now just rectified this, but should first acknowledge one exceptional product that made on-line conferences a fruitful and enjoyable on-line experience, in the absence of real attendees. We had attended the excellent iGGi conference, which acts as a bridge between industry and academia, linking up industrial needs with academic prowess. This is dominated by the games industry. AI Factory has both participated and benefited from this over a decade (see below).
This conference ran on the impressive conference software "gather town" that allows delegates to wander around a virtual 2-D space as if in a platform game. The world you explore can be directly modelled on the real conference centre or some invented layout. This is populated with meeting and lecture rooms with specific seating and with other meeting areas outside the rooms. You can move your delegate around, sitting next to who you want to sit with. If your delegate approaches delegates you can start to hear what they are saying. If right up close you can see them in a mini zoom-like window and the volume goes up. This allows pairs or groups of delegate to casually chat. You can also just wander off with a delegate to some other area for private chat. The environment is filled with sounds that grow louder as you approach such as (for example) a fountain and even bird song if outside.
This all works remarkably well. It's fun and feels personal. You can wander into an area and see who is there and who is talking to who. It does not suffer from the sterile feel of more common remote conference offerings. You can get to easily bump into and join group conversations in a way that conventional conferencing software just does not allow. All credit to Gather Town.
However we finally managed to attend our first real live conference in 3 years, this time at Develop Brighton. After such a long absence it had a slight feeling of "uncanny valley" as it was both familiar and strangely unfamiliar, as-if visiting some dream one once had. I had to miss the last day but every ounce of my days there were worthwhile, re-connecting with colleagues and ready to energetically brainstorm unexplored ideas. It was inspirational :)
As for work completed this year, a significant proportion of AI work has been into the further enhancement of Backgammon, which now beats the old program by a substantial margin. Since this is auto-tuned it has something of the common characteristic of self-learning AI that it becomes a black box that does not so easily reveal why it is choosing a play. This is not as opaque as deep learning as the tuned parameters do slot into a linear evaluation, so the parameters are exposed for scrutiny. As alluded in the earlier comments, our Backgammon does not fully follow the rules of our game architecture, so it had to be shoe-horned into our testbed tools. It also means that aspects of game control had to migrate to the UI. This invites issues as it needed some hacks to mimic conformity, which slowed things up and invited mistakes, but again exposed the benefits of good conformity. The reason for this shortfall was that the original engine was an inherited source. The new engine is a completely new architecture, but still needs to talk to the UI designed for the non-conforming old game engine.
Of the rest of our apps, Sudoku has added 1000 new puzzles. Chess multiplayer is being filled out and, on a trivial matter, all our apps have had to be re-named to conform with the new Google rules, so "Chess Free" is now "Chess" and "Chess" is now "Chess Pro". The vast bulk of releases though have been to conform with Google and advertising SDK and coding updates.
August 2022
