Artificial Intelligence

Home / Blog / Artificial Intelligence

Artificial Intelligence

Introduction

As you have already noticed, AI in Brass is not very strong. However this is not because we didn’t manage to develop stronger AI. Actually, we targeted and planned for this AI level in the game. It’s a software development classic: “this is not a bug, it’s a feature”.

It is not a bug

Let me explain. As you may remember from our network game post, the main reason for creating Brass Mobile was to allow network play. Our assumption was that most of you will want to play with real opponents and the AI exists to assist fresh players in learning the game. Our first usability tests and interviews we conducted during closed beta seemed to confirm this initial hunch. Most of the players reasoned that Brass is too complex and great AI wasn’t expected. This feedback bolstered our planning.

It was only during public beta that we started to hear differently. Some players do feel that the AI isn’t strong enough. This was the first sign that we made some wrong assumptions and that some people actually would preference the AI game over the network game (mainly to avoid all the logistics and back and forth dealing with timeouts and players who might have left the game). We also heard from people telling us that the game of Brass is quite fast due to the newspaper  feature that allows you to skip the AI animations. This is another incentive to play with AI as you can theoretically finish the game in several minutes and not 2 hours as in the traditional board game.

We’ve listened and made a few improvements to the AI before the release. The AI in public beta could win a 4 AI game with a 70 points on average. The improvements we made moved the AI score to around 90 points. This is the average score most new Brass players will get in their first games before they really learn to play. Our current statistics are not far from this. Here is a snapshot of the win to lose ratio among all games played with AI since the launch:

As you can see only 56.3% of games end with a human player winning the game. Another screenshot represent the average scores in a 4-player offline game:

The average points the winner has is a little above 110 points (not too far from AI scores). This is an average score so it means that some players are much worse and some are much better. Our record holder has 198 points. I believe that many of you end your games in the 160-180 points range. For you the AI Level – Hard is a must.

The issue we have with the current AI is that we can’t improve it further without a heavy redesign. Again, this is a feature not a bug. When we planned the AI we knew that at some point we would extend the game with additional maps or a 2 player variant. Our goal was to have a single AI that can handle many maps and the number of players. The elegant way to achieve this is to make a more generic AI. And by generic we mean that it is not profiled for the Lancashire map. For example: the perfect map spots are not hardcoded, we try to detect them using an algorithm (ie: number of connections to region or distance from Port). In this way it will work on any map. It will not be as smart as hardcoded values but it will work. For this same reason we can’t implement the strategy to block Barrow-In-Furness or Birkenhead as th are unique to the Lancashire map. Remember, we believed players were mostly interested  in the network game so we didn’t put the same level of detail into AI.

Summary

Now we know that AI Level-Hard is a must-have feature and we do plan to implement it. However, the project is a bit complex and will take some time. We need to redesign the AI and change the way it works so we can give it some planning ability and map specific behaviour. This means we can’t give it to you immediately. We’re grateful for your patience on this issue and will let you know the timeframe for implementation when we have it.