Hey everyone! My name is Matt Brand. I work at GameTheory in Burlington, VT, and I am the primary developer on the digital version of Dominant Species by GMT Games. I am finishing up the Artificial Intelligence (AI) portion of the game right now, and wanted to give a little window into the creation of the AI. I’m going to give a little summary of the complex nature of the AI for this game, and then some of the strategies that Kai Jensen (Developer of the Dominant Species boardgame ) and I came up with in order to facilitate an interesting and fun play experience.

Dominant Species (DS) is a game with a lot of complex decision points. It may not seem that way when playing the game as a human, but the ways in which some decisions are made as a player involve a lot of underlying logic and assumptions that need to be taken into account when making an AI come to similar conclusions. Even some seemingly simple moves take layers of analysis in order for the AI to arrive at a similar decision that a human player will make.

The main reason for the deep complexity in DS is the breadth of types of actions that the game involves. For example, choosing and placing an element in Abundance, which really comes down to what elements can support the player the best on the board, is relatively simple, while moving animals around during Migration in order to maximize different results gets way more complex. Both involve a basic goal of maximizing the amount of tiles on earth that the AI desires to dominate, but the ways that is accomplished in each of those 2 types of actions are very different. And those are only 2 actions out of the 12 types, not to mention the Dominance Cards, which involve another entire range of different types of behaviors and decisions.

Another reason for the complexity is the turn structure. DS is unlike most other games. Generally in a game each player takes one turn, does an action, and the game advances to the next player. There are the same number of turns per round per player, and the actions done in each turn are relatively similar. DS is very different because it has varying amounts of actions per player, which changes over the course of the game. It progresses in Initiative order in the Planning Phase, each player taking a turn to place an Action Pawn on the desired action. But then once it gets to the Execution Phase, the turns are made in the order the players have designated, so the same player could take 2, 3 or even more turns in a row. (This also presents a big challenge as we design the multi-player functionality we’re adding to the game later this summer, but that’s another article.)

These complexities make for a challenging design experience for the AI . They also give rise to a great amount of variation that can be capitalized on in order to make a dynamic and gratifying experience for the players!

Kai and I came up with a basic flow for the AI. The first few rounds involve certain decisions in the Planning Phase which we call the early game strategy. During that time, the AI prioritizes certain actions over others, and chooses the prioritized actions, depending on what makes sense given the actual layout of the board and current states of the players. An action is only chosen if it hits a percentage roll, which is determined by different criteria for each action. For example, Adaptation is towards the top of the priority list in the early game strategy. When deciding whether or not to use Adaptation, the AI will generate a percentage through a calculation of whether or not it has already used Adaptation this round, which Adaptation slots have already been taken by other players, what Adaptation elements are available, and how many elements the player already has. From those criteria the AI generates a base number that it must hit in a random throw from 1 to 100, and if it hits, the action is taken. This makes it so that the AI will generally behave in certain ways, but it will not repeat things exactly due to the bit of randomness. This is really important to make an AI that is replayable, so that players don’t get bored with a robotic experience.

This is an early game view of the board. Note that the “AI Strategy” text will not be in the final version. We’re using those for testing to identify which Strategies the various AI players use in different circumstances.

After the first few turns, the AI evaluates itself and the state of the board, and decides on an overall strategy to use. We won’t spoil your fun by revealing all, but some of these strategies are:

• Tundra – where the AI prioritizes dominating on Tundra tiles in order to get points from Glaciation and the Survival Card. This strategy won’t get a lot of points early in the game, but as play progresses and more earth converts to Tundra, the player using this strategy will start to see better returns.
• Super Adaptation – where the AI has a wide range of elements that it can survive on. When the AI is using this strategy, it prioritizes Regression, Abundance, Wasteland and Wanderlust in order to maintain its advantage.
• Piggyback – where the AI rides on the coattails of the dominant player(s), and shoots for second place on as many tiles as possible. This allows for a decent amount of points during Domination, while taking the heat off the AI player, because they will not be the target for Competition and other actions that will directly hurt another player.

Here’s a pic from the Domination Phase. Note the card pop-up that allows a player to browse then select a card from those available.

Regarding the strategies used for actions during the Execution Phase, each action has a different set of logic that gets used to determine what the AI will do. The general idea for all actions is that a set of possible actions are amassed, in different prioritized lists, and then an action is chosen from the lists that makes the most sense. The lists are categorized into all possible actions, the best actions, good actions and bad actions. If the list of best actions contains possibilities, the AI chooses one. This sometimes involves a random factor, again in order to create a varied game experience; however, sometimes a particular action is clearly the obvious choice, and this action will stand on its own, outside of the prioritized lists. The AI will make the clear choice, so that it does not look foolish.

Overall this has been the biggest challenge in my career for creating an Artificial Intelligence. This is my third game AI that I have worked on, and it has been a difficult and also rewarding experience. I hope that Dominant Species players, both new and veteran, enjoy it thoroughly!

– Matt Brand

————————————-

Note from Gene: We are nearing (within a couple weeks) release of this game on the Apple Store. We want to remind you of two things:

1. It will be a free upgrade for all of you who already own the game.

2. For those of you new to Dominant Species on the iPad, the price of the game will be $6.99. The previous version was $4.99. Until we upload it to Apple, you can buy the old version at $4.99 and then get the new one as a free upgrade.

Enjoy the games!

Gene

Author: Matt Brand