I’ve written before about one of the major design decisions that went into this strategy game. It’s generally called hidden information. It’s the equivalent of the two cards that are face down in a game of Texas Hold’em. If you were playing a real boardgame against a “mechanical turk” that was acting as your opponent then in most games you would see all his playing pieces on the board. In a game of Chess there are no hidden pieces. The artificial player makes a move and you can usually quickly assess where it lies on the bell curve of competency. With hidden information things become clouded. Uncertainty is much harder for an AI to deal with than for a human. It can also make an AI look stupid when it’s being rational and smart when it’s being stupid. I got an email today from a customer (We’ll call him player X) with a legitimate question about why the AI waited five turns to attack him to finish him off.
“Anyways, I have just played my first HUGE game vs 3 AI opponents and got creamed 😛
Nice job on the AI…however……they just won’t finish me off!
I have attached the save file. They seem to just sit there and not attack.”
Right afterwards he sent this:
“LOL…well I decided to advance the game a few more turns and sure enough they finally attacked me a finished me off.
It just seemed strange that I had the weakest unit defending the base and the AI just sat there for maybe 5-6 turns doing nothing 😛
Anyway, be good to hear what you have to say about that”
I was curious myself and of course the stomach starts to turn thinking there is some major goof on my part. To be honest there are still major goofs in the machine I am sure, but with such a complex contraption there are bound to be. You just have to keep whittling away. Anyway, I loaded up the save game and flipped the switches so I could have the “God View” and see everybody. I had to chuckle because it immediately became apparent what was going on as the hidden world beyond Player X’s lone isolated and besieged stronghold was revealed. The middle of the map was a cauldron of conflict. The Xenopods who were besieging Player X’s outpost with two armies one of which had a Corrupter Bio Mecha and a Psyker unit. He was down to a lone mobile supply unit. To the north however, the Free Mutants had just captured an indie facility and it was obvious that the Xenopods were starting to queu up some defensive goals to respond to the Mutant incursion. But the biggest problem was right next door unseen to Player X.
Tyranicus had shown up with his legion to lay claim to the Imperial Stronghold of Player X as well. The Xenopods were busy with bigger fish to fry and the 3 AP to assault the place were being spent elsewhere. Now this is interesting as well. What would a human do? He could decide to just sit there and see what Tyranicus does. The forces are pretty evenly matched. He could quickly take the Stronghold and get it’s defensive bonus. He could move another strike force towards Tyranicus and make the odds favorable if he brings him to battle. One of the key conclusions to make was that the human position was insignificant but whether or not the stronghold itself was important was subject to debate. Supply was not an issue for either side. The AI decided on the last one and spent several turns gearing up the war machine towards the point of contention. I think I would have chosen to take the Stronghold immediately. Why did the AI decide to do that. Well it has to do with priorities and goals and the deterministic way they are processed. The AI was weighing Mutant incursion and Machine threat and deciding that it needed to get some firepower into the area to make the odds better. The goal priorities and the Action Point costs associated with them didn’t leave 3 action points left over to conduct the assault. The consoling factor is that some human players might have made this choice. It’s not clear cut smart vs. bonehead.