good AI, bad AI

I recently played quite many games of Race for the Galaxy and Blue Moon against computer opponents, using programs downloaded from keldon.net. I have been enjoying the games very much, which surprised me a little. I have written about why some computer versions of boardgames spoil the games for me. So I wondered why I enjoyed these games so much.

The main reason I dislike playing boardgames against AI's is it makes the game feel very mechanical and formulaic. Real computer games, e.g. Civilization, Pharaoh and its siblings, the Total War series, Europa Universalis 2, are enjoyable because despite their AI's not matching human intelligence, they do many things that boardgames can't do. They handle many complex calculations behind the scenes allowing complex simulations, they can manage a real-time aspect easily, they run complex scenarios and allow much content to be programmed and presented, and of course, they have good graphics and sounds too. As for why some computerised boardgames turn me off but some hook me, I think it boils down to just one simple reason - hidden information.

St Petersburg (computerised version) didn't click because throughout the game, the cards available to be bought is open information, and cards that have been bought is open information too. So after a while you can more or less figure out how the AI's make their decisions. Same case for Yspahan. There is randomness in these games, but after cards are drawn or dice are rolled, it's all open information. However, in Blue Moon and Race for the Galaxy, you can't see the AIs' cards. I guess in a way I'm saying they appear more intelligent than they actually are because of hidden information. I can't easily figure out how they "think" because I don't know what they know. But still, they are pretty good and they give me a good challenge. Actually, I should be saying the programmer who programmed them is pretty good. In the recent 58 games of Blue Moon played against AI's, I lost 60% of them. In the recent 88 games of Race for the Galaxy with 3 players (myself and two AI's), I won 32% of the time, came in second 42%, came in last 26%.

When playing Blue Moon, once I find a combination that is hard to play, I keep playing it so that I can learn and improve. I find playing Khind against Mimix quite hard, and playing Flit against Terrah even more difficult. Sometimes I swap races, so that I can learn from how the AI plays the "weaker" race. It turns out that the AI can beat me too, but I do learn a few tricks in the process. In Blue Moon it is slightly easier to see through the program logic, because each race deck has a fixed set of 30 cards. So after a while you know some of the common tricks when playing one specific race against another, e.g. playing a character card requiring your enemy to play a booster card, when you know your enemy deck has no or few booster cards. Race for the Galaxy has many many more cards than Blue Moon, so it is not as easy to work out how the AI thinks.

I enjoy learning from the AI's. I only play these games with very few people, so there are many things that we have not explored or even thought about trying. Despite my 400+ plays of Race for the Galaxy, I still think I'm a mediocre player.

Sometimes I'm sloppy when I play Race for the Galaxy against the AI's. I don't watch their tableaus, I don't pay attention to what actions they have chosen and I don't bother guessing what they will choose. I find that when I'm lazy, I tend to lose more. Sometimes when I play a long stretch of games, I win more in the earlier games, and as I become sloppy and impatient, I lose more and more of the later games. I think this shows that Race for the Galaxy should not be played in a solitaire way. You do need to leech off your opponents' actions in order to do better than them.

In conclusion, I lied about good AI's and bad AI's. The AI's are all good. It's just how we perceive them. But then "not-so-transparent AI's and transparent AI's" is not exactly a catchy title.

Playing Blue Moon against the AI. I played Flit against Terrah. It sucks when all your cards are non-character cards. You are forced to retreat. See how my cards are all greyed out.

A game of Race for the Galaxy that I won using a novelty goods consume strategy. I had both Free Trade Association and Consumer Markets, and I had four novelty goods worlds. I also had the Alien Toy Shop, which could consume for 4pts (if I double-consumed).

I thought I did pretty well in this game, but I still lost. I had three 6-cost devs, Terraforming Guild (10pts), Prospecting Guild (14pts!) and Rebel Alliance (6pts). The other two AI's also had three 6-cost devs each! Scores were 57, 51 (me), 48.

This is one pretty extreme game which I won 81, 23, 15. I played Galactic Federation very early, and eventually had five 6-cost devs. I also won three first-to objectives and one most-of objective. Playing a 6-cost dev early is a strategy I have never thought of myself, even after 400+ games. I learned this from the AI. I used to think that 6-cost devs are just too expensive and would stunt further progress. Now I realise they can be very powerful under the right circumstances.

A close game in which I lost by 1pt. In the last round, I had the choice of ending the game there and then, or extending it by another round. I developed Uplift Code. I had the choice of whether to discard R&D Crash Program to pay for it. If I had done that, my tableau would not have reached 12 cards yet, and there would be another round, in which I could then develop Pan-Galactic League (worth at least 7pts), and possibly settle one more world if one of the AI's chose Settle. I was too lazy to do detailed analysis and just decided to end the game early. And I lost by 1pt. In hindsight, I probably should have stalled one more round.

Takeover in action (I had the Imperium Seat, which could takeover Rebel worlds). I have played many games with the Rebel vs Imperium expansion before, most with takeovers on, but even so takeovers were rare. I think I have only done it once with the physical copy of the game. So it was fun to be able to do a takeover again. Naturally I targeted the AI which was closer to me in points (top left).

This takeover did not changed the result of the game. I would have won anyway.

Another extreme game where at one point 7 out of 9 cards in my hand were 6-cost devs.

I eventually developed two of them, but I lost by a big margin, 57-45-39(me). The winning AI had four 6-cost devs.

One very blue game. Six novelty goods worlds and Free Trade Association. Other than my start world (yellow - Alien), everything else was blue or blue related.

Another takeover. Before...

... and after.

A developer dream come true. I think I chose Develop for all but 2 rounds (or thereabouts). By mid game, whenever I chose Develop, I would draw 3 cards, and get a -4 discount for developing. Also both Galactic Federation and Galactic Bankers rewarded me for developments.