AI is terrible

[furrykef]

I've played three games (once as the US, twice as the USSR) against the AI, and even though I've never played before, I won all three times. Sure, I had some help from twilightstrategy.com, but only so much. I'll admit I won my second game by a fluke (we headlined "We Will Bury You" and "Duck and Cover" at the same time), but I also had a very strong VP lead (16 points; I was headlining WWBY to put me at 19) as well as a solid board position, so I was quite likely to win anyway.

I'm not too disappointed in my purchase, but I do think a game at this price could stand to have AI a little better than this.

[Antti]

The AI has DEFCON suicided on me a few times in what can't possibly have been unavoidable situations (e.g. AI is playing USSR, coups a BG in AR1 and then plays CIA on AR2 at DEFCON 2).

[KustomShoota]

This is what I was also wondering...I played through 5 turns for the very first time on the physical board game last friday being half taught for 6 hours, a few days later I bought the game and won 2 matches using the twilight strategy website for each and every decision, also taking 6ish hours to complete a single game.

At this point, about 4-5 days later, I'm not saying I win every time, but I can play a game without the website open and win most of the time in about an hour 15 mins to hour and half. As the US in case it matters. The first two 6 hour games I won as USSR because that's what I used for board game experience.

I wouldn't have expected the AI to be all that great since this game has so much nuance, but I sort of wonder what the equivalent handicap is to make it more like playing a good player. Since I have a long way to go for that obviously.

[Bleh!]

Great another Ai thread, dude if programers could create human like intelligence we would be in a new golden age of technology. This game is very deep and if you didn't essentially cheat using a strategy guide then it would of taken you quite a lot longer to win your first game. I purposefully didn't look at any strategy guides and it took me 6 games and about 7 or 8 hours to beat the Ai for the first time playing as the US, I think you are expecting a bit much from a $15 title.

[Narkulus]

The devs acknowledged the AI was weak at launch and said they would improve it in the future. I think atm they are probably concentrated on getting iOS out the door.

[furrykef]

Originally posted by Bleh!:

Great another Ai thread, dude if programers could create human like intelligence we would be in a new golden age of technology.

I am a game programmer and have programmed AI, so I know a little about this subject. This AI stinks.

I would happily program a better one if only I had an API for doing so.

if you didn't essentially cheat using a strategy guide

That's a little strong, I think. Am I cheating if I consult a book on chess strategy to learn how to play chess? I'll admit it's a little cheaty to be doing it during a game (and definitely a no-no against a human if you don't have their permission), but I'm getting to where I don't need to consult the guide during the game anymore. I'd think of it as using training wheels.

[KustomShoota]

Yea I wasn't critiquing the game in any way with my post, so I don't want to be associated with that notion, I was just wondering what the starting handicap should be to equate to playing a relatively experienced player. And then for fun, what it would be for a tournament level player? lol.

-----------------

I just had USSR AI on turn 1 try to space race DeGaulle... am I missing something? lol

[jcrory]

I agree that it is easy to beat the AI, but TS really is a complex game...I look forward to an AI improvement in the future. I am an experienced player (200+ live games) so right now I am giving the AI a +7 handicap for a challenge.

I think the AI's biggest strategy problem right now is that it plays into midwar regions on turns 1 and 2, when it is necessary to play into early war regions that have not yet been scored. It also has a habit of couping non-battlegrounds in situations when defcon is above 2, allowing the opponent to get coup important battlegrounds. It wastes turns trying realignment rolls that it is not favored to win. And of course it just does dumb things sometimes like commit defcon suicide unncessarily.

[Threeist]

Full Critique of the AI, using the Turn Zero expansion and additional cards:

1) DEFCON suicide blunders. There are a lot of DEFCON suicide Cards: CIA Created/Lone Gunman, Duck and Cover/We Will Bury You, Ortega, KAL Shot Down, Grain Sales, Missile Envy, Kremlin Flu, Tear Down This Wall. The Soviet Player, short of DEFCON dropping to 2 in the headline phase, always has the chance to give up a coup attempt and disarm a suicide card, but this is never done. I'd go further and add the AI even space races annoying cards over DEFCON Suicide ones.

2) Bad Card Misplays. DeStalinization? If you are playing against the Soviet AI, go for it...the AI will never make you pay for it. Five Year Plan? This is a beautiful way to dump very toxic scoring cards for the Soviet Player, and the AI never does this. Perhaps more bizarre, the AI doesn't seem to wait on using cards like Suez Crisis or DeGaulle to maximize their effect.

3) The two to one trade. The AI becomes extremely predictable when it wants a battleground to be disputed, constantly paying 2 influence to fight for control. This allows for full out influence piracy.

4) European Control. This is a win condition of the game, and yet the easiest way for the Soviets to score a win is Europe. The AI doesn't seem to much worry about a East German/Austria/Yugoslavia buildup with the intent to simply realign Italy, Germany and march west. The Early War is strong for the Soviets, and it makes sense that optimal play in Europe can really put the USA on the ropes, but this isn't that. There is no all-out fight to break European Control by the American AI, and the Soviet AI generally doesn't try all that hard to seize Europe either. I've never lost a game because of European Control to the AI.

5) Finally, the AI never considers the combination of cards to increased effect. The AI doesn't do Red Scare/Purge + Blockade, or Bear Trap/Quagmire. The AI doesn't try to throw Sadat or Camp David before Muslim Revolution, or use Captured Nazi Scientist before its space race attempt (increased odds of two steps that way).

The AI has won many games against me, but generally only because of poor DEFCON management on my part or a 20 VP Dash I couldn't stop. That in itself says much.

[Benkyo]

Only a couple of points to add:

1) The claim by the developers that play data would be used to improve the AI "later" is a vacuous one, IMO. The data alone won't suffice, the AI would have to be significantly updated to make any real use of it, and I suspect that investment won't happen.

2) The guy who thinks he could write a good AI: I don't doubt you (many people) could write a better one that would still be very weak, but I don't think a good one is possible by anyone without a near chess-like database of plays to draw from and some seriously top-tier AI programming skills.

3) There is no handicap that can make the AI challenging to play against. The AI doesn't make good use of a +/-10 handicap, so once you hit a 99.99 % win rate against the AI without a handicap, you'll still have a 99.99 % win rate against the AI with a +/-10 handicap.

4) Play people, people are good.

[Riddley Walker]

They could start with some very basic things to improve the AI-

1) for instance occasionally having it actually investing in regions it holds a scoring card for, instead of just wasting scoring cards on regions dominated by the player.
I could swear it DID invest in regions, until recently. Was there an update? Am I imagining? Maybe it has to do with the optional cards DLC I'm using? Weird.

2) for some reason in the very last and crucial round in my games the AI always wastes cards on the space race.

3) AI investing in non-battleground countries even though they are not a gateway to anywhere nor will change the score.

4) AI wasting very good events for influence

5) AI using too low of influence cards to conduct coups

I'm sure there are many other things.