Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[gordythegopher]

Don`t know whether this counts as bizzare, but the bots just don`t know when they`re beaten. After clearing out 2 piles, and with only 1 worker`s village left, Defender Bot just kept stubbornly drawing victory point cards instead of ending the game. The game finally ended when I`d emptied the coppers pile, which allowed me to amass 219 points. And I`m not even a good player as you can probably tell....

[Drab Emordnilap]

Don`t know whether this counts as bizzare, but the bots just don`t know when they`re beaten. After clearing out 2 piles, and with only 1 worker`s village left, Defender Bot just kept stubbornly drawing victory point cards instead of ending the game. The game finally ended when I`d emptied the coppers pile, which allowed me to amass 219 points. And I`m not even a good player as you can probably tell....

I think it's entirely reasonable to code an AI to never take an action that will result in a guaranteed loss.

[DG]

I think it's entirely reasonable to code an AI to never take an action that will result in a guaranteed loss.

I raised this problem in Beta for the bots not finishing games. You could argue that bots shouldn't ever buy the last card to lose a game, however the game is being run for the benefit of the human customer. If the bot creates a tedious protracted endgame when it has no mathematical chance of winning the game, except by the human paymaster quitting in boredom, then it should instead lose graceful and quickly. There is nothing to be gained for anyone from bot stubbornness. (In some earlier versions the bot would actually buy a card for a close loss when continuing might have found a win, but persevere forever in games that were entirely lost.)

In multiplayer games the situation is more complex but if the bots employ a simple strategy of maximizing their score then they avoid king making problems.

[Drab Emordnilap]

You could argue that bots shouldn't ever buy the last card to lose a game, however the game is being run for the benefit of the human customer. If the bot creates a tedious protracted endgame when it has no mathematical chance of winning the game, except by the human paymaster quitting in boredom, then it should instead lose graceful and quickly.

If it's so easy for the bot to end the game, why is it impossible for the human player to end the game instead?

In multiplayer games the situation is more complex but if the bots employ a simple strategy of maximizing their score then they avoid king making problems.

I think there's a few different ways they can go here -- one can aim for maximum personal score; one can shoot for improving one's placement relative to others, or one can go for first place. If I'm in a 3 player game and I can end the game now by buying the last Province, putting me from 3rd to 2nd place, I'd be much more inclined to try and play for first still. It depends on what the intended goal is.

[DG]

If it's so easy for the bot to end the game, why is it impossible for the human player to end the game instead?

Once you've got 11 provinces in your deck you might not find it easy to buy the twelfth. I come across this problem against the bots more often then you might think.

[Taco Lobster]

Someone mentioned this in another thread, and I tried it last night to great amusement: using Masquerade to give a bot Rats.  The bot has no idea what to do with Rats, and just plays it because it's a cantrip.  Within a few turns, the bot will have a deck of all Rats and the one Copper it can buy each turn.

[theory]

Someone mentioned this in another thread, and I tried it last night to great amusement: using Masquerade to give a bot Rats.  The bot has no idea what to do with Rats, and just plays it because it's a cantrip.  Within a few turns, the bot will have a deck of all Rats and the one Copper it can buy each turn.

The thematic implications of this are hilarious.

[aaron0013]

What? A taco lobster masquerading as a rat?

[Morgrim7]

Many of the bots I've played seem to choose random strategies out of the blue, like rushing Moneylenders.

[jotheonah]

So my coding experience is all of one class in college on javascript, but it seems like the Goko bots are already coded with a brute force code that thinks things like "IF Cantrip or Money, THEN Buy as many as you can."  It doesn't seem like rocket science to have a "Buy 1" category for cards like Moneylender and Chapel.

The big problem I've noticed is they don't know/care about their money density. So when they get into a position where they consistently hit less than 3, they buy Coppers and Estates, making their deck worse and worse. Or they buy lousy 3's that don't help their deck value over Silver.

[Kuildeous]

The big problem I've noticed is they don't know/care about their money density. So when they get into a position where they consistently hit less than 3, they buy Coppers and Estates, making their deck worse and worse. Or they buy lousy 3's that don't help their deck value over Silver.

Although, coding the AIs to base their buying decisions on money density could be pretty nasty, since what is tricky for humans to track would be a doddle for a computer to do.

It wouldn't be as simple as saying "If the new card does not increase my average coin, then don’t buy it," because victory cards do decrease your average coin and will need to be bought at some point.

I suppose there could just be a tipping point. The bot can be programmed to increase average coin and then when it goes green, that conditional no longer applies.

[jotheonah]

That is a good point.  I think there ought to be some money density it won't go below ever, if the idea is to force them to never get into negative feedback loops where all they can afford are Estates so they just buy more and more Estates.

[WallyNate]

In a recent game, the bot ruthlessly chapel'd down his deck until he only had a silver and pearl divers left.  Then every single turn, he played every pearl diver he had, then bought another pearl diver.  When the pearl divers finally ran out, he bought a copper to start building up again.

[Beyond Awesome]

In a recent game, the bot ruthlessly chapel'd down his deck until he only had a silver and pearl divers left.  Then every single turn, he played every pearl diver he had, then bought another pearl diver.  When the pearl divers finally ran out, he bought a copper to start building up again.

I hate bot games like that. They are long and tedious and you are essentially playing solitaire at that point.

[Schneau]

In a recent game, the bot ruthlessly chapel'd down his deck until he only had a silver and pearl divers left.  Then every single turn, he played every pearl diver he had, then bought another pearl diver.  When the pearl divers finally ran out, he bought a copper to start building up again.

Pearl Diver is so OP. I'm sure you lost this game.

[SirPeebles]

Someone mentioned this in another thread, and I tried it last night to great amusement: using Masquerade to give a bot Rats.  The bot has no idea what to do with Rats, and just plays it because it's a cantrip.  Within a few turns, the bot will have a deck of all Rats and the one Copper it can buy each turn.

Here we are, a month and a half later and if Rats end up in a bot's deck it's game over for them.  And apparently if I switch tabs over to here so that I can complain to people who aren't responsible for it, the bot waits for me to tab back before continuing to play the 20 Rats.  It's as though Goko wants to discourage from complaining.

[Lashof]

In a recent game, the bot ruthlessly chapel'd down his deck until he only had a silver and pearl divers left.  Then every single turn, he played every pearl diver he had, then bought another pearl diver.  When the pearl divers finally ran out, he bought a copper to start building up again.

This also happens with Chapel + Spy.

Regarding it waiting while you switch tabs:  If you actually use a separate window instead of a separate tab, it continues to run.

[LastFootnote]

My favorite is when it continues to buy Warehouses whenever it hits $3, always plays Warehouses, and never discards Warehouses when playing a Warehouse. The bot just plays 4 or 5 Warehouses in a turn and buys nothing. Classic!

[Kirian]

The big problem I've noticed is they don't know/care about their money density. So when they get into a position where they consistently hit less than 3, they buy Coppers and Estates, making their deck worse and worse. Or they buy lousy 3's that don't help their deck value over Silver.

Although, coding the AIs to base their buying decisions on money density could be pretty nasty, since what is tricky for humans to track would be a doddle for a computer to do.

It wouldn't be as simple as saying "If the new card does not increase my average coin, then don’t buy it," because victory cards do decrease your average coin and will need to be bought at some point.

I suppose there could just be a tipping point. The bot can be programmed to increase average coin and then when it goes green, that conditional no longer applies.

Actually, consider the code written for the simulators out there.  Not only is this relatively easy to do for a bot, it's readily codeable.

[Donald X.]

My favorite is when it continues to buy Warehouses whenever it hits $3, always plays Warehouses, and never discards Warehouses when playing a Warehouse. The bot just plays 4 or 5 Warehouses in a turn and buys nothing. Classic!

I have reported this one a couple times.

I am told that they have an improved AI in-house; dunno if it fixes this.

[pinkymadigan]

My favorite is when it continues to buy Warehouses whenever it hits $3, always plays Warehouses, and never discards Warehouses when playing a Warehouse. The bot just plays 4 or 5 Warehouses in a turn and buys nothing. Classic!

I have reported this one a couple times.

I am told that they have an improved AI in-house; dunno if it fixes this.

Good thing they are keeping it to themselves. We wouldn't want that going public.

[jotheonah]

The big problem I've noticed is they don't know/care about their money density. So when they get into a position where they consistently hit less than 3, they buy Coppers and Estates, making their deck worse and worse. Or they buy lousy 3's that don't help their deck value over Silver.

Although, coding the AIs to base their buying decisions on money density could be pretty nasty, since what is tricky for humans to track would be a doddle for a computer to do.

It wouldn't be as simple as saying "If the new card does not increase my average coin, then don’t buy it," because victory cards do decrease your average coin and will need to be bought at some point.

I suppose there could just be a tipping point. The bot can be programmed to increase average coin and then when it goes green, that conditional no longer applies.

Actually, consider the code written for the simulators out there.  Not only is this relatively easy to do for a bot, it's readily codeable.

I find coding problems highly interesting on a very abstract level, but my actual knowledge of code is negligible.

[DG]

Of all the dodgy bot strategies I've seen, I feel that using a rebuild to trash a duchy and gain an estate is probably the worst .

[ftl]

Hah. I can even imagine why that happens. The Bot correctly figures that if it gains the last province then it loses, and so it gains an estate instead since Duchies are out?

[DG]

No. With half a dozen estates, duchies, and provinces in the supply it decided to rebuild a duchy into an estate, twice. It had been successfully rebuilding some estates into estates prior to that, so this was obviously an ambitious leap forward.