Forums

Search
Forums » Small World - Digital » Suggested Improvements for AI's Strategy
Show: Today's Posts 
  
AuthorTopic
renvhoek
Junior Member

Posts: 8
Registered:
July 2010
Suggested Improvements for AI's Strategy Sun, 25 July 2010 00:32
Hi All:

Perhaps players can post here recomendations on where the AI appears weak, so the developers can get ideas on how to improve it for a future version.

So far, the biggest weakness I see from the AI is that it tends to not move into decline when it should. In repeated games, I've seen it try to hang on when it should decline and move on. In one game, it still wouldn't decline even when it had only one token left.

A smaller AI issue - is it doesn't study the board well enough when entering and I've got up a halfling block. It will sometimes come on the wrong side of the "block" and severely reduce its ability to succeed with its new race.

- Jon
      
eric
-= Crew =-
Advanced Combat Training

User Pages
Posts: 3198
Registered:
October 2002
Re:Suggested Improvements for AI's Strategy Mon, 26 July 2010 19:27
Agreed, the timing of the decline is not always optimal and something that we will continue tweaking.

On the issue of a halfling block, unless you have the Halflings with Two Heroes, how do you establish a block? there are only two regions you could make impassable, and you'd need a minimum of three to cut a portion of the map off. Or did we miss something?

      
renvhoek
Junior Member

Posts: 8
Registered:
July 2010
Re:Suggested Improvements for AI's Strategy Wed, 28 July 2010 06:34
Hi:

What I meant by the halfling block, is choosing two vertical lands either below or above the center lake, thus blocking off that side of the map.

In the case I was thinking of, I picked the two below the lake, and and then went northeast, taking another land there.

The AI decides to come in from the southeast, and takes 3 or 4 lands. I was able to keep it bottled up in that corner for the rest of the game, taking advantage of both weaknesses - that it picked a bad area to start its race, and that it wouldn't give up when it was obviously hopeless.

- Jon
      
eric
-= Crew =-
Advanced Combat Training

User Pages
Posts: 3198
Registered:
October 2002
Re:Suggested Improvements for AI's Strategy Thu, 29 July 2010 01:11
ok understood. It's a non trivial issue, so no promise. Confused
      
Rikdom
Junior Member
Cadet

User Pages
Posts: 1
Registered:
June 2005
Re:Suggested Improvements for AI's Strategy Sat, 14 August 2010 12:41
Hi

I have had a similar problem the AI playing sorcerers picked the bottom left corner despite 2 hobbit holes effectively blocking the whole board.

The bot then stacked and re-stacked the single zone for the rest of the game.

Otherwise the bot is a good match for me as a beginner . .

Rik

[Updated on: Sat, 14 August 2010 14:10]

      
Lillumultipass
Junior Member

Posts: 12
Registered:
August 2010
Re:Suggested Improvements for AI's Strategy Sun, 15 August 2010 18:34
Hello,

I have played around 10 games over the past few days against the AI, and having won 9-1 (even though I had never played the board version before) I think the AI might be improved in certain areas:

1. I agree about the decline thing: in my testing, I found that the AI really tends to stick for too long with the same race, even though its prospects are grim.

2. On the contrary, in my last game, the AI chose Stout human in turn 9 and didn't go into decline at the end of the turn, which is a bit surprising.

3. What I found the most detrimental to the AI's success is the fact that, in my games at least, it seldom pays VPs to select better race combinations at the bottom of the queue. For instance; in one game, it was lucky to start with Thief Amazons but I retaliated with Hord (?) Ratmen and it was pretty much even. Then, I went into decline before him so that the race at the top of the list was Cursed Ghouls with 3 VPs on it. CLearly, it could have chosen better options down the list, but it chose the Ghouls and I pretty much decimated it.
It's not the only example I have: in all my games, I have never seen the AI pay 4-5 VPs to get what I consider awesome combos, above all at the start, i.e., it once passed on Marauder Ratmen on its first turn as a first player! I then payed 4 VPs to get the ratmen and the game was pretty much already decided!

These are the most glaring weaknesses I found, but I will post anew if I find others.
As a final thought, it would be nice to have different AI levels because it is true that for new players, the AI is already quite good. Indee, I lost a bit at the beginning, but after playing a few games with a friend who didn't know Smallworld but is usually very good at board games, I improved quite a bit thanks to our discussions and I now see the mistakes the AI is making.
      
MikkelOgRasmus
Junior Member

Posts: 1
Registered:
January 2010
Re:Suggested Improvements for AI's Strategy Sat, 20 November 2010 23:22
I played a game today and noticed two strategic errors made by the AI:

1) I was playing as Sorcerers, yet the AI continually let one of its mountain regions next to me be held by a single token, giving me the option to convert it. It did have 2 tokens in a region I could by no means get to however. I couldn't quite see the point in that.

2) I started the game, therefore the AI had the very final conquering round. It was Swamp-Something and entered the map after declining in the swamp in the south east corner, next to the water area. All regions in the neighborhood had 1 token on 'em. Yet instead of conquering its way to the swamp two regions west, it conquered the forest to the north east, thus wasting 1 VP.

      
    
Previous Topic:Essen splash screen
Next Topic:Not available on the South African iPad Store?
Goto Forum: