Suggestions to improve the A.I

Discussion and questions about the latest version of Chaos Reborn. Not for bugs, but for comments about the game play.
User avatar
goldendragon76
Posts: 441
Joined: Mon May 19, 2014 2:02 pm

Suggestions to improve the A.I

Post by goldendragon76 » Mon Dec 29, 2014 8:41 pm

I have been playing a lot of games against the A.I lately in between my async games and I have a few (well ten!) suggestions to improve the A.I's win percentage...

1. Use mounts better - Not being able to use mounts would seriously harm any player's win percentage and the A.I is no exception. I've only seen the A.I get on a mount twice but I think it just did it by accident both times.

2. Better use of Disbelieve - When any creature is cast it may be an idea for the A.I to label it with a suspicion rating and then refer to this when it chooses to use Disbelieve and be more likely to target a more suspicious cast. The labels could range from 0 (most definitely not an illusion) to 9 (very likely to be an illusion). A skeleton cast when there is a 50% chaos bias would be ranked 0 on the suspicion rating as a simple sum would tell you it was a 100% cast. An elephant summoned would be a 5 and a sapphire dragon summoned with zero law bias would be a 9. You could also factor in what staff the casting wizard has so staff of neutrality would push the elephant cast down to a 3 and staff of law or dragons might lower the dragon summon down to a 7. You would not want the A.I to disbelieve strong creatures only as you would just game it by making all your weak creatures illusions but I think it would play a smarter game if it tried to disbelieve a dragon more often than a rat.

3. Choose targets for Disbelieve/magic attacks based on risk to wizard - i.e don't target creatures on the other side of the map that are currently no threat to you when there are other nearer creatures posing a more imminent threat to your wizard. If I was in a battle against player 2 and players 3 & 4 were in their own battle on the other side of the map I would be focused on my fight and I would leave players 3 & 4 to theirs.

4. Keep wizard away from danger - The A.I does a pretty good job of keeping out of bolt range and away from danger but occasionally it seems to throw caution to the wind and move in for an attack on a creature where it would have been better to retreat.

5. Randomise starting order so A.I wizard does not always go last - I've read through a bunch of threads about the advantage the player going first has and forcing the human player to always go before the A.I wizard does not help its chances. The starting order should be randomised in games against the A.I and games against human players.

6. Act according to threat - i.e don't waste bolt on a Rat or treat a tree as if it was a Giant about to stomp on the wizard.

7. Take notice of V.P's - If V.P's are on and within 5 turns of end then retreat and defend if well ahead, advance and attack if well behind. Not many games against the A.I go the distance but as far as I can tell the A.I does not currently do this.

8. Use mana well - Sometimes to boost spells, sometimes save for mega spell. I can not see how well the A.I is using its mana but I have not seen it use a mega spell once even when I know it must have loads of mana.

9. Take notice of elevation - Try to position units on higher elevation for ranged and normal attacks to get the highest possible kill percentage.

10. Implement 8 player games - It is easy to beat a wizard that is flailing about erratically but good luck against 7 wizards doing this.

User avatar
Tess
GOD
Posts: 2217
Joined: Sat Feb 01, 2014 7:23 pm
Location: Truro

Re: Suggestions to improve the A.I

Post by Tess » Mon Dec 29, 2014 10:16 pm

I agree with all of those. The most obvious change required is better analysis of threats both when casting disbelieve and when deciding what to attack. Disbelieve should take account of the casting chance of the creature but also how much of a threat it poses to their wizard or creatures. That's not just a matter of attack values but also distance away and how far it can move in a turn and whether it has ranged attacks, or is undead etc. Should be possible to create a 'threat value' for each creature as you say, which will dictate which creature the wizard targets with spells or creatures.

Number 10 is obviously the most important :)

I'm finding the AI offers a fair challenge in 4-wizard battles, simply because I'm usually running out of spells by the time it comes down to 1v1 and have to take risks.

User avatar
NoWorries
GOD
Posts: 5039
Joined: Fri Apr 18, 2014 4:17 am
Location: Baldrlux.com
Contact:

Re: Suggestions to improve the A.I

Post by NoWorries » Mon Dec 29, 2014 10:26 pm

These are all good suggestions. But the best way to improve the A.I. is to advance development of the A.I. customization features promised for Wizard Lords and above who are set to guard regions or realms. If these setting options could be made fairly granular (e.g., a player could configure their character's A.I. such that if creature with base chance of 30% or less is cast within the first two turns, and no threats within 5 hexes, then cast disbelieve) or at least modular, then during this Early Release phase Snapshot could collect hard data on which A.I. configurations have the best win/loss ratios. Hypotheses are great. But empirical data will likely surprise us... and deliver more competitive results! :)

Meanwhile, perhaps Snapshot (if they're not already) could have A.I.'s configured to have randomized action/reaction criteria, compile win/loss data, and configure accordingly.
___________________________
  • One Theme to rule them all, One Theme to find them, One Theme to bring them all and in the Chaos bind them.
    NoWorries plays as Baldr, God of Wonder & Progress | Compete together with Baldr in the Order of Baldrlux

User avatar
Tess
GOD
Posts: 2217
Joined: Sat Feb 01, 2014 7:23 pm
Location: Truro

Re: Suggestions to improve the A.I

Post by Tess » Mon Dec 29, 2014 10:54 pm

AI that learns by playing against you (or indeed by playing against everyone) would be amazing of course.

User avatar
NoWorries
GOD
Posts: 5039
Joined: Fri Apr 18, 2014 4:17 am
Location: Baldrlux.com
Contact:

Re: Suggestions to improve the A.I

Post by NoWorries » Mon Dec 29, 2014 11:38 pm

A.I. wrote:I'm sorry, Tess. I'm afraid I can't do that.
___________________________
  • One Theme to rule them all, One Theme to find them, One Theme to bring them all and in the Chaos bind them.
    NoWorries plays as Baldr, God of Wonder & Progress | Compete together with Baldr in the Order of Baldrlux

User avatar
SpiteAndMalice
DEMIGOD
Posts: 6276
Joined: Sun Apr 06, 2014 10:45 pm
Location: UK
Contact:

Re: Suggestions to improve the A.I

Post by SpiteAndMalice » Tue Dec 30, 2014 11:22 am

I think it would be good if the AI accidentally made mistake in the way that human players do.

i.e. Accidental dismounts.
Image

The Guild of the Cephalopods is now recruiting - We embrace all.

- Chief Tentacle Arranger in The Guild of the Cephalopods.

Granthor
Posts: 377
Joined: Sun Dec 14, 2014 11:42 pm

Re: Suggestions to improve the A.I

Post by Granthor » Tue Dec 30, 2014 11:36 am

Is the likelihood that you would try a disbelieve related to threat though? For me, I'll only disbelieve a charging creature if I have no other options - even if I suspect it might be an illusion, it's often a safer bet to use a bolt, subvert or ranged attacks than risk a disbelieve failing. I tend to just ping disbelieves across the map if I have nothing vitally important to do that turn....although this is descending into a strategy discussion here...


Completely agree with you on how they should prioritise their use of magic attacks better. Shadow tree's movement of 0 and tangle vine's attack of 0 should absolutely make them low priorities for AI, but that doesn't seem to be the case now.

User avatar
Tess
GOD
Posts: 2217
Joined: Sat Feb 01, 2014 7:23 pm
Location: Truro

Re: Suggestions to improve the A.I

Post by Tess » Tue Dec 30, 2014 1:24 pm

Granthor wrote:Is the likelihood that you would try a disbelieve related to threat though? For me, I'll only disbelieve a charging creature if I have no other options - even if I suspect it might be an illusion, it's often a safer bet to use a bolt, subvert or ranged attacks than risk a disbelieve failing. I tend to just ping disbelieves across the map if I have nothing vitally important to do that turn....although this is descending into a strategy discussion here...
Definitely. It's one thing to identify the size of the threat, another thing to decide how to deal with it (disbelieve, creatures, magic attacks, bolt, wizard attack etc).

User avatar
Ped209
Posts: 1622
Joined: Wed Mar 26, 2014 1:40 pm
Location: UK
Contact:

Re: Suggestions to improve the A.I

Post by Ped209 » Tue Dec 30, 2014 2:59 pm

NoWorries wrote:These are all good suggestions. But the best way to improve the A.I. is to advance development of the A.I. customization features promised for Wizard Lords and above who are set to guard regions or realms. If these setting options could be made fairly granular (e.g., a player could configure their character's A.I. such that if creature with base chance of 30% or less is cast within the first two turns, and no threats within 5 hexes, then cast disbelieve) or at least modular, then during this Early Release phase Snapshot could collect hard data on which A.I. configurations have the best win/loss ratios. Hypotheses are great. But empirical data will likely surprise us... and deliver more competitive results! :)

Meanwhile, perhaps Snapshot (if they're not already) could have A.I.'s configured to have randomized action/reaction criteria, compile win/loss data, and configure accordingly.
I would've thought the customization of the AI would be limited to sliders for things like: Aggression, Suspicion, Risk Taking, Magic/Creatures, Sprite Lure(?)

This would give the algorithms numbers to factor into decision making. Setting individual behaviours for certain situations might be a bit too much as there would be thousands of situations to account for and each of these rules would have to factor in with each other and all of the other AI algorithms to figure out the best move.

I think ultimately while these sliders will allow variety in opponents there needs to be solid decision making algorithms that take into account all of the necessary factors in the first place and only snapshot can put these in place.

But of course this is only draft one and I have every faith that this will be achieved.
Chaos Reborn Community/Tournament Website: http://www.lawandchaos.co.uk

User avatar
NoWorries
GOD
Posts: 5039
Joined: Fri Apr 18, 2014 4:17 am
Location: Baldrlux.com
Contact:

Re: Suggestions to improve the A.I

Post by NoWorries » Tue Dec 30, 2014 3:25 pm

Oh, to be clear, I concur that in the final release sliders and such make sense. But generally those of us who have spent $50+ to back Chaos Reborn may be a wee bit more obsessed than the casual gamer. Thus, we're more likely to be enthused when Snapshot conspicuously engages our assistance to play as play-testers in configuring our AI-controlled Wizard Lords, Kings, etc. during the Early Release.

So, if the AI box was opened up as much as possible to adjust as many different settings as make sense during the Early Release phase, then Snapshot can obtain data on what settings have the highest win percentages. And then we'll also have better data on for what the optimal settings which might be used when labeling the sliders that no-doubt will need to come in to play after the game is finalized for release.

In other words, Snapshot should baldly exploit us as playtesters... just give us the tools to playtest. :)
___________________________
  • One Theme to rule them all, One Theme to find them, One Theme to bring them all and in the Chaos bind them.
    NoWorries plays as Baldr, God of Wonder & Progress | Compete together with Baldr in the Order of Baldrlux

Locked