How to make AI strong without cheating

For what's not in 'Top Priority Game Design'. Post your ideas, visions, suggestions for the game, rules, modifications, etc.

Moderator: Oberlus

Post Reply
Message
Author
elfstone
Space Squid
Posts: 78
Joined: Fri Jun 27, 2003 9:00 am
Location: Germany

How to make AI strong without cheating

#1 Post by elfstone »

I've got an idea how to improve the strength of the AI, so its not soo easy to crush. Maybe this would fit in the lengthy thread above, but since the whole thread has gone OT (from Saveplanes to Planet-AI and Enemy-AI and Cheating) so i create a new thread, to prevent further off-topicking.

A player that wants a challenge would pretty sure play against more than 1 AI, maybe 3-4 AIs.
Every AI could make a look into the future (explained later) and see if it has a change or not. If one player seems to be extremly strong, other AIs should try and stop him together. So, if a player (no matter if AI or human) is getting too dangerous the other AIs might communicate with each other, to attack that player.
Since there's nothing known about diplomacy, the AI will be able to send messages, so it could send a message to any friendly AI like "this player is dangerous, lets kill him, before he kills us". If the other AI also thinks the player is dangerous, they could ally against him. In a 1Player <-> 4 AIs the Player could soon face quite a strong enemy.

On the other hand, if the player is weak, and one AI growth too strong, the player has the chane to ally with other AIs to attack that "too-strong" AI.


How to tell, if a player is dangerous
=======================

Since AIs are totally dumb, and i don't know which information will be available, i have to guess a little about informations:

Number of Starsystems
Population
Tech
Number of Ships
Military Strength

At least some of those informations will be available i think, so if a one player P is best in 3 of 5 categories he would be quite a risk, so if AI A doesnt like him, he could suggest to AIs B,C,D and E to make war on him. Depending on theire diplomatic status to P they could agree or not.
Maybe even more important is the change of those informations, so if the

Speed of expanding
Population-growth
Research
Number of new-build ships

Are high, it would also indicate danger.

This way, we could help weaker players a little, and make games interesting for good players. The drawback is, that if you are much better than the AI, almost every game will sooner or later come to "me vs. rest of the galaxy". This would emphasize the diplomatic aspect more, since its important to make good friends.

I know this is far in the future, but i wanted to post the idea, since it's somehow related to the "chating AI or not" discussion going on.

OceanMachine
Pupating Mass
Posts: 95
Joined: Thu Jun 26, 2003 12:09 pm
Location: Chicago

#2 Post by OceanMachine »

Deciding who the AI should go to war with is not difficult. Many games use this sort of gang up and the leader strategy. The difficult part is making AI's wage war effectively. If you can't do that, it doesn't matter who declares war on the player.
Programming Lead

User avatar
utilae
Cosmic Dragon
Posts: 2175
Joined: Fri Jun 27, 2003 12:37 am
Location: Auckland, New Zealand

#3 Post by utilae »

That notority idea I thought, of applies here. The AI would think a player is notorious, ie very powerful, ie they notice that player as the biggest threat. And so they take action to destroy.

Look in the notority thread (in brainstorming) for more background on notority.

Impaler
Creative Contributor
Posts: 1060
Joined: Sun Jun 29, 2003 12:40 am
Location: Tucson, Arizona USA

#4 Post by Impaler »

I think one of the major flaws with computer AI's is its inability to seaze an oportunity or weakness. So a player can leave themselves open for attack and gain a huge advantage by directing all their energy towards growth.

Their should be some algotrith that claculates force-location-turn. Basicaly it calculates the forces that each side can bring to a system and on what turn they arive. Players do this all the time. I notice that one of the computers systems is unprotected, my fleet can get their in 3 turns, their defendinf fleet will need 5 turns so I attack. The same algorithum run in reverse shows me whare I should defend.
Fear is the Mind Killer - Frank Herbert -Dune

User avatar
utilae
Cosmic Dragon
Posts: 2175
Joined: Fri Jun 27, 2003 12:37 am
Location: Auckland, New Zealand

#5 Post by utilae »

Sounds like a good idea.

User avatar
skdiw
Creative Contributor
Posts: 643
Joined: Mon Sep 01, 2003 2:17 am

#6 Post by skdiw »

I agree. Also the AI should be more dynamic and interacts depending what other players do. If the human are mobilizing a certain ship like missile boats for hit-and-run raids, the AI should be smart enough to counter with anti-missile orbitals or ships.

Tekwrecker@atx
Space Krill
Posts: 7
Joined: Mon Sep 08, 2003 6:03 pm
Location: Austin, TX

#7 Post by Tekwrecker@atx »

greetings...

I like the recent trend in games for AIs to incorporate anti-routine type programming....

ie if a player does the same type of thing over and over, the AI changes up to make it more difficult...

I think players should have to stay on their toes and constantly evolve their tactics... perhaps set it up at teh beginning of each scenario for certain victory conditions to be easier, and for others to be much more difficult...
It's tough at the top, tough at the bottom, hard to survive, in a world of destruction

Sh.Tac.
Space Floater
Posts: 19
Joined: Mon Sep 15, 2003 1:16 pm
Location: Moscow-Russia

#8 Post by Sh.Tac. »

Hmm, I'd preffer that my AI does the routine, so I can dedicate to space battles or research ;-)
On the other thing, a honest AI sounds well, but heavy to realize in practice, the main approach consists in making harder start conditions for the player.
This is what you get ...

Post Reply