AI only human?

Proposals for AI

from GalCiv3 Forums

Having played the previous GC and other strategy games (like most of the people here I am sure) I am looking forward to this games AI with all the posts that have been running around the forums since last year. Here are some of my though that will hopefully help someone in some way:

  1. Exploits weeded out via testing: No more armies of transports wandering around without escort letting themselves blown up time after time
  2. Let us deactivate "cheating" such as seeing player in fog of war and getting additional resources. It is best to know the enemy plays with the same rules as the player else it gets uninteresting. If the enemy does get handicap that needs to be highlighted in the beginning
  3. I like the concept of improving the AI with the database of player strategies past but I do not think the AI will ever be unbeatable without cheating. That should also not be the goal. Goal should be that the AI puts up a good fight from beginning to end meaning: it does not "give up" in mid-game or after mayor losses
  4. "Only Human" parameter of the AI: It is fine if the AI is smart enough to calculate the statistically best course of action in a given situation but applying this becomes repetitive and exploitable. The AI should have the option to take a less preferable course of action even if it knows that is not optimal. e.g.: sending a transport unarmed is a risk a "smart" AI would not do but if the player does not notice the one that is sent payoff if larger since it took less resources. This aspect is something that should be applied in small amounts but lends the AI much by making it unpredictable. In a complex games such as GC is there is no best course of action like in StarCraft or in chess - at least diplomacy throws that concept out in the window
  5. Sliders would be good just as in GC2 to set all mayor parameters of the AI
  6. Cease ineffective tactics - If the attack of the AI is not efficient sending the same waves of ships at a planet defense it should recognize it as such and change strategy else it becomes repetitive and is only buffing up defense ships with experience

 

Best would be 4 difficulties:

  • Beginner: AI has reduced resources. All sliders are per-configured. Aimed towards newcomers
  • Normal: AI has the same resources as the player and uses per-defined algorithm. All sliders are pre-configured. Aimed towards normal gamers who do not want to tweak nor have internet connection :)
  • Hard: AI has the same resources as the player and uses analysis of past games from net. All sliders are pre-configured
  • Custom: Every major aspect such as handicaps, and usage of central DB for statistics is selectable and adjustable via sliders - this is for those who like to tweak the AI to their fancy

GC2 was one of may favorites due to the AI. It was not perfect but it was better then that of most other strategy games. I can hardly wait to find out in what ways GC3 is improving on it :)

15,443 views 4 replies
Reply #1 from GalCiv3 Forums Top

i believe they mentioned in one of the earlier dev-streams that the A.I. would always play to the best of its ability and the difficulty slider set at the beginning just gives the A.I more or less resources compared to you

Reply #2 from GalCiv3 Forums Top

As for your title, (and I know this is a bit of a stretch) you could say that the AI IS only human, since it is being programmed by a human (or two, led by Brad Wardell).

Offworld Trading Company
Reply #3 from GalCiv3 Forums Top

As far as the second idea I consider cheating to additional economic bonuses or penalties for no reason. I would like to see a more difficult ai instead. I disagree with the fourth idea. Why would the an take a less preferable course of action. That is why you should play an easier level. Also I think i t should be the most preferable course of action against the person it is playing against instead of a random strategy. I also would like to be able to download the multiplayer ai that is designed te same way to learn from the tactics of the player it is playing against. Besides rmembering the strategies of the players it is playing against. The air could try new stuff until it finds a strategy that works. They also need to have an option to reset the an in the options menu. I don't like the difficulty ideas I would rather have easier and harder ais instead of penalties and bonuses. Not just te strategies of multiplayer but your own also. Programmed by a human but should be adaptable to your tactics.

Reply #4 from GalCiv3 Forums Top

"i believe they mentioned in one of the earlier dev-streams that the A.I. would always play to the best of its ability and the difficulty slider set at the beginning just gives the A.I more or less resources compared to you"

Found this also mentioned in the forum, and like the idea if for not other reason then only due to the sheer effort that needs to go into this concept from DEVs. Think it is better to put that effort in refining the AI to be more capable

 

"you could say that the AI IS only human, since it is being programmed by a human (or two, let by Brad Wardell"

Agreed if the AI is using the script of a DEV since I do not think that the game will build an analytical DB and filling it with data during the game :) It is supposed to do just that with the online connection and pulling statistics and tactics from past games though.

It is of course difficult to say how it will turn out without knowing what exactly the plan is but analyzing past game statistics can easily lead to a repetitive and even less capable since in complex situations there is usually no perfect way of doing things. In such situations if the AI goes for a pattern, the user simply takes another pattern. In a game of rock-paper-scissors there is no winning pick but if someone always makes the same pick or can be easily lured into making that pick or is deductible - there is no challenge

I was actually trying to say that being random or taking the less logical route in some cases makes you more competitive. If the AI were always another human it would almost be the perfect AI to play against :)

 

"Why would the an take a less preferable course of action"

Because there is no perfect strategy to go with and if the AI thinks there is, it will probably loose to an average player due to being per-determined.

It also becomes a problem if 80 AIs from 88 are following the same strategy because that is according to statistics the way to win (it may be statistically correct but 79 of them will still loose :))

Also for making the game more life-like. Intelligent beings are not logical but highly efficient partly because making mistakes does not always lead to loosing altogether. Consider a game where the AI is hammering planets with effective wings of rocket ships and in accordance with statistics and empirical effects it will continue to produce the same ships only to be annihilated by a single new type of ship the player has produced. Works 10/10 in GC2 - need to invest into defense against the preferred weapons type of the AI and produce 2-3 such ships, experience will take care of the rest. Now consider an AI changing or even mixing its layout seemingly in random interval and random strategy

If the AI goes after statistics it will always accustom its strategy to the past which in a rock-paper-scissor fight does not help much ;)