Quite often in a game I conquer an AI developped planet that has specific tiles, like bonus to population, bonus to economy, etc, being misused.
It may be that the AI was desperate and had to use these tiles for something else, or it may be that the AI is ignoring the tile's bonus. For example, a market center built on an influence tile. A research center built on a economy tile. Little stuff like that annoys me. I like efficiency, the AI does not seem efficient
I'm playing on normal, so maybe it's only something that kicks at an higher difficulty level?
I build my planets to specialize. Basically by mid game my home planet is manufacturing, 0-2 are income and the rest are all research. One nice planet maxed on manufacturing can put out a ship every 2 turns. I typically have one farm and one hospital on each planet, regardless of type. Later the hospital can be replaced with a farm if the planet is maxed.
On tile selection through soil/terraforming/etc i first look to create a perfect circle for max benefit of a coordination or other bonus building. Then I go for an oval for 2 max used coordination buildings adjacent. Barring that I pick the tile that has the most neighboring tiles.
What I want is to see is the data and conclusions Brad gets from those collected games. That could be very interesting.
First, you teach the AI the fundamentals. That is what is happening now. Then, you let the AI execute the fundamentals with nano-management, a turn by turn examination of every planet, ship, shipyard and starbase. (or whatever database you think needs to be checked) You know your AI is never going to actually be as clever and innovative and reactive as a human, so you use the nano-management as a way to make up for that. That makes a lot of sense to me.
After that comes endless fine-tuning. Dedicated play testing by millions of volunteers would be best. Stardock is using its customer base. This may be Brad's cleverest trick. The present state of AI in general cannot really learn, though they are getting closer to simulating that. Game engines have even more challenges due to smaller code and hardware. So, Brad steps in and does the learning on behalf of the AI and then teaches the AI what he can. And he does what we do, learn by studying the opponent. All this poetic philosophy is then painfully ground into actual code and we get an update.
The only problem is that it takes patience on our part. Forums don't do patience very well. I don't think there is enough notice of what has improved already and how subtle it is. I don't think Stardock gets enough credit for how much work they are actively putting into a difficult and long term project. However, all these ideas, points and conversations are a vital part of the process. Not all of these ides will make it into the final AI, but the mixture of possibilities is where the true strength that Galciv will be lies. Keep it up, folks! Just give Stardock a little credit for what they have done, what they are doing, and that they listen so much to a bunch of 4X crazed fans.
I am so excited by what I have seen so far, but it just makes me more excited about what is to come. I can understand the impatience.
whew, good to know Skynet is still a ways away. If the AI can't optimize buildings in galciv3, they are not ready to optomize human killing strategies and successfully adapt to John Connor type human resistance.
I know this is off topic, but there is a great fear of robot workers replacing humans in fast food. While its true a robot can pump out more burgers per hour and not need breaks and such, vandals could easily sabotage an all robot fast food place. It would be harder with human employees who could more easily fight back, call police, etc. At the end of the day, AI has to be programmed to think about all of these things. While human minds can not do singular things like solve calculus equations in micro seconds, we have remarkable flexibility in being able to adapt to complex situations. Maybe our genetic code is just insane in that we are actually rigoursely programmed for everything just like robots, or maybe there is more to it than that.
There are many great features available to you once you register, including:
Sign in or Create Account
