I set the Drengin to tough. This is their best game without getting freebie money and such. Only around 5% of users should be able to beat it at this level.
|
This is what I expected. Somehow I don't think the 5% number holds.
|
Well if it said, 5% of users playing their first or maybe second game, i would say maybe. But like any AI, that is not adaptable with time, once you play enough you start spotting it's weaknesses, for example, most players I think after a couple of games get a sense on the conditions it will declare war, and work to make sure it never happens etc. T
If you know generally how your opponent is going to react, it's a HUGE advantage.
Personally I think i was a bit scared off by the reputation of Galciv's AI in my first 2 games, so I gave up too early. Plus in the first game, I didn't even figure out how to invade colonies until much later.
In fact, it's highly beatable even from an inferior (not vastly of course) position, but I'm seldom in that position for TBS games. I I've replayed some of my saved games, and it's beatable with a quick transport blitz.
You're mistaken. If it wasn't scripted in any way, it would be a self-learning neural net, which it is clearly not.
While it's not scripted, as in "do this then that then that", its possibilites are scripted.
For example, the AI has scripts telling it to put ships in fleets, and scripts telling it to send transports to the enemy planets.
|
Probably a semantic matter. By not scripted, most people mean that it does not do x,y,z actions regardless of what is happening, but rather it takes into account a lot of factors in its decision.
In that sense, it's much like a human being, like when choosing to research technology, we take into account the overall position, whether we are at war, our overall victory plan ,whether we can or should get it from trading etc.
I suppose these are embodied in APIs like CalcFullAttack(), CalcFullDefense().
I doubt if Galciv2 is the first game AI to do this of course, but properly done it can lead to a respectable opponent.
You are right in that the AI cannot from game to game, add its own APIs, while humans can figure out new strategies, take into account new factors from game to game, but that is not a bar from high performance, since top Chess AIs can't do this either.
Now, what would a truly non-scripted AI be?
An non-scripted AI would probably only given the rules and objectives of the game, and develop its own ideas of how to play it.
A non-scripted AI would be able to understand and counter any human strategy.
A non-scripten AI would be able to use any human tactic including the cheesy ones.
Such an AI is impossible to make for GC2 of course. |
A human isn't scripted (by your definition)? I would argue that using a wide definition of 'scripts', even neural nets are fundamentally, 'scripts' at the lowest levels, just doing their tasks stupidly...
The nice thing to have of course is some kind of persistent self learning program, that can figure out it's errors, and adapt to the human. Chess AIs generally don't do this , they have positional learning but this applies only to a specific position, which is less useful since it can't generalize.
Experiments in neural nets and other forms of machining learning for chess AI have generally not being as successful as standard alpha-beta programs with hand tuned Evaluation functions. That's basically the same as GalcivII really.
