GC2: Online (Not multiplayer, sort of)

I love GC2, the AI is the best in any computer games I have played. But one thing is missing from the AI that make it perfect: the ability to learn from game to game (pls correct me if I'm wrong, I will be delighted if GC2 AIs do indeed learn from the human player)

Can Stardock develop a learning AI? One that can learn from the player. It need not be able to counter the player strategy in a single game, but over a few games, It is able to learn from the player style and strategy, develop it's own strategy and counter the player.

If this can be done, I will not only trying to outsmart Stardock computer gurus in strategy, I'm also trying to beat myself in everygame, cos the AI will adapt to my play style.

In another aspect, I'm also training my own GC AI. A host system can be setup to let player's trained AI battle it out without human interference and have a ranking of player's AI like Meterverse. The system should also allow player to upload their "pet" AI for other to download and play against.

Several AI of different difficulty levels and play styles can be host on Stardock website or SDC. Player can download different AI to play against. These "official" AI can/should be release on a regural basis.

If Stardock is able to provide this, I'm willing to pay monthly/yearly suscription and not buy another computer game.
19,359 views 11 replies
Reply #1 Top
No, neural nets are a completely different system. The game would require substantial reworking from the evaluations it currently does to evaluating a players style, and it probably would be very CPU intensive.
Reply #2 Top
Yeah, but a game learning AI wouldn't work in such detail, you'd simply derive certain elements from each game, like "He did X so I tried counter Y and in 40/45 games it didn't work... stop doing it". What to bother monitoring would be the trick.
Reply #3 Top
Learning AI is one one of those things that is easy to say, but much, much harder to do.

As far as getting a challenging opponent goes, I reckon that the current method that Stardock uses is much better- updating the AI in patches. In a way, that is a kind of learning AI, since it is always being tweaked to counter common strategies.
Reply #4 Top
Yeah the poster above is right, its really hard to develop an AI and then let it "learn" your playstyle. This game already has really good AI and its better to tweak it with patches...
Reply #5 Top
The AI isn't that impressive, considering that even at the hardest levels it is easy to fool and exploit.

At which level of difficulty does the AI start to get economic bonuses?

I want to fight the AI 100% - Economic Bonus to see just how good it is.

I did try the absolutely highest setting the other day, and to be fair it did all but defeat me since I quit. The fact I recognised the fact that regardless of how hard I fought I'd slowly but surely be defeated shows my forward thinking. Its sad but to be fair the only reason I was slowly being forced back was purely due to the unfair economic bonuses. I was clearly ahead in technology, economics, and population as well as planets and influence but they kept just pumping out damn ships bigger than the last ones.

I guess they simply clawed it back.

-J
Reply #6 Top
If it wasn't for the fact that I _know_ Stardock is manually updating the AI, I'd swear I have HAL living in my computer.

Because the AI is not scripted (meaning it's closer to a true AI), I wonder if more processing power--like an Athlon64--makes it more lethal by giving it a better capacity for decision making.
Reply #7 Top
Because the AI is not scripted (meaning it's closer to a true AI)


Even AI experts argue endlessly about what a phrase like "true AI" might mean.

Just because it's written in a slightly more complex language than those normally dubbed "scripting" languages doesn't mean it isn't scripted in AI terms. Even Frog openly agrees it isn't really an AI, that's just a convenient term. It's probably a state machine, which means it's responses are predictable.

A non-scripted AI would involve learning/training algorithms of some kind, whether they're backprop/feedforward, or Bayesian statistical probability distributions, or the more specialized Bayes classifier method, or things of that nature. That would be something you could claim was close to true AI, given the current state of the art.

The GC2 AI is closer to a scripted AI than anything else. The only thing more CPU power is going to do is make it reach the same conclusions faster. It isn't spending every spare moment "thinking" about what to do.
Reply #8 Top
This of course brings up a rather nasty issue: the fact that surgeons know more about the human body, which we did not create, than computer programmers know about computers, which we DID create.

Why is it that if I have an internal injury, a surgeon can x-ray me, find the problem, cut me open, fix the problem, and send me out within a week or two, but if I have a windows problem, the answer is "reinstall, update your drivers, reboot, reformat"??

I think that computer programmers should be renamed to "high priests of technology". And they should all be forced to wear a giant pope hat and speak in Latin. Its not like we can understand their technobabble anyway .
Reply #9 Top
How long can/will Stardock continue to better the GC2 AI? Another month? 3 months or until the 300K after release budget dry up?

I'm thinking more of a different version of GC2. If developing 9 Races are too much, why not reduced to 3 or even 1? If game rules are too complex for today technology, why not simplify it? Stardock set the rules.
Reply #10 Top
I think with the AI there's maybe 3 core AI, and the rest are simply derivatives of those three.

Terrans (sure of), Drengin (sure of), and maybe Torian or Thalan?
Reply #11 Top
I want to fight the AI 100% - Economic Bonus to see just how good it is.

Set all AIs to intelligent.