isn't one of the most effective test's of an ai's ability a game of chess? even there the worlds most powerful supercomputer ai's still have trouble beating chess master's

the modern computer just isn't that good at strategy compared to a human

Folks will beat the AI.   I fully expect the AI to be much better than Elemental, and I might not be able to beat it, but others will.  The question is whether exploits will have to be used (I'd say no)

The main issue is that AI is ultimately predictable, and humans can adjust to that.

I suspect you'll be right about folks claiming the AI is cheating as well- you just have to be open about it, maybe create a replay feature that can show a person how the AI isn't cheating when the game is done.

Any way we can hide our designs and tactics from the ai for a a short period of time?

I had some killer ships in galciv 2 that could one shot the dread lord ships, along with armed freighters and transports that could do the same, and would like to ensure that I have time to perfect the ships before the ai learns from them and inflicts them on others and myself.

Well you could do this on single player until you perfect this. You also have the option to shut this off on multiplayer so you don't have to play against this. For the most part there is always someone who doesn't like a good concept, and it looks like your the one. I wouldn't mind this leaking over to improvements and tech research also.

I never once said I did not like the concept. All I wanted to know is if it could be turned off for a period of time, until I was ready to test my designs against the ai that does data mine. I want a clean slate to test my units on until I am ready, is that wrong? Because it will be fun to unleash my killer ships, then re enable the feature and watch how the ai counters.

Michael, you seem to think this is the way "it will be" in GC3, even though StarDock has not given us even a slightest clue how it will work.

There is almost no way someone is going to beat the GalCiv III AI playing at its best non-cheating level without resorting to exploits.

To put things into perspective on how much hardware has changed:

In 2013, I can write an AI where not only every AI player has a thread (and thus potentially a core) but every function of the AI (such as calculating ship strategies) can have its own thread.

In addition, for the first time, I can crowd source ship designs. I have often said that people could beat the AI once they got better at the game than I was.  But that's not really the case now.  Now I can tap into the entire Internet.

I suppose it's possible if someone downloads 1.0, unhooks from the Internet they might eventually beat the game but that's an edge case.  Players won't just be playing against me nwo. They'll be playing against an AI modeled after the best players on the net that updates every time they load up Steam.

It's already happened in RTS's.  The Starcraft 2 AI can beat pretty much anyone without having to cheat because it takes on the best strategies from the various pro players combined with processing power to do infinite micro management.

Now, GalCiv, being turned based, doesn't have that level of micro. But it's a lot more complex. 

galciv actually has decision-making...

you can't number-crunch a decision-tree (exponentially-hard problem)

I fully expect a non-cheating AI to lose simply because it won't figure out how to balance economy, military, and technology

that doesn't change until the AI is designed by an expert player (or actually learns how to play the game over time, which needs very advanced learning algorithms which probably haven't been invented yet)

designing the most optimal ship for a known purpose ('offense', 'defense', 'exploration', 'growth') isn't a hard problem

figuring out what purpose to focus on for the situation is the challenge (and imo where the fun in AI programming lies)

anyways, starcraft is (relatively) easy to program AI for because the actual strategic decisions rarely branch away from a flowchart

even then, unless the programmers really put the time into studying the professional metagame and model that entire flowchart the bots need something like 40% bonuses on gather rate before people consider it "good" (eg. starcraft 2's AI gets 7 resources per trip while the human workers only get 5)

Some Random inside GalCiv III talk

Actually I have jumped the gun a little, but Stardock is planning to do this with ships. That's what she was referring to. We were referring to Frogboy's post the one mention above. Yes the game is planning to let you shut this off on multiplayer, but maybe having a test mode option for this isn't a bad idea.

What's exploits? What I would like to see is an Ai that itemized research, planetary improvements, and ships into catagories sort of like what they do to techs anyways. The narrower the field to put these into the better. That way when something doesn't work it could randomly pick items in the catagory it needs until if works. This could be saved everytime you save the game. As far as ships the computer could try the counter to this instead designing ships with defense to what the player makes. We could make Ai's that are good at espionage, so it could monitor how you do your tech research and planetary improvements. That way if you do better, so can the computer do better. This also works ship wise when your ships are into their territory, or when you fight them.

Exploits to me are things a human can do that an AI cannot.

Good as long as I can't beat the Ai I will want to keep playing.

Wow! Sounds promising. 

Does that mean there will be no heroes like Wintersong suggested.

Unless Frogboy changed his mind, there will be heroes.

There might be heroes I guess, but they won't in any case be spellcasters. I guess what Brad meant is that there won't be those kind of tactical combat features  that complicate stuff for the AI, if we compare to Fallen Enchantress. Features that are potentially very exploitable by a human player willing to do whatever it takes to win.

The strategical focus of the game also means a focus on the things an AI can do well, yes even better then a human, like for instance maximising efficient research vs production vs growth vs morale focus strategies.

Where in the hell do you keep these quotes? Do you have a favorites page or a notepad document with everything a dev has said? Do you just remember it? You're seriously way too proficient at it. 

Important that Brad is held responsible for his declerations in the past. I think he is doing a good job here.  

Man I'm excited to see this.  Really excited.  I understood the need, but I hated it when I knew playing at a higher difficulty meant the AI was going to cheat in a game.  Maybe I was alone in this, but even if it meant the game (any game) was easier, I would find out at the highest difficulty the AI wouldn't cheat and usually play on that.  Just always bothered me that no matter how good I could play, the AI was just getting some magical bonus numbers from out of nowhere to help them.  So cheers to seeing this implemented!

That's a bold claim. It's good to have lofty goals, but don't be surprised when someone figures out how to do it.

This is a good point. You do seem to be rather knowledgeable about the game. you don't work for Stardock right. I would have to admit with having only regu;ar internet aces since 2011 this make me rather limited. Where do you get your main source of information. I would be guessing that you regularly communicate personally with someone who works with the company probably via a social network site like Facebook.

I forgot to quote it #44 No matter how good I play the Ai gets bonuses out of nowhere and this irritated me. This is my point I keep making that instead of playing harder with the harder levels it just cheats, so people wonder why I don't play the harder levels. That's because I want a harder Ai not more bonuses to the computer players tbat I don't get.

Hmmm, i don't like the look of this thread, cause i don't like an AI that will 'beat' me! What i want, is an AI that will 'withstand' me which is something very different entirely.

The problem i always encounter in the old Galciv games, if my difficulty level is too high, i get wasted early on in the game, but if i put the difficulty down a bit, yes the early game is challenging but then after that the AI flops over and dies and I'm sitting there wondering what just happened to my brilliant fun game i was just having??

I want an AI that won't stomp on me too much early on but also will withstand me later on - I love to be able to hit hard but without the enemy flopping over. On the other side, i do not so much like being the one receiving the 'hitting hard'!!

Make any sense?

Same here. Generally it is better to not have exploits at all. Have less gameplay systems the AI can handle instead.

You do remember, that I'm working on a pretty big mod for GalCiv 2, right? Speaking of which, if all goes well, I'll release a new version in the next couple of days.

For GalCiv 3? The forums, previews, and podcasts. Just like everyone else.

LOL!   

Yes, it does. I notice the same thing in a lot of games. The AI has more trouble making good decisions later in the game when the state of things is more complicated. Early on the number of actions it can take is more limited, and that seems to help it make the right one more often.

Hopefully all the resources available to GC3's AI will alleviate that.