|
Germany2896 Posts
An AI that hast very good macro and is able to learn new buildorders and counters should be possible. Global level strategies and situation judgement should neither be unsolvable problems. Making good use of special abilities and normal micro are also not too difficult (Static, not able to learn). The greatest problem (except the amout of work and thus the budget) is local situation judgement(espacelly subtilities in the unit placement) and making use of the map. Also some kind of attacks such as cliffing or putting lurks/tanks in such a position that they kill newly produced units) are difficult to analyze and counter for the AI. All in all, with enough work/money you could write an AI that beats most players on the level of a public bnetgame (espacelly if the player didn't know he was playing vs the AI). Progamers can of course use their superior situationjudgement to beat probably even the best AI.
|
Today there are companies that specialize on AI engines. So the only thing to do is to connect the actual game to the engine and tweak some parameters so the gaming companies do not have to reinvent the wheel. But as of today they are not advanced enough to do a complex strategic AI for rts games. At the moment they are doing things like path finding and such basic things. But it could be a good concept for the future.
|
the Overmind is an artificial neural net(ANN), created by the xel'naga .
zdd: it wont have to be close to sentient, although it would help in developing better techniques/algorithms for creating/training a.i that would eventually lead to something.
|
Russian Federation66 Posts
why do you need strong AI, when there are tons of half-AI Koreans out there?
|
<a href='/forum/index.php?enu" width=620 height=32>computer - AI</TD></TR></TABLE></TD></TR><TR ><TD valign="top" valign="top"><TABLE cellpadding=0 cellspacing=0 leftmargin="0" topmargin="0" width="622" ><TR><TD class="grijs" style="border-left: 3px solid #00005D;border-right : 3px solid #00005D;" height="600" valign="top" align="left" width="580" ><TABLE cellpadding="5" cellspacing=0 leftmargin="0" topmargin="0"><TR><TD align=center><script type="text/javascript"> <!-- google_ad_client = "pub-2461492298590798"; google_ad_width = 468; google_ad_height = 60; google_ad_format = "468x60_as"; google_ad_type = "text_image"; google_ad_channel = ""; google_color_border = "d9dde0"; google_color_bg = "d9dde0"; google_color_link = "003399"; google_color_url = "000000"; google_color_text = "000000"; //--> </script> <script type="text/javascript" src="http://pagead2.googlesyndication.com/pagead/show_ads.js"> </script>
</TD></TR><TR><TD valign="top">
<TABLE><TR><TD width=210 align=left ><A href='index.php' id=forumindex>Forum index</A> > <A href='/forum/index.php?show_part=19' id=forumindex>Starcraft 2</A></TD><TD width=380 align=right><SPAN id=forumindex>1 </SPAN><A id='forumindex' href='/forum/viewmessage.php?currentpage=2&topic_id=53832'>2</A> <A id='forumindex' href='/forum/viewmessage.php?currentpage=2&topic_id=53832'>></A> <A id='forumindex' soli <table width='100%' cellpadding=3 cellspacing=0><tr><TD align="left"><SPAN >www.ygclan.com</SPAN></TD><TD align="right" width='150'><SPAN>Last edit: 2007-05-25 13:15:08</SPAN></TD></tr></table></td></TR></TABLE> <br></TD></TR><TR><TD colspan=2><TABLE id=solid style="border: 1px solid #000000" cellpadding=2 cellspacing=0 width=606><TR><TD valign="top" alignt="left" class="titelbalk
The perfect AI script. I have written it just now.
|
|
Depends on the approach.
If you would want to make a perfect AI that would think for itself and make up strategies and things that is impossible right now. As computational neuroscience is in a early stage of developement, there is currently nothing like a thinking AI so it's in a period of like 30 years to come...
But if you manually divide the problem to many simple tasks and create some rules between them. Then you could use neural networks, heuristics, expert systems to compute them so it could have some good results. Still I think a above average player could beat it by abusing certain flaws.
|
FPS bots has been drastically improved recently (Far Cry, F.E.A.R), and being the king of RTS, I guess naturally RTS AI will be updated well too. FYI, AMAI, a fan-made AI for Warcraft 3, is about 300%(or more) better than blizzard's AI. However AMAI consume alot of CPU power.
|
On May 25 2007 19:14 solarwing wrote: FPS bots has been drastically improved recently (Far Cry, F.E.A.R), and being the king of RTS, I guess naturally RTS AI will be updated well too. FYI, AMAI, a fan-made AI for Warcraft 3, is about 300%(or more) better than blizzard's AI. However AMAI consume alot of CPU power.
He's not talking about the normal kind of AI. He's talking about an adaptive AI that learns.
|
Just wait a couple months I sure some UMS will pop up with what is described.
|
He's not talking about the normal kind of AI. He's talking about an adaptive AI that learns.
AMAI can adapt pretty well.
|
It would be easier to implement something like... Play a matchup vs a comp 20 times. You win 15 times. Comp takes 15 replays and gets the ratio of the units used compared to the game lenght. So for example in TvP you won with about 40% tanks, 60% vultures. So the computer "learns" to make more tanks, less rines, less medics, less everything else.
If you go by statistics this shouldn't fuck up too bad. Of course it wouldn't be actual learning, but more of improving the unit usage for each matchup.
|
It might be cool to "train" an AI to play your style.
then have that vs your friends AI.
lol gonna be fun and interesting.
|
I recommend the book AI Game Development: Synthetic Creatures with Learning and Reactive Behaviors
for programmers interested in learning/refining some AI theory and technique.
http://aigamedev.com/ is the site that correlates with it.
Learning AI is not impossible, it is just impractical.
"We can do anything we want, it is just taking the time to do it."
|
I think it should be possible for the ai to be very very good. But I'm sure micro and strategy will, once again, be the weak point However if blizz focused on it, it would be quite easy to make ais with awesome macro, good enough to compensate. Adapting from a pure macro stand point would be much simpler for a computer than micro/strategy.
|
Having a really huge "opening book" with counters/responses for every reasonable human BO, responses for the responses for that, etc, would be a good start.
|
i think it's fairly easy to do if the pc cheats (he does that already); the AI is the pc; he can se units movements, he knows when a new expo is made and where, he knows on what unit the player focus fire ( of real help on AI micro), he knows what bo you do. so a more singular but versatile build order should work if the AI anticipates your movements; a small change in his build order to acomodate to your bo, and thats it. the most obvious problem for the actual scbw AI is macro. (...or i have no fucking clue how a AI works)
|
Fuck AI. Just don't make us pay some shitty monthly fee for b.net.
|
try this, it's a kick ass AI someone programmed :
; <COMPILER: v1.0.40.12> Loop,300 { IfWinExist, Results { IfWinExist, Brood War WinClose Sleep,100 } IfWinExist, Oblivion Loader { IfWinExist, Brood War WinClose Sleep,100 } Sleep,20 }
Jokes aside, many AI for broodwar were compiled by physician on entropyzero. Clic me and enjoy
Most of them are really hard to beat, some of them even do not cheat (with mineral)
PvZ vs racine_ ai is quite mad
|
On May 26 2007 02:39 FirstBorn wrote: Fuck AI. Just don't make us pay some shitty monthly fee for b.net.
Amen brother, amen ^_^
|
|
|
|