My Rift solo build was a ranger marksman assassin.
I sent my pet in with a click, pressed 1 1 1 1 1 2 and CTRL+1 when my pet hp went low, tabbed to the next mob I wanted to kill and pressed 1 1 1 1 1 2 and CTRL+1 if needed.
I could have pressed more buttons but I didn't really need to, so I just skipped through with my 1s and 2s, and cried boredom but told my self, hey the end game will be fun im sure of it!
Do you know of any existing games, present or past, that have a non-cheating AI that can beat a skilled human? I ask this because if AI were really easy to build, like you are theoretically arguing, then you would think there would be at least one game out there that has a good non-cheating AI. The only games I can think of that can beat skilled players without cheating are chess and a few card games.
I personally don't know of a single complex ruled game that has an AI that can beat a skilled player without cheats. Do you?
Kidding, right? Any first-person shooter fits those criteria. It is not difficult to have a bot fire with perfect precision at the first instant that a single hittable pixel of your avatar is visible. And unless there is a maximum turning speed, it can look in forward and backward (or up) on alternate frames so that you can't approach it from any angle.
It shouldn't be too hard to figure out why developers don't bother to put that sort of AI in their games.
FPS remains a popular genre but nearly all modern FPSes pit the player against AI opponents who are much weaker in terms of firepower and health/armor. They go down in one headshot but never (or almost never) kill you with a single shot. Quake III Arena and Unreal Tournament bots just don't compare; they can't provide the same experience. They're good examples of what can happen when you put "Good" AI in the wrong kind of game. Because they have health and weaponry equal to your own, it's plainly obvious that you are only capable of beating them because they deliberately perform poorly.
No I'm not kidding. The AI in every FPS I've played, and I've pretty much played them all, is terrible compared to human AI.
I still haven't heard anybody list a single existing game (again, other than the very simple rule games in simple environments like chess or card games) that has a non-cheating AI that can beat a skilled human player.
List the game and the difficulty level please.
The reason it doesn't exist is AI in a complex ruled game with complex environments is extraordinarily difficult to program, and any effective algorithm if you could come up with one would take extraordinary amounts of computation power that would make it innefective in real time. In fact, you can mathematically prove that most brute force algorithms that involve the number of variables in a typical complex ruled game are computationally infeasible with any Turing based machine in our universe. Now if quantum computers come to reality the question is open for debate again as they break the rules of Turing machines.
And developers don't really even try that hard, so it is moot even talking theory. Just look how crappy a simple concept like pathfinding is in even modern games. Heck look how long it took before AI could compete with the great chess players, and chess rules are very simple and the grid is only 8x8.
GW2 "built from the ground up with microtransactions in mind" 1) Cash->Gems->Gold->Influence->WvWvWBoosts = PAY2WIN 2) Mystic Chests = Crass in-game cash shop advertisements
No I'm not kidding. The AI in every FPS I've played, and I've pretty much played them all, is terrible compared to human AI.
Ever played an early Rainbow Six game, specifically the 3rd one: Raven Shield? That game is impossible, even today(re-installed recently & regret it haha), I cannot peak around a corner in that game without having my head blown away and have no idea where the shot came from lol.
No I'm not kidding. The AI in every FPS I've played, and I've pretty much played them all, is terrible compared to human AI.
Ever played an early Rainbow Six game, specifically the 3rd one: Raven Shield? That game is impossible, even today(re-installed recently & regret it haha), I cannot peak around a corner in that game without having my head blown away and have no idea where the shot came from lol.
I swear those AI use wall-hacks!
There are definately FPS games where beating a mode at a certain level is difficult, but it is usually because you are one versus many and your damage tolerance is lowered. I have never had a problem beating a FPS bot in a one-on-one scenario for instance in a game that has bots play by the same rules as the humans. Bot AI was a big topic of research back in the Quake days. There were even some college theses out there on it believe it or not.
Now of course you can come up with a scenario like drop yourself and a bot in a straight hallway and he can beat you, but make the map complex at all and the weaponry varied then the bots get smoked every time...
GW2 "built from the ground up with microtransactions in mind" 1) Cash->Gems->Gold->Influence->WvWvWBoosts = PAY2WIN 2) Mystic Chests = Crass in-game cash shop advertisements
Do you know of any existing games, present or past, that have a non-cheating AI that can beat a skilled human? I ask this because if AI were really easy to build, like you are theoretically arguing, then you would think there would be at least one game out there that has a good non-cheating AI. The only games I can think of that can beat skilled players without cheating are chess and a few card games.
I personally don't know of a single complex ruled game that has an AI that can beat a skilled player without cheats. Do you?
Yeah... Pretty sure what you are experiencing is largely an artifact of the scriptors purposefully not putting in unbeatable gameplay modes.
One specific example that comes to mind for me is Soulcalibur (Dreamcast). As anyone who's played the game extensively can tell you there is no "game mode" which is "difficult" as no difficulty setting prevents the AI from falling into easy tech traps which through correct punishment trivializes even the supposedly "difficult" games such as "Survival Mode".
However, if you go into practice mode with a CPU set on the highest difficulty setting you may notice a rather interesting property... Pretend the game is essentially "survival mode" where the only way to win is to knock the opponent out of the ring. The first ~10 times you knock the cpu out unanswered will be relatively trivial. The next 10 times will take you nearly three times as long. By ~50 the CPU becomes a DEMON. It will flawlessly execute guard impacts/parries, cancel 100% of all throws and punish absolutely anything that can be punished. From that point I believe the game has essentially capped out (I've made it to ~75 and it didn't feel vastly different) but the CPU is SO good at that point that it is basically inevitable that you will get knocked out at some point (resetting you back to 1 and the CPU back to "easy mode") so it's entirely possible it just keeps getting more and more "perfect" from there.
TL;DR
Clearly it was possible for them to make a ridiculously good AI for a fighting game that is pretty widely considered to have one of the largest skill barriers found in a 3d fighter (and it's not like fighters are a genre known for low skill barriers to begin with...) even given the constraints of the Dreamcast's pretty feeble processing power.
I just wish that AI mode hadn't been confined purely to practice mode gated behind a ~20 minute time sink knocking out chump AIs. Silly developers and their insistance that their games be "winnable". ;p
Do you know of any existing games, present or past, that have a non-cheating AI that can beat a skilled human? I ask this because if AI were really easy to build, like you are theoretically arguing, then you would think there would be at least one game out there that has a good non-cheating AI. The only games I can think of that can beat skilled players without cheating are chess and a few card games.
I personally don't know of a single complex ruled game that has an AI that can beat a skilled player without cheats. Do you?
Yeah... Pretty sure what you are experiencing is largely an artifact of the scriptors purposefully not putting in unbeatable gameplay modes.
One specific example that comes to mind for me is Soulcalibur (Dreamcast). As anyone who's played the game extensively can tell you there is no "game mode" which is "difficult" as no difficulty setting prevents the AI from falling into easy tech traps which through correct punishment trivializes even the supposedly "difficult" games such as "Survival Mode".
However, if you go into practice mode with a CPU set on the highest difficulty setting you may notice a rather interesting property... Pretend the game is essentially "survival mode" where the only way to win is to knock the opponent out of the ring. The first ~10 times you knock the cpu out unanswered will be relatively trivial. The next 10 times will take you nearly three times as long. By ~50 the CPU becomes a DEMON. It will flawlessly execute guard impacts/parries, cancel 100% of all throws and punish absolutely anything that can be punished. From that point I believe the game has essentially capped out (I've made it to ~75 and it didn't feel vastly different) but the CPU is SO good at that point that it is basically inevitable that you will get knocked out at some point (resetting you back to 1 and the CPU back to "easy mode") so it's entirely possible it just keeps getting more and more "perfect" from there.
TL;DR
Clearly it was possible for them to make a ridiculously good AI for a fighting game that is pretty widely considered to have one of the largest skill barriers found in a 3d fighter (and it's not like fighters are a genre known for low skill barriers to begin with...) even given the constraints of the Dreamcast's pretty feeble processing power.
I just wish that AI mode hadn't been confined purely to practice mode gated behind a ~20 minute time sink knocking out chump AIs. Silly developers and their insistance that their games be "winnable". ;p
Now I do admit I am not big into the Japanese style fighting games, so I am not a good test subject for the AI there.
And to be honest that may be an example of a simple ruled game in a simple environment. As I recall most of those games are 2 dimensional on one screen with basically no obstacles. And your choice of moves is relatively small as well. A pure twitch game where the AI is probably limited to simple rules like "If move A counter move B, etc"
But that is the reason I asked the question, as it would be interesting to list all the games that have AIs capable of beating humans. I just know that in the genres I play I haven't seen it yet: FPS games, strategy games, and RPG games mostly (in addition to MMOS where I prefer the PVP aspect so AI isn't usually an issue for me).
I was probably to hasty to generalize AI as only good in chess and card games, as there are probably a number of games that break down into very simple ruled games in simple environments. Japanese fighting games probably an example of that.
GW2 "built from the ground up with microtransactions in mind" 1) Cash->Gems->Gold->Influence->WvWvWBoosts = PAY2WIN 2) Mystic Chests = Crass in-game cash shop advertisements
Do you know of any existing games, present or past, that have a non-cheating AI that can beat a skilled human? I ask this because if AI were really easy to build, like you are theoretically arguing, then you would think there would be at least one game out there that has a good non-cheating AI. The only games I can think of that can beat skilled players without cheating are chess and a few card games.
I personally don't know of a single complex ruled game that has an AI that can beat a skilled player without cheats. Do you?
Kidding, right? Any first-person shooter fits those criteria. It is not difficult to have a bot fire with perfect precision at the first instant that a single hittable pixel of your avatar is visible. And unless there is a maximum turning speed, it can look in forward and backward (or up) on alternate frames so that you can't approach it from any angle.
It shouldn't be too hard to figure out why developers don't bother to put that sort of AI in their games.
FPS remains a popular genre but nearly all modern FPSes pit the player against AI opponents who are much weaker in terms of firepower and health/armor. They go down in one headshot but never (or almost never) kill you with a single shot. Quake III Arena and Unreal Tournament bots just don't compare; they can't provide the same experience. They're good examples of what can happen when you put "Good" AI in the wrong kind of game. Because they have health and weaponry equal to your own, it's plainly obvious that you are only capable of beating them because they deliberately perform poorly.
No I'm not kidding. The AI in every FPS I've played, and I've pretty much played them all, is terrible compared to human AI. -You mean "intelligence", right?
I still haven't heard anybody list a single existing game (again, other than the very simple rule games in simple environments like chess or card games) that has a non-cheating AI that can beat a skilled human player.
List the game and the difficulty level please.
The reason it doesn't exist is AI in a complex ruled game with complex environments is extraordinarily difficult to program, and any effective algorithm if you could come up with one would take extraordinary amounts of computation power that would make it innefective in real time. In fact, you can mathematically prove that most brute force algorithms that involve the number of variables in a typical complex ruled game are computationally infeasible with any Turing based machine in our universe. Now if quantum computers come to reality the question is open for debate again as they break the rules of Turing machines.
And developers don't really even try that hard, so it is moot even talking theory. Just look how crappy a simple concept like pathfinding is in even modern games. Heck look how long it took before AI could compete with the great chess players, and chess rules are very simple and the grid is only 8x8.
I agree with gainevilleg. AI in FPS games is not smart. It is unnaturally quick and accurate, and cheats, but it is not smart. Its function is strictly limited to moving around the map, maybe along some waypoints, pointing any game characters and pressing "fire". Differences in AI in FPS games come from their ability to adjust to the player, other game characters near them and the surrounding environments.
Chess has a limited ruleset which makes it easy for an AI to play it. It is a whole different thing to program an AI that acts like a human player without cheating in for example an FPS game.
I skate to where the puck is going to be, not where it has been-Wayne Gretzky
Have you ever made an actual AI? Designed one? Do you even know what's involved? Do you know what a decision tree is? Do you know what exponential growth is? Do you know what neural networks are? Do you know what computational complexity is?
No offense but you don't know what you're talking about. You're talking as if AI is possible that can act like human players and be as smart or even smarter then they are.
An aimbot is not an AI. An aimbot either scans for pixels on each already drawn with colors that are mostly unique to players and aims for them, it modifies your directX libraries so it can grab the 3D coordinates of anything being drawn with a player model and skin or they directly modify the game file to allow for extra functionality. Either by "patching" the game file on your hard drive or by altering your CPU's instruction cache real-time to modify the program flow.
There's not a single decision tree, neural network or pattern recognition in an aimbot.
The sims is not an example of amazing AI. A sim is based on a few very simple values. It's got it's hunger, sleep, entertainment etc. values and it's got it's relationship values. If one of it's hunger, sleep etc. values gets too low then it scans the environment for something that has a positive effect on that value and it uses a simple A* pathing alghorithm to get there. If it encounters another sim it possibly starts a conversation based on his relationship value. I'm sure the entire story is a bit more complex then that but it's not some sort of revolutionary AI.
Computers are incapable of such AI as of yet. AI so far is only capable of doing a very limited task in a strictly controlled environment.
Yes, you can make a pet that doesn't attack crowd controlled targets. Now try making a pet that does that along with a thousand other things and is capable of deciding within 5ms of which one of those thousand things it has to do. It can't get access to all the statistics because that would take too long. It can't predict the outcome of every possible action because that would take too long. It has to deal with unknown variables because there's no time to wait until you know what they are. It has to know which variables are important and which aren't even before it knows what the values of them are. It has to be able to adapt and recognise when it's chosen action didn't have the desired effect. But it can't do a full check wether or not the desired effect was achieved because there's no time. It has to change the conditions of an action constantly but never too much. It has to be able to execute actions even when not all conditions are met. It has to be able to decide between 2 actions that both have their conditions fulfilled. It can't calculate too much though, there's no time.
And no, even with a quadcore of those nice and shiny 5,2 GHz CPUs there's still not enough time.
Ok then the conclusion is solved , you can´t do it , Neither can I cause it´s not my specialty or field .
But I do know people who are able to easily do it , but nobody plays there games , cause just like your post , who is nothing but admiting defeat and trying to push off responsibility . the same EGO or ERGO problems presist , that there are people who can do it , just you and me are taken out of the equation .
Final , more you rant about how difficult it is , the more nonesense it sounds , offcourse it is a LOT OF WORK .
Like any programming , debugging , modding , cause you don´t want or cannot do it , doesn´t mean it is not possible.
So stop making tons of excuses , like boring programmers these days , who keep saying impossible .
Have you ever made an actual AI? Designed one? Do you even know what's involved? Do you know what a decision tree is? Do you know what exponential growth is? Do you know what neural networks are? Do you know what computational complexity is?
No offense but you don't know what you're talking about. You're talking as if AI is possible that can act like human players and be as smart or even smarter then they are.
In order that would be: "Yes", "Yes", "Yes", "Yes", "Yes", "Yes", "Yes", "Probably true", "It can't".
I didn't make this post to trivialize the difficulty of making intelligent AI because that is truly difficult. I made it simply to point out that I think you are giving WAY too much credit to the amount of "intelligence" required to be good in limited environments such as games.
Three years ago the general consensus of people in the AI field was that "Go" (a popular board game in asia) would NEVER be playable at any skill level by a computer because its complexity was divergent. Thanks to Monte Carlo search even at 19x19 it can now beat all but a few professional players. In another year or two humans will no longer be able to beat computers at Go.
Many FPS games (especially those allowing headshots) are nothing more than an exercise in reaction speed and quickness of aim (something aimbots have been exploiting for years).
Many fighters have strictly defined enough rules that a computer can be designed to play them perfectly.
Platformers are so easy to beat with AI that designing solutions for them is frequently assigned as undergrad work. (A real good friend of mine "solved" "Mario: Lost Levels" for his undergrad final project actually, so I can cite specific examples of this one...)
TL;DR
There almost certainly ARE common game archetypes which are difficult for current AI (and of course if that ever ceases to be true it should still be trivially easy to purposefuly design a game type which it is "bad at" at least for some time to come) but I would argue that there are many (if not most) archetypes which dominate the market and yet are simply "too easy" for AI.
EDIT: And yes... Ken Jennings already got wtf pwned by a computer on Jeopardy. While exceptionally cool that game is probably not of the type that the OP is talking about.
You can easily make an AI that will stomp all over human players. It doesn't require the AI to be as smart as a human, just smart enough because tthe AI has huge advantages.
1st it makes its decisions near instantaneously
2nd it makes its response selection near instantaneously
3rd it never accidentally hits the wrong button
4th if it's designed to for example not hit someone CCd it never will, but humans definitely will...a lot.
etc...
1 and 2 is a lot bigger than most people think. A very good gamer will be able to make his mind for an action and actually physically hit the button they want, in say 0.2 seconds. This is generous. The computer takes 0.00001 seconds to do this. In a fight with 15 actions that equates to 3 seconds the computer gains on a human opponent. More than enough to win nearly any fight. Yes, of course you can dumb down the AI to slow its response time, but that's not the argument here.
3 and 4 are big for other reasons.
Anyway, the computer AI doesn't need to think overly complex thoughts in order to kick the snot out of a human player at least most of the time. This isn't what people want though. They want to feel godly compared to all the peon monsters out there. Ooooohhhh, I can destroy 8 mobs at once, and kill 10000s of mobs without ever dying. They wouldn't like it too much if the mobs killed them 8 out of 10 times they tried to fight one, and don't even think about fighting 2 mobs at once.
With a global cooldown on almost all abilities in mainstream mmos point number 2 is irrelevant. Regarding number 3 and 4, they only apply if you're not playing with a good group of dedicated players. Compare running with friends to running with randoms. We never had accidental cc breaks, and we maybe hit the wrong button once every hour or so... Doesn't really change the outcome of a fight..
Now I do admit I am not big into the Japanese style fighting games, so I am not a good test subject for the AI there.
And to be honest that may be an example of a simple ruled game in a simple environment. As I recall most of those games are 2 dimensional on one screen with basically no obstacles. And your choice of moves is relatively small as well. A pure twitch game where the AI is probably limited to simple rules like "If move A counter move B, etc"
I'm starting to suspect that your definition for "simple ruled game in a simple environment" is no different than "games that an AI can master". You've formed a tautology. You're asking us to give an example of an AI that can beat skilled players in a non-simple game, but for any game that someone holds up as an example of that, you can just respond "well if the AI can beat people then that must be a simple game".
The whole point of the thread was to show that the tank and spank combat is half of the reason why we have dull combat, many forget that. But I do see that AI does bog down server resources. I do have an idea for a tier up on combat for both melee and spell casting.
I am talking about AI making decisions within combat more so than, "I am dying, I need to flee or yell for my comrades." The way I figured how a more intelligent AI would make decisions and to adapt to a player is essentially by collision detectors. There would be a very complex "if than statements" and the AI would calculate the best decision to adapt, they would also detect patterns within the players strategy to adapt. So in theory not as complex but the coding could be complex.
Some of you say that, because the WOW crowd whined, and moaned about a challenging combat with Catacysm, Blizzard nerfed it. Gameplay like what I am describing is not for the WoW player who wants their hand held. This gameplay is for players to use their brain once and a while and create strategy when needed, a challenge, a challenge in which this genre is lacking. I guess you'll have to see my combat mechanic design to full understand.
You can easily make an AI that will stomp all over human players. It doesn't require the AI to be as smart as a human, just smart enough because tthe AI has huge advantages.
1st it makes its decisions near instantaneously
2nd it makes its response selection near instantaneously
3rd it never accidentally hits the wrong button
4th if it's designed to for example not hit someone CCd it never will, but humans definitely will...a lot.
etc...
1 and 2 is a lot bigger than most people think. A very good gamer will be able to make his mind for an action and actually physically hit the button they want, in say 0.2 seconds. This is generous. The computer takes 0.00001 seconds to do this. In a fight with 15 actions that equates to 3 seconds the computer gains on a human opponent. More than enough to win nearly any fight. Yes, of course you can dumb down the AI to slow its response time, but that's not the argument here.
3 and 4 are big for other reasons.
Anyway, the computer AI doesn't need to think overly complex thoughts in order to kick the snot out of a human player at least most of the time. This isn't what people want though. They want to feel godly compared to all the peon monsters out there. Ooooohhhh, I can destroy 8 mobs at once, and kill 10000s of mobs without ever dying. They wouldn't like it too much if the mobs killed them 8 out of 10 times they tried to fight one, and don't even think about fighting 2 mobs at once.
With a global cooldown on almost all abilities in mainstream mmos point number 2 is irrelevant. Regarding number 3 and 4, they only apply if you're not playing with a good group of dedicated players. Compare running with friends to running with randoms. We never had accidental cc breaks, and we maybe hit the wrong button once every hour or so... Doesn't really change the outcome of a fight..
I don't think you understand. The computer AI knows exactly the moment a cooldown ends, and responds with the next spell or ability instantly. For a human it requires you to see the cooldown end, process the information, select next spell/ability(you may have already decided upon this), and then physically press the button to use your decided upon skill/ability. This takes you anywhere from 0.2-0.5 seconds to do. It doesn't matter if it's global or not, you still have to do this. Just try pressing a stopwatch button to start it and stop it as fast as you can. It takes about 0.2 seconds for you to do because your body is only physically so fast, and that's without having to make any decision or take in any external input, like visual or sound cues. This is the same reason the IBM Jeopardy AI always beat the humans to the buzzer if it had found an acceptable anwer. IT IS FASTER. With 15 abilities in a fight this really adds up, even if you are extremely fast, the AI will just beat you to it.
For example say in WOW you enter an instance where you and your party have a little cutscene and then a fight begins against an AI controlled party. The AI controlled party will know exactly the moment the cutscene ends and the fight begins, and begin casting their CC spells immediately after the cutscene. Even the best of the best human would take 0.2-0.4 seconds to see the visual cue the cutscene has ended and press the button to start their CC. Guess who gets their CCs off first...EVERYTIME. Now with your party CCd the AI would then just tag team your healer, kill him and move to the next clothie. By the time your party is even in shape to fight the battle is essentially over. Not because there was some super complex AI in place. All it has to do is react instantly, CC everyone it can, and focus fire on healer, and then as long as it doesn't act completely stupid the rest of the fight, you lose.
Kidding, right? Any first-person shooter fits those criteria. It is not difficult to have a bot fire with perfect precision at the first instant that a single hittable pixel of your avatar is visible. And unless there is a maximum turning speed, it can look in forward and backward (or up) on alternate frames so that you can't approach it from any angle.
It shouldn't be too hard to figure out why developers don't bother to put that sort of AI in their games.
FPS remains a popular genre but nearly all modern FPSes pit the player against AI opponents who are much weaker in terms of firepower and health/armor. They go down in one headshot but never (or almost never) kill you with a single shot. Quake III Arena and Unreal Tournament bots just don't compare; they can't provide the same experience. They're good examples of what can happen when you put "Good" AI in the wrong kind of game. Because they have health and weaponry equal to your own, it's plainly obvious that you are only capable of beating them because they deliberately perform poorly.
As I've said before. Aimbots aren't AI.
Aimbots solve the following problem:
If a player is at position (x,y) then where do I shoot? Answer: position (x,y)
They cheat. They have access to information players don't. If players could macro their "shoot" key to automatically headshot then who do you think would win? Those bots or human players?
AI with equal stats and information playing to win can not beat human players with equal stats and information playing to win.
AI with superior stats and superior information can easily win, unless they're faced with human players that also have superior stats and superior information ( AKA open console. AI.setHealth(0) ).
Aimbots aren't good AI. They're not even AI to begin with.
We are the bunny. Resistance is futile. ''/\/\'''''/\/\''''''/\/\ ( o.o) ( o.o) ( o.o) (")("),,(")("),(")(")
Some of the hardest (and most interesting) fights in WoW were the PvP like fights. You got a random mix of opponents, and those opponents would fight you much like actual players would. There was one in Magister's Terrace and one in another instance that escapes me. Those were the encounters that I actually remember and I enjoyed them a lot. Of course, I enjoy PvP a bit more than PvE too, so it makes sense that I would enjoy those encounters.
Better, more human like AI can and has been done. You can't script the encounters though. The Scripted Boss Encounter is a staple of raiding...people like it or they expect it. It allows guilds to spend time learning encounters and getting it right. People who don't know the encounters need to spend time learning them. I think that's the main thing...with the human like AI, you could walk in and one shot encounters right off the bat. With scripted encounters, there's always a learning curve, time spent getting the encounter right.
I think the better AI for specific mobs makes sense and if you designed your game with that as a basic mechanic instead of the scripted encounter as a basic mechanic. In games where you learn the scripts for fights and spend a bunch of time gearing up for the fights, it doesn't make nearly as much sense. Perhaps if you had a mix of the two...some scripted mechanics you have to learn, and some better AI for the people who can just react.
I can not remember winning or losing a single debate on the internet.
Kidding, right? Any first-person shooter fits those criteria. It is not difficult to have a bot fire with perfect precision at the first instant that a single hittable pixel of your avatar is visible. And unless there is a maximum turning speed, it can look in forward and backward (or up) on alternate frames so that you can't approach it from any angle.
It shouldn't be too hard to figure out why developers don't bother to put that sort of AI in their games.
FPS remains a popular genre but nearly all modern FPSes pit the player against AI opponents who are much weaker in terms of firepower and health/armor. They go down in one headshot but never (or almost never) kill you with a single shot. Quake III Arena and Unreal Tournament bots just don't compare; they can't provide the same experience. They're good examples of what can happen when you put "Good" AI in the wrong kind of game. Because they have health and weaponry equal to your own, it's plainly obvious that you are only capable of beating them because they deliberately perform poorly.
As I've said before. Aimbots aren't AI.
Aimbots solve the following problem:
If a player is at position (x,y) then where do I shoot? Answer: position (x,y)
They cheat. They have access to information players don't. If players could macro their "shoot" key to automatically headshot then who do you think would win? Those bots or human players?
AI with equal stats and information playing to win can not beat human players with equal stats and information playing to win.
AI with superior stats and superior information can easily win, unless they're faced with human players that also have superior stats and superior information ( AKA open console. AI.setHealth(0) ).
Aimbots aren't good AI. They're not even AI to begin with.
You could easily make an AI that doesn't cheat by just shooting player at (X,Y) and instead "sees" the player and then shoots, and have it never miss. That would be trivial, and it needs no more information than would be available to you.
In order that would be: "Yes", "Yes", "Yes", "Yes", "Yes", "Yes", "Yes", "Probably true", "It can't".
I didn't make this post to trivialize the difficulty of making intelligent AI because that is truly difficult. I made it simply to point out that I think you are giving WAY too much credit to the amount of "intelligence" required to be good in limited environments such as games.
Three years ago the general consensus of people in the AI field was that "Go" (a popular board game in asia) would NEVER be playable at any skill level by a computer because its complexity was divergent. Thanks to Monte Carlo search even at 19x19 it can now beat all but a few professional players. In another year or two humans will no longer be able to beat computers at Go.
Many FPS games (especially those allowing headshots) are nothing more than an exercise in reaction speed and quickness of aim (something aimbots have been exploiting for years).
Many fighters have strictly defined enough rules that a computer can be designed to play them perfectly.
Platformers are so easy to beat with AI that designing solutions for them is frequently assigned as undergrad work. (A real good friend of mine "solved" "Mario: Lost Levels" for his undergrad final project actually, so I can cite specific examples of this one...)
TL;DR
There almost certainly ARE common game archetypes which are difficult for current AI (and of course if that ever ceases to be true it should still be trivially easy to purposefuly design a game type which it is "bad at" at least for some time to come) but I would argue that there are many (if not most) archetypes which dominate the market and yet are simply "too easy" for AI.
EDIT: And yes... Ken Jennings already got wtf pwned by a computer on Jeopardy. While exceptionally cool that game is probably not of the type that the OP is talking about.
We're not talking about platformers. We're not talking about shooters. We're not talking about board or card games.
We're talking about a MMO server executing an AI that's on par with players PvPing. Not only does an MMO have so many more variables then shooters, platformers and board games have but it also has to be executed for thousands of mobs.
And I don't doubt that in 10 years we may see MMOs that can do those things. Just like we now have AIs that were impossible 10 years ago.
But current day it's impossible to make an AI that can beat human players in an MMO without greatly inflating their stats. I'm even leaving the superior information out. You can have that.
But even so an MMO has simply way too many variables for it to work. You're not dealing with 10 possible weapons and 10 possible moves. You're dealing with 100s of possible abilities and moves.
And you have to fit all that into something that can be executed thousands of times per second.
We are the bunny. Resistance is futile. ''/\/\'''''/\/\''''''/\/\ ( o.o) ( o.o) ( o.o) (")("),,(")("),(")(")
I would love to play a mmo where you have diverse dynamic gameplay and combat from mobs dynamic and unpredictable behavior and attacks. And they could make mobs in all kinds of different lvls of difficulty so you have varity of fights easy to strong mobs .
But with current pc's people have at home majority still have a pc from 5-6 years ago and run XP 32bit and laptops who realy are terible for gaming, lol plus avarage connection of 2mbit i realy doub we see soon more complex games. My prediction is maybe in 5 years time we slowly get more complex games. But on the other hand game developers and publishers are also looking to reach more people to sell games and with pc losing more and more prolly games even getting dumb down and we see more complex games maybe in 10 years time?
Evasia, I am envisioning something very similiar to what you see here. I do agree that if AI becomes a major deterrent to factor in new combat and a more indepth gameplay experience, it wont be something coming out in the next 5 years. So I think you've hit the nail on the head. I have design combat mechanics that would compliment such an endeavor, but its all in high concept document form. Need to bring it to life to see if theory works.
My Rift solo build was a ranger marksman assassin.
I sent my pet in with a click, pressed 1 1 1 1 1 2 and CTRL+1 when my pet hp went low, tabbed to the next mob I wanted to kill and pressed 1 1 1 1 1 2 and CTRL+1 if needed.
I could have pressed more buttons but I didn't really need to, so I just skipped through with my 1s and 2s, and cried boredom but told my self, hey the end game will be fun im sure of it!
This is exactly the problem and something I am trying to fix. Mindless and no purpose driven combat. I'd personally rather become a tactician and strategize against the opposition then consistently, repeat, snare + frost blast + fireball + stun + fireball + snare + lightning bolt all the time to kill a mob. That could work sometimes but I'd like to be challenged. If people would just pay attention to the purpose of the thread instead of taking a different street, they could see what I am talking about, adaptbility.
You could easily make an AI that doesn't cheat by just shooting player at (X,Y) and instead "sees" the player and then shoots, and have it never miss. That would be trivial, and it needs no more information than would be available to you.
Thus explaining the fact that we currently have AIs capable of instant object recognition in graphical pictures in nearly all circumstances....
I wonder why google is still looking at the pages around images and other text instead of just having this AI scan the images and instantly tell you what it is.
We are the bunny. Resistance is futile. ''/\/\'''''/\/\''''''/\/\ ( o.o) ( o.o) ( o.o) (")("),,(")("),(")(")
You could easily make an AI that doesn't cheat by just shooting player at (X,Y) and instead "sees" the player and then shoots, and have it never miss. That would be trivial, and it needs no more information than would be available to you.
Thus explaining the fact that we currently have AIs capable of instant object recognition in graphical pictures in nearly all circumstances....
I wonder why google is still looking at the pages around images and other text instead of just having this AI scan the images and instantly tell you what it is.
Way to go. You just related a game with a very very very very limited number of objects, and even fewer objects that actually move with the real world which has billions upon billions of objects and said the recognitiion needed for the AI would be the same in both.
Is it possible to talk about AI behavior within combat making decisions and reacting within combat? Is it possible to get this thread back on it's rails?
Is it possible to talk about AI behavior within combat making decisions and reacting within combat? Is it possible to get this thread back on it's rails?
We are talking about it. Some of us believe they could make AI that would be pretty good at this already, but they don't because people would just whine and complain about it so instead they make "fun" AI instead of "good" AI. Others think that a "good" AI is currently unnattainable which is why developers opt for the "fun" AI. This argument is central to your point.
You could easily make an AI that doesn't cheat by just shooting player at (X,Y) and instead "sees" the player and then shoots, and have it never miss. That would be trivial, and it needs no more information than would be available to you.
Thus explaining the fact that we currently have AIs capable of instant object recognition in graphical pictures in nearly all circumstances....
I wonder why google is still looking at the pages around images and other text instead of just having this AI scan the images and instantly tell you what it is.
Way to go. You just related a game with a very very very very limited number of objects, and even fewer objects that actually move with the real world which has billions upon billions of objects and said the recognitiion needed for the AI would be the same in both.
fewer object moving?, wow your ignorance is amusing, even if you are not moving or a object is not moving that don't mean the computer stop calculating that object, also the simple fact you move make he calc new object who enter in your view and recalc the old object who is your view, just because you moved.
you simple are lost on your own wish, first you wanted a AI who would make the game fun, now you just want to have a strong NPC, like you said you can make a bot who land 100% of thet time and react the moment you enter his LOS, then what? you will die, no change of reaction, is that what you want? for that you don't need a true AI.
again I will say go study about AIs before commenting anything new, you don't even KNOW what you want.
I fail to see what 2mb lines have to do with ai. o.o
This whole thing depends on the type of combat. If its WoW, there are those things like murlocs running for help, enemies with counterspell, etc.
If youre talking about AI on a solo gameplay basis then theres not much you can do about it but maybe add a small algorithm for ranged enemies to run around a bit. This running around part would take a toll on the servers if there are a lot of ranged enemies. For example, trigerring the murloc running away thing when ranged enemies get hit with melee (would make it too annoying for melee chars)
The main problem for this becomes that...The AI has to be really smart if youre thinking in terms of soloing. In group play, its a lot easier to make it feel more dynamic by tweaking the aggro system so that the guy who does the least damage but has high defense gets ignored a lil more. Again someone stated this already but people yelled nerf for it in cataclysm.
If you were to just try to improve the enemies ai in terms of groups then the game would have to be built where the combat lasts maybe 15-20 seconds and where everyone will put on their toes. This would mean that tanks will get a constant aoe aggro that affects maybe 2-3 enemies and works mostly on melee focused characters and that ranged enemies or certain melee enemies having a higher tendency (maybe resistance or immunity) to the taunt and will most likely go for the most dangerous one in your group, etc.
Having the combat react that way is actually very plausible. You could take WoWs current AI, do a couple of tweaks on the numbers, add a few taunt immunities and maybe add a little code to make it so that threat is always going if a person doesnt do as much and then youll have quite a dynamic group based ai. If the players are using the trinity system, either let the AI use it as well (wouldnt you like to be forced to attack the toughest guy while their hard damage dealers go for your priest?) or at the very least make combat not as reliant on the constant where the enemies will always without a doubt be on your tank. That way, everyone will be on their toes as to who to kill, how to protect the slightly squishier people, etc.
AI for this kind of combat will be as simple as WoW and would bring out quite a dynamic combat system. The problem with it is designing epic content with it. How would you like it if you were fighting a big dragon and it always seems to smack your tank once then proceed to kill everything else?
You could argue that "Oh that doesnt work for solo and itll still be easy mode" But really theres not much to do about soloing. Causing enemies to circle strafe you or run away annoys the !@#$ out of people because of the adds. Giving enemies access to a lot of skills would make them too hard for a lot of people and it really would NOT be solo friendly. Letting the solo enemies run away and lure players to traps will not be feasible in an mmo because of adds and the fact that people while cry nerf. So the only way to have the dynamic system if its part of the core system and a lot of the content will be designed that way (just like a lot of the systems you see now is basic tank n spank + 1 or 2 gimmicks)
Also with this kind of system means that classes will have to work differently and roles will be very broken off from the regular trinity game. Which would mean that the combat system will be more innovative and we all know how innovation gets handled.
So no...Linear AI is not the reason for me. Its the Tank and spank system that makes it extremely dull for me.
''/\/\'' Posted using Iphone bunni ( o.o) (")(") **This bunny was cloned from bunnies belonging to Gobla and is part of the Quizzical Fanclub and the The Marvelously Meowhead Fan Club**
Comments
My Rift solo build was a ranger marksman assassin.
I sent my pet in with a click, pressed 1 1 1 1 1 2 and CTRL+1 when my pet hp went low, tabbed to the next mob I wanted to kill and pressed 1 1 1 1 1 2 and CTRL+1 if needed.
I could have pressed more buttons but I didn't really need to, so I just skipped through with my 1s and 2s, and cried boredom but told my self, hey the end game will be fun im sure of it!
No I'm not kidding. The AI in every FPS I've played, and I've pretty much played them all, is terrible compared to human AI.
I still haven't heard anybody list a single existing game (again, other than the very simple rule games in simple environments like chess or card games) that has a non-cheating AI that can beat a skilled human player.
List the game and the difficulty level please.
The reason it doesn't exist is AI in a complex ruled game with complex environments is extraordinarily difficult to program, and any effective algorithm if you could come up with one would take extraordinary amounts of computation power that would make it innefective in real time. In fact, you can mathematically prove that most brute force algorithms that involve the number of variables in a typical complex ruled game are computationally infeasible with any Turing based machine in our universe. Now if quantum computers come to reality the question is open for debate again as they break the rules of Turing machines.
And developers don't really even try that hard, so it is moot even talking theory. Just look how crappy a simple concept like pathfinding is in even modern games. Heck look how long it took before AI could compete with the great chess players, and chess rules are very simple and the grid is only 8x8.
GW2 "built from the ground up with microtransactions in mind"
1) Cash->Gems->Gold->Influence->WvWvWBoosts = PAY2WIN
2) Mystic Chests = Crass in-game cash shop advertisements
Ever played an early Rainbow Six game, specifically the 3rd one: Raven Shield? That game is impossible, even today(re-installed recently & regret it haha), I cannot peak around a corner in that game without having my head blown away and have no idea where the shot came from lol.
I swear those AI use wall-hacks!
There are definately FPS games where beating a mode at a certain level is difficult, but it is usually because you are one versus many and your damage tolerance is lowered. I have never had a problem beating a FPS bot in a one-on-one scenario for instance in a game that has bots play by the same rules as the humans. Bot AI was a big topic of research back in the Quake days. There were even some college theses out there on it believe it or not.
Now of course you can come up with a scenario like drop yourself and a bot in a straight hallway and he can beat you, but make the map complex at all and the weaponry varied then the bots get smoked every time...
GW2 "built from the ground up with microtransactions in mind"
1) Cash->Gems->Gold->Influence->WvWvWBoosts = PAY2WIN
2) Mystic Chests = Crass in-game cash shop advertisements
Yeah... Pretty sure what you are experiencing is largely an artifact of the scriptors purposefully not putting in unbeatable gameplay modes.
One specific example that comes to mind for me is Soulcalibur (Dreamcast). As anyone who's played the game extensively can tell you there is no "game mode" which is "difficult" as no difficulty setting prevents the AI from falling into easy tech traps which through correct punishment trivializes even the supposedly "difficult" games such as "Survival Mode".
However, if you go into practice mode with a CPU set on the highest difficulty setting you may notice a rather interesting property... Pretend the game is essentially "survival mode" where the only way to win is to knock the opponent out of the ring. The first ~10 times you knock the cpu out unanswered will be relatively trivial. The next 10 times will take you nearly three times as long. By ~50 the CPU becomes a DEMON. It will flawlessly execute guard impacts/parries, cancel 100% of all throws and punish absolutely anything that can be punished. From that point I believe the game has essentially capped out (I've made it to ~75 and it didn't feel vastly different) but the CPU is SO good at that point that it is basically inevitable that you will get knocked out at some point (resetting you back to 1 and the CPU back to "easy mode") so it's entirely possible it just keeps getting more and more "perfect" from there.
TL;DR
Clearly it was possible for them to make a ridiculously good AI for a fighting game that is pretty widely considered to have one of the largest skill barriers found in a 3d fighter (and it's not like fighters are a genre known for low skill barriers to begin with...) even given the constraints of the Dreamcast's pretty feeble processing power.
I just wish that AI mode hadn't been confined purely to practice mode gated behind a ~20 minute time sink knocking out chump AIs. Silly developers and their insistance that their games be "winnable". ;p
Now I do admit I am not big into the Japanese style fighting games, so I am not a good test subject for the AI there.
And to be honest that may be an example of a simple ruled game in a simple environment. As I recall most of those games are 2 dimensional on one screen with basically no obstacles. And your choice of moves is relatively small as well. A pure twitch game where the AI is probably limited to simple rules like "If move A counter move B, etc"
But that is the reason I asked the question, as it would be interesting to list all the games that have AIs capable of beating humans. I just know that in the genres I play I haven't seen it yet: FPS games, strategy games, and RPG games mostly (in addition to MMOS where I prefer the PVP aspect so AI isn't usually an issue for me).
I was probably to hasty to generalize AI as only good in chess and card games, as there are probably a number of games that break down into very simple ruled games in simple environments. Japanese fighting games probably an example of that.
GW2 "built from the ground up with microtransactions in mind"
1) Cash->Gems->Gold->Influence->WvWvWBoosts = PAY2WIN
2) Mystic Chests = Crass in-game cash shop advertisements
I agree with gainevilleg. AI in FPS games is not smart. It is unnaturally quick and accurate, and cheats, but it is not smart. Its function is strictly limited to moving around the map, maybe along some waypoints, pointing any game characters and pressing "fire". Differences in AI in FPS games come from their ability to adjust to the player, other game characters near them and the surrounding environments.
Chess has a limited ruleset which makes it easy for an AI to play it. It is a whole different thing to program an AI that acts like a human player without cheating in for example an FPS game.
I skate to where the puck is going to be, not where it has been -Wayne Gretzky
Ok then the conclusion is solved , you can´t do it , Neither can I cause it´s not my specialty or field .
But I do know people who are able to easily do it , but nobody plays there games , cause just like your post , who is nothing but admiting defeat and trying to push off responsibility . the same EGO or ERGO problems presist , that there are people who can do it , just you and me are taken out of the equation .
Final , more you rant about how difficult it is , the more nonesense it sounds , offcourse it is a LOT OF WORK .
Like any programming , debugging , modding , cause you don´t want or cannot do it , doesn´t mean it is not possible.
So stop making tons of excuses , like boring programmers these days , who keep saying impossible .
Then you hire another who does it in 2 weeks .
In order that would be: "Yes", "Yes", "Yes", "Yes", "Yes", "Yes", "Yes", "Probably true", "It can't".
I didn't make this post to trivialize the difficulty of making intelligent AI because that is truly difficult. I made it simply to point out that I think you are giving WAY too much credit to the amount of "intelligence" required to be good in limited environments such as games.
Three years ago the general consensus of people in the AI field was that "Go" (a popular board game in asia) would NEVER be playable at any skill level by a computer because its complexity was divergent. Thanks to Monte Carlo search even at 19x19 it can now beat all but a few professional players. In another year or two humans will no longer be able to beat computers at Go.
Many FPS games (especially those allowing headshots) are nothing more than an exercise in reaction speed and quickness of aim (something aimbots have been exploiting for years).
Many fighters have strictly defined enough rules that a computer can be designed to play them perfectly.
Platformers are so easy to beat with AI that designing solutions for them is frequently assigned as undergrad work. (A real good friend of mine "solved" "Mario: Lost Levels" for his undergrad final project actually, so I can cite specific examples of this one...)
TL;DR
There almost certainly ARE common game archetypes which are difficult for current AI (and of course if that ever ceases to be true it should still be trivially easy to purposefuly design a game type which it is "bad at" at least for some time to come) but I would argue that there are many (if not most) archetypes which dominate the market and yet are simply "too easy" for AI.
EDIT: And yes... Ken Jennings already got wtf pwned by a computer on Jeopardy. While exceptionally cool that game is probably not of the type that the OP is talking about.
With a global cooldown on almost all abilities in mainstream mmos point number 2 is irrelevant. Regarding number 3 and 4, they only apply if you're not playing with a good group of dedicated players. Compare running with friends to running with randoms. We never had accidental cc breaks, and we maybe hit the wrong button once every hour or so... Doesn't really change the outcome of a fight..
I'm starting to suspect that your definition for "simple ruled game in a simple environment" is no different than "games that an AI can master". You've formed a tautology. You're asking us to give an example of an AI that can beat skilled players in a non-simple game, but for any game that someone holds up as an example of that, you can just respond "well if the AI can beat people then that must be a simple game".
The whole point of the thread was to show that the tank and spank combat is half of the reason why we have dull combat, many forget that. But I do see that AI does bog down server resources. I do have an idea for a tier up on combat for both melee and spell casting.
I am talking about AI making decisions within combat more so than, "I am dying, I need to flee or yell for my comrades." The way I figured how a more intelligent AI would make decisions and to adapt to a player is essentially by collision detectors. There would be a very complex "if than statements" and the AI would calculate the best decision to adapt, they would also detect patterns within the players strategy to adapt. So in theory not as complex but the coding could be complex.
Some of you say that, because the WOW crowd whined, and moaned about a challenging combat with Catacysm, Blizzard nerfed it. Gameplay like what I am describing is not for the WoW player who wants their hand held. This gameplay is for players to use their brain once and a while and create strategy when needed, a challenge, a challenge in which this genre is lacking. I guess you'll have to see my combat mechanic design to full understand.
I don't think you understand. The computer AI knows exactly the moment a cooldown ends, and responds with the next spell or ability instantly. For a human it requires you to see the cooldown end, process the information, select next spell/ability(you may have already decided upon this), and then physically press the button to use your decided upon skill/ability. This takes you anywhere from 0.2-0.5 seconds to do. It doesn't matter if it's global or not, you still have to do this. Just try pressing a stopwatch button to start it and stop it as fast as you can. It takes about 0.2 seconds for you to do because your body is only physically so fast, and that's without having to make any decision or take in any external input, like visual or sound cues. This is the same reason the IBM Jeopardy AI always beat the humans to the buzzer if it had found an acceptable anwer. IT IS FASTER. With 15 abilities in a fight this really adds up, even if you are extremely fast, the AI will just beat you to it.
For example say in WOW you enter an instance where you and your party have a little cutscene and then a fight begins against an AI controlled party. The AI controlled party will know exactly the moment the cutscene ends and the fight begins, and begin casting their CC spells immediately after the cutscene. Even the best of the best human would take 0.2-0.4 seconds to see the visual cue the cutscene has ended and press the button to start their CC. Guess who gets their CCs off first...EVERYTIME. Now with your party CCd the AI would then just tag team your healer, kill him and move to the next clothie. By the time your party is even in shape to fight the battle is essentially over. Not because there was some super complex AI in place. All it has to do is react instantly, CC everyone it can, and focus fire on healer, and then as long as it doesn't act completely stupid the rest of the fight, you lose.
As I've said before. Aimbots aren't AI.
Aimbots solve the following problem:
If a player is at position (x,y) then where do I shoot? Answer: position (x,y)
They cheat. They have access to information players don't. If players could macro their "shoot" key to automatically headshot then who do you think would win? Those bots or human players?
AI with equal stats and information playing to win can not beat human players with equal stats and information playing to win.
AI with superior stats and superior information can easily win, unless they're faced with human players that also have superior stats and superior information ( AKA open console. AI.setHealth(0) ).
Aimbots aren't good AI. They're not even AI to begin with.
We are the bunny.
Resistance is futile.
''/\/\'''''/\/\''''''/\/\
( o.o) ( o.o) ( o.o)
(")("),,(")("),(")(")
Some of the hardest (and most interesting) fights in WoW were the PvP like fights. You got a random mix of opponents, and those opponents would fight you much like actual players would. There was one in Magister's Terrace and one in another instance that escapes me. Those were the encounters that I actually remember and I enjoyed them a lot. Of course, I enjoy PvP a bit more than PvE too, so it makes sense that I would enjoy those encounters.
Better, more human like AI can and has been done. You can't script the encounters though. The Scripted Boss Encounter is a staple of raiding...people like it or they expect it. It allows guilds to spend time learning encounters and getting it right. People who don't know the encounters need to spend time learning them. I think that's the main thing...with the human like AI, you could walk in and one shot encounters right off the bat. With scripted encounters, there's always a learning curve, time spent getting the encounter right.
I think the better AI for specific mobs makes sense and if you designed your game with that as a basic mechanic instead of the scripted encounter as a basic mechanic. In games where you learn the scripts for fights and spend a bunch of time gearing up for the fights, it doesn't make nearly as much sense. Perhaps if you had a mix of the two...some scripted mechanics you have to learn, and some better AI for the people who can just react.
I can not remember winning or losing a single debate on the internet.
You could easily make an AI that doesn't cheat by just shooting player at (X,Y) and instead "sees" the player and then shoots, and have it never miss. That would be trivial, and it needs no more information than would be available to you.
We're not talking about platformers. We're not talking about shooters. We're not talking about board or card games.
We're talking about a MMO server executing an AI that's on par with players PvPing. Not only does an MMO have so many more variables then shooters, platformers and board games have but it also has to be executed for thousands of mobs.
And I don't doubt that in 10 years we may see MMOs that can do those things. Just like we now have AIs that were impossible 10 years ago.
But current day it's impossible to make an AI that can beat human players in an MMO without greatly inflating their stats. I'm even leaving the superior information out. You can have that.
But even so an MMO has simply way too many variables for it to work. You're not dealing with 10 possible weapons and 10 possible moves. You're dealing with 100s of possible abilities and moves.
And you have to fit all that into something that can be executed thousands of times per second.
We are the bunny.
Resistance is futile.
''/\/\'''''/\/\''''''/\/\
( o.o) ( o.o) ( o.o)
(")("),,(")("),(")(")
Evasia, I am envisioning something very similiar to what you see here. I do agree that if AI becomes a major deterrent to factor in new combat and a more indepth gameplay experience, it wont be something coming out in the next 5 years. So I think you've hit the nail on the head. I have design combat mechanics that would compliment such an endeavor, but its all in high concept document form. Need to bring it to life to see if theory works.
This is exactly the problem and something I am trying to fix. Mindless and no purpose driven combat. I'd personally rather become a tactician and strategize against the opposition then consistently, repeat, snare + frost blast + fireball + stun + fireball + snare + lightning bolt all the time to kill a mob. That could work sometimes but I'd like to be challenged. If people would just pay attention to the purpose of the thread instead of taking a different street, they could see what I am talking about, adaptbility.
Thus explaining the fact that we currently have AIs capable of instant object recognition in graphical pictures in nearly all circumstances....
I wonder why google is still looking at the pages around images and other text instead of just having this AI scan the images and instantly tell you what it is.
We are the bunny.
Resistance is futile.
''/\/\'''''/\/\''''''/\/\
( o.o) ( o.o) ( o.o)
(")("),,(")("),(")(")
Way to go. You just related a game with a very very very very limited number of objects, and even fewer objects that actually move with the real world which has billions upon billions of objects and said the recognitiion needed for the AI would be the same in both.
Is it possible to talk about AI behavior within combat making decisions and reacting within combat? Is it possible to get this thread back on it's rails?
We are talking about it. Some of us believe they could make AI that would be pretty good at this already, but they don't because people would just whine and complain about it so instead they make "fun" AI instead of "good" AI. Others think that a "good" AI is currently unnattainable which is why developers opt for the "fun" AI. This argument is central to your point.
fewer object moving?, wow your ignorance is amusing, even if you are not moving or a object is not moving that don't mean the computer stop calculating that object, also the simple fact you move make he calc new object who enter in your view and recalc the old object who is your view, just because you moved.
you simple are lost on your own wish, first you wanted a AI who would make the game fun, now you just want to have a strong NPC, like you said you can make a bot who land 100% of thet time and react the moment you enter his LOS, then what? you will die, no change of reaction, is that what you want? for that you don't need a true AI.
again I will say go study about AIs before commenting anything new, you don't even KNOW what you want.
I fail to see what 2mb lines have to do with ai. o.o
This whole thing depends on the type of combat. If its WoW, there are those things like murlocs running for help, enemies with counterspell, etc.
If youre talking about AI on a solo gameplay basis then theres not much you can do about it but maybe add a small algorithm for ranged enemies to run around a bit. This running around part would take a toll on the servers if there are a lot of ranged enemies. For example, trigerring the murloc running away thing when ranged enemies get hit with melee (would make it too annoying for melee chars)
The main problem for this becomes that...The AI has to be really smart if youre thinking in terms of soloing. In group play, its a lot easier to make it feel more dynamic by tweaking the aggro system so that the guy who does the least damage but has high defense gets ignored a lil more. Again someone stated this already but people yelled nerf for it in cataclysm.
If you were to just try to improve the enemies ai in terms of groups then the game would have to be built where the combat lasts maybe 15-20 seconds and where everyone will put on their toes. This would mean that tanks will get a constant aoe aggro that affects maybe 2-3 enemies and works mostly on melee focused characters and that ranged enemies or certain melee enemies having a higher tendency (maybe resistance or immunity) to the taunt and will most likely go for the most dangerous one in your group, etc.
Having the combat react that way is actually very plausible. You could take WoWs current AI, do a couple of tweaks on the numbers, add a few taunt immunities and maybe add a little code to make it so that threat is always going if a person doesnt do as much and then youll have quite a dynamic group based ai. If the players are using the trinity system, either let the AI use it as well (wouldnt you like to be forced to attack the toughest guy while their hard damage dealers go for your priest?) or at the very least make combat not as reliant on the constant where the enemies will always without a doubt be on your tank. That way, everyone will be on their toes as to who to kill, how to protect the slightly squishier people, etc.
AI for this kind of combat will be as simple as WoW and would bring out quite a dynamic combat system. The problem with it is designing epic content with it. How would you like it if you were fighting a big dragon and it always seems to smack your tank once then proceed to kill everything else?
You could argue that "Oh that doesnt work for solo and itll still be easy mode" But really theres not much to do about soloing. Causing enemies to circle strafe you or run away annoys the !@#$ out of people because of the adds. Giving enemies access to a lot of skills would make them too hard for a lot of people and it really would NOT be solo friendly. Letting the solo enemies run away and lure players to traps will not be feasible in an mmo because of adds and the fact that people while cry nerf. So the only way to have the dynamic system if its part of the core system and a lot of the content will be designed that way (just like a lot of the systems you see now is basic tank n spank + 1 or 2 gimmicks)
Also with this kind of system means that classes will have to work differently and roles will be very broken off from the regular trinity game. Which would mean that the combat system will be more innovative and we all know how innovation gets handled.
So no...Linear AI is not the reason for me. Its the Tank and spank system that makes it extremely dull for me.
''/\/\'' Posted using Iphone bunni
( o.o)
(")(")
**This bunny was cloned from bunnies belonging to Gobla and is part of the Quizzical Fanclub and the The Marvelously Meowhead Fan Club**