Curse

Fril Estelin · Nov 28, 2008, 09:13 AM // 09:13

Quote:

Originally Posted by aapo

Question is how difficult it would be to code chess-winning bot and how difficult it would be to code GW-winning bot.

The change between the 2 games is not exponential, they're of the same class of problems and thus require roughly speaking the same difficulty for programming bots (of course, that's ignoring the 2.5D nature of GW1 and technical computing stuff). In simple words: the fact of adding more pieces to the chess game and a bigger chessboard (new pieces=skills, interrupt, dodge, etc.) doesn't change the nature of the game. As IBM's Deep Blue has shown, it's about sheer processing power, memory and strategising/prioritising moves. In theory Anet can create an AI team that'll beat any human team, but the cost is very, very, very prohibitive. So in practice, humans are much better/faster/efficient at playing the game that heroes (when they know what they're doing, which is a lot ...).

tmakinen · Nov 28, 2008, 09:18 AM // 09:18

Quote:

Originally Posted by aapo

Question is how difficult it would be to code chess-winning bot and how difficult it would be to code GW-winning bot.

And the answer is: writing a bot capable of playing chess on grandmaster level is easy since chess is a discrete state game with a very easily searchable tree. Getting better is just a matter of throwing in more computational power. Download some freely available chess bot like Fritz if you want to be soundly beaten. Writing a bot capable of independently playing GW PvP on top level just isn't feasible. And to make it clear, this is not a question about aimbots (computer augmented human play) which are just cheating. Granted, AI can have superior reflexes but it completely lacks insight and is thus unable to anticipate future actions if they are not easily numerable like in chess.

Quote:

Originally Posted by Fril Estelin

The change between the 2 games is not exponential, they're of the same class of problems and thus require roughly speaking the same difficulty for programming bots

That's not true. Chess is infinitely (used in a mathematical sense) simpler than GW. Chess is turn based, has discrete states, each state has a only a small number of possible moves and the relative merit of each state is fairly straightforward to evaluate. Compare this to GW which is real time, does not have discrete states and at each instance there is a mindbogglingly huge number of possible moves (as a combination of all possible moves of every player) making it impossible to search the movement tree much less evaluate successive positions.

Fril Estelin · Nov 28, 2008, 09:28 AM // 09:28

Quote:

Originally Posted by tmakinen

Writing a bot capable of independently playing GW PvP on top level just isn't feasible. And to make it clear, this is not a question about aimbots (computer augmented human play) which are just cheating. Granted, AI can have superior reflexes but it completely lacks insight and is thus unable to anticipate future actions if they are not easily numerable like in chess.

You're theoretically wrong tmakinen, it's perfectly doable, just too costly (and not very useful...). Advanced AI like neural or evolutionary programming can already do some nice stuff, but expert knowledge can lead to extremely competent AI, which combined with superior speed and memory will outperform a human team. But we're talking supercomputer-like computing (which could be done with a Mac G5) running very, very costly software here. As I said above, the state space of the problem hasn't changed in nature between chess and GW, sure the CPU-memory cost is growing linearly which means that usual programming will require exponential growth in CPU or memory, unless you're doing very smart stuff.

So in practice your point stands, but not in theory

. (which means that you may be able to get close enough to human-like abilities, see the Turing Test for the Loebner Prize)

aapo · Nov 28, 2008, 09:35 AM // 09:35

Quote:

Originally Posted by [DE]

You are assuming that guild wars is a one-dimensional game. And the idea of anyone, even progamers from korea who play starcraft, microing 7 heroes + there own bar to do everything an 8 man team does it GvG is extremely ridiculous. It's hard enough to play one bar in GvG. But you're suggesting that you can micromanage EVERYTHING. You're gonna need more than 400+ APM and a much superior brain I'm afraid.

- I just thought that maybe we have a bit different definition for the word 'skill'. 400 button presses per minute requires nimbleness and endurance. However, the game of Starcraft is largely pre-defined, so such thing as button presses per minute directly correlates with your ability to play the game. I'm thinking skill as in how many different factors player has to consider. In Starcraft some units are more effective against certain other units. In GW, you can't know what sort of armor your opponent has and thus you're stuck with choices you made on outpost before starting to play. When playing Monk, 'skill' factors are awareness of battlefield and knowing which skill is most appropriate for countering which threat, something that would be easy to code on bot. After this it's a question of latency, fatigue, communication etc. which aren't really what I would consider "skill". Some people get along well and thus raise their ranking in guild ladder, but that doesn't make the game to be about skill.

tmakinen · Nov 28, 2008, 09:50 AM // 09:50

Quote:

Originally Posted by Fril Estelin

You're theoretically wrong tmakinen, it's perfectly doable

Chess is a tree-searchable game because of the qualities I listed above. Let's see what happens if we try the same approach with GW.

Since GW runs on computers which are discrete, the game only simulates continuity and thus there are states which can be counted. Let's say that the discrete unit of time in GW is around 10 ms, i.e., the game updates 100 times per second. Let's also assume that the average degrees of freedom per player at each unit time is 20 (i.e., at any given point of time, you as a player can decide to initiate about 20 different types of action). There are 8+8 = 16 players in GvG, thus providing 20^16 = 6.5e20 degrees of freedom at every 'turn'. If every atom in the known universe was a supercomputer at your disposal, you might be able to search half a second forward from a given position before the very last proton decays and the universe as we know it ceases to exist. It so isn't doable.

Shasgaliel · Nov 28, 2008, 09:55 AM // 09:55

Quote:

Originally Posted by aapo

- Are you sure you know it can be adjusted? ANET had to limit hero numbers in HA and GvG because they were too good.

Who told you that? They limited them to prevent title point farming not because they were good. If they havent limit them it would be again just boring hero battles in a bigger scale.

Quote:

Originally Posted by aapo

- That's not even the case. What makes good team in your opinion? I know: communication, coordination, ...

No it is not true. They are deciding factors in case some other requirements are met. The most important is area and situation awareness. Not always the least protected target should be attacked etc, also you need to be able to predict what other people can do. Skills are only a small part of gvg and HA what decides is adaptability. The problem why AI fails fighting with humans is because AI is predictable while humans are not. Try to teach your heroes that... I would like to see your warrior hero doing Chizu dancing for example.

Quote:

Originally Posted by aapo

The game works simply based on those errors. Take away source of errors (the human player) and you'll win.

Depends on the errors. Sometimes errors lead to win since they may open a possibilities not available before. How will you recognize that the error is not for example a simple trap? So far no AI can do that.

Quote:

Originally Posted by Fril Estelin

You're theoretically wrong tmakinen, it's perfectly doable, just too costly (and not very useful...). Advanced AI like neural or evolutionary programming can already do some nice stuff, but expert knowledge can lead to extremely competent AI, which combined with superior speed and memory will outperform a human team. But we're talking supercomputer-like computing (which could be done with a Mac G5) running very, very costly software here. As I said above, the state space of the problem hasn't changed in nature between chess and GW, sure the CPU-memory cost is growing linearly which means that usual programming will require exponential growth in CPU or memory, unless you're doing very smart stuff.

So in practice your point stands, but not in theory

. (which means that you may be able to get close enough to human-like abilities, see the Turing Test for the Loebner Prize)

There is no CPU able to handle unlimited possibilities. So to make decision they will need to rely on estimations. Which are biased by default. You can optimize but the random factor still counts. Therefore there is no way Chess is similar to GW. In theory you can make perfect chess bot but for GW not. Self learning AI does not help. Just make a situation relatively new and you have great chance to win. Of course next time your trick will not work but you can do another one.... The main argument in my opinion is that humans will be always better at predicting other humans than any AI which does not matter much in top level chess but matters a lot in GW pvp.

aapo · Nov 28, 2008, 10:02 AM // 10:02

Quote:

Originally Posted by tmakinen

And the answer is: writing a bot capable of playing chess on grandmaster level is easy since chess is a discrete state game with a very easily searchable tree. Getting better is just a matter of throwing in more computational power. Download some freely available chess bot like Fritz if you want to be soundly beaten. Writing a bot capable of independently playing GW PvP on top level just isn't feasible. And to make it clear, this is not a question about aimbots (computer augmented human play) which are just cheating. Granted, AI can have superior reflexes but it completely lacks insight and is thus unable to anticipate future actions if they are not easily numerable like in chess.

- GW doesn't have many actions, most which have only one consequence:

Movement (forward, backward, left&right -strafe)
Skills (1-8)
Selecting target (player-controllable)
Weapon slots (1-4)
The attack button
(total 17)

There's 20 possibilities for the first move in chess. It should be noted that movement is secondary to objective in Guild Wars, i.e. it makes little difference whether you kite this or that way, but formation is the key in chess. It's more difficult to calculate which piece should move in chess and in what direction, than it is to either approach or keep away from certain enemies in GW.

tmakinen · Nov 28, 2008, 10:27 AM // 10:27

Quote:

Originally Posted by aapo

- GW doesn't have many actions, most which have only one consequence (...) (total 17)

There's 20 possibilities for the first move in chess.

Ok, there are 20 possibities for the first move in chess. There are 17^16 = 5e19 possibilities for the first move in a GvG match (if there are 16 players, each of them having 17 possible actions). An average chess match contains about 100 turns (actually half-turns since only one side moves at a time). An average GvG match contains about 120000 turns. Tell me again which game is linear?

Well, although this is certainly game theory, it is also as much out of topic as you seem to be out of depth here.

aapo · Nov 28, 2008, 10:40 AM // 10:40

Quote:

Originally Posted by tmakinen

Since GW runs on computers which are discrete, the game only simulates continuity and thus there are states which can be counted. Let's say that the discrete unit of time in GW is around 10 ms, i.e., the game updates 100 times per second. Let's also assume that the average degrees of freedom per player at each unit time is 20 (i.e., at any given point of time, you as a player can decide to initiate about 20 different types of action). There are 8+8 = 16 players in GvG, thus providing 20^16 = 6.5e20 degrees of freedom at every 'turn'. If every atom in the known universe was a supercomputer at your disposal, you might be able to search half a second forward from a given position before the very last proton decays and the universe as we know it ceases to exist. It so isn't doable.

- More like discrete unit of time should be 200ms, which is the human reaction time. Besides the server doesn't even register that frequent movement. Degrees of freedom are 17, which it much less than chess. Besides your argument is flawed. Program doesn't have to anticipate like that, when all necessary information is available just like for us humans.

Example: How to defend
Calculate how many strikes of adrenaline enemy Warrior has by adding together connected hits and subtracting used adrenaline skills (which are shown). Calculate threat level estimation based on that. Calculate recharge times for every offensive spell enemy has used. Danger-level for own character depends on proximity to enemy. As enemies turn towards target they're about to spike, the program can anticipate it and use appropriate pre-prot method. If skillbars are correct, nothing dies (or dies less often than enemies).

Attack works like the best of Ritspike teams used to do: targets don't stay at full health, so it's easy to scan for low health targets and initiate necessary damage spells to kill them.

Remember: It doesn't have to strive for complete information. It only has to strive for necessary information. I'd probably start by coding spike capability and then start alpha-testing. But that's already available as macro for our current heroes.

tmakinen · Nov 28, 2008, 11:14 AM // 11:14

Quote:

Originally Posted by aapo

Degrees of freedom are 17

17 per character, I have now told it three times in a row and it still doesn't seem to register on planet aapo? There are 16 characters which can act simultaneously, thus the complete number of degrees of freedom in a situation is 17^16 = 5e19. You can't evaluate the merit of any single action in isolation from other actions.

Quote:

Originally Posted by aapo

which it much less than chess.

This is patently wrong. In any given situation, GW has several orders of magnitude more possible moves than chess.

Quote:

Originally Posted by aapo

Program doesn't have to anticipate like that

Well, that statement pretty conclusively disqualifies you from the discussion if there were any lingering doubts about it left.

*lemming* · Nov 28, 2008, 11:41 AM // 11:41

Talk about a major derail.

aapo · Nov 28, 2008, 12:33 PM // 12:33

Quote:

Originally Posted by tmakinen

17 per character, I have now told it three times in a row and it still doesn't seem to register on planet aapo? There are 16 characters which can act simultaneously, thus the complete number of degrees of freedom in a situation is 17^16 = 5e19. You can't evaluate the merit of any single action in isolation from other actions.

- You've gotta be kidding me. What happens when you take your three heroes to explorable area with 200 monsters? Universe implodes? Maybe you didn't get the memo that a computer program calculates only what it's programmed to calculate. Your heroes do nothing but follow your character around, except when enemies appear, when they will use their skills and possibly kite away. When this primitive setup can autopilot through all of hard mode with proper build, it's not very difficult task to program some additional features like skill recognition to meet competitive standards.

Fril Estelin · Nov 28, 2008, 12:34 PM // 12:34

Quote:

Originally Posted by tmakinen

Well, although this is certainly game theory, it is also as much out of topic as you seem to be out of depth here.

I'm taking this offline with you tmakinen, sorry for the derail guys.

pumpkin pie · Nov 28, 2008, 12:52 PM // 12:52

please move your personal non topic related argument, elsewhere lol. Thank you very much, and happy turkey day! except the turkeys

I was enjoying reading this thread and have actually given it a long thought, the conclusion is you cannot hypothesize a community of possibily 5 million players or even 1 million, supposing the other 4 are all mule account or xunlai prediction account.

that is why all the points brought forth are agreeable to me. lol

need more time to come to a conclusive "hypothesis" and Arena Net or any other online gamming company would kill to get a hand on that informations

DreamWind · Nov 28, 2008, 10:59 PM // 22:59

Quote:

Originally Posted by aapo

- I just thought that maybe we have a bit different definition for the word 'skill'. 400 button presses per minute requires nimbleness and endurance. However, the game of Starcraft is largely pre-defined, so such thing as button presses per minute directly correlates with your ability to play the game. I'm thinking skill as in how many different factors player has to consider.

I refrained from entering this thread, but this was too much. You don't know what the hell you are talking about sir.

aapo · Nov 29, 2008, 11:12 AM // 11:12

Quote:

Originally Posted by DreamWind

I refrained from entering this thread, but this was too much. You don't know what the hell you are talking about sir.

- In game of Starcraft, building units on your barracks and hatchery is always a good idea. The more you build, the better your odds of winning. In Guildwars, spamming Fireball (if you have that on your bar) on recharge is always a good idea. The more you spam the faster enemies die.

In Chess, capturing opponent pieces is not always a good idea. Capturing them might expose your more valuable pieces. In Poker, playing out every hand is not always a good idea. If you play every hand, you lose money if opponents have better hands.

Do you see the fundamental difference between these games? In latter two, what you do defines your success. In former two, there's clear objective and way of playing which players must strive to achieve. If you sit around with your energy full, you're not being effective player. If your skills are not recharging all the time, those skills are not being used effectively. Simple as that.

Avarre · Nov 29, 2008, 12:02 PM // 12:02

Quote:

Originally Posted by aapo

- In game of Starcraft, building units on your barracks and hatchery is always a good idea. The more you build, the better your odds of winning. In Guildwars, spamming Fireball (if you have that on your bar) on recharge is always a good idea. The more you spam the faster enemies die.

Unless, of course, you're cutting workers for faster unit production, or cutting army production for a double expansion/workers, or saving larvae for muta builds...

Pretty sure everyone knows apm is not related to skill.

In the case of Guild Wars, you don't spam attacks. DPS does not kill people, burst damage does - hence the spike-dependent meta. Thus it is not preferable to simply spam, especially in the case of any form of conditional/interrupt skill.

Abedeus · Nov 29, 2008, 01:02 PM // 13:02

Quote:

In the case of Guild Wars, you don't spam attacks. DPS does not kill people, burst damage does - hence the spike-dependent meta. Thus it is not preferable to simply spam, especially in the case of any form of conditional/interrupt skill.

What about pressure builds?

Sure, they have burst DPS, but it's not that big. Just enough to finish off targets one by one.

Avarre · Nov 29, 2008, 01:10 PM // 13:10

Quote:

Originally Posted by Abedeus

What about pressure builds?

Sure, they have burst DPS, but it's not that big. Just enough to finish off targets one by one.

Even pressure builds generally rely on varying bursts of damage. Maybe some older builds (KGYU) are exceptions, but the bulk of pressure builds involve spikes of some sort. I can't think of any builds that involve outright spamming skills on recharge as was suggested - there's proper timing, targeting, and other factors to include.

maraxusofk · Nov 29, 2008, 01:38 PM // 13:38

Quote:

Originally Posted by aapo

- In game of Starcraft, building units on your barracks and hatchery is always a good idea. The more you build, the better your odds of winning. In Guildwars, spamming Fireball (if you have that on your bar) on recharge is always a good idea. The more you spam the faster enemies die.

i actually thought u were serious until i saw this. u sure had me fooled.

Quote:

Originally Posted by Avarre

Even pressure builds generally rely on varying bursts of damage. Maybe some older builds (KGYU) are exceptions, but the bulk of pressure builds involve spikes of some sort. I can't think of any builds that involve outright spamming skills on recharge as was suggested - there's proper timing, targeting, and other factors to include.

i do believe old school hex builds got the shaft quite badly, so pure pressure builds dont exist anymore (iono bout now i quit since march). oh and iway was the first major build that used spamming skills on recharge for the most part, but that was because back then atk spd wasnt capped at 133%, evic was at 6, and dying did not reset ur adred,