A better question is why you're supposing it was about anything but PvP :P I think I mentioned quite a few times earlier that the issues we're discussing are really only problems in PvP. It's rather easy (or at least, much easier) to balance human vs. mob.



Indeed you do, but see KaVir's amusing RPS example as to why this is not fun.



I think the very definition of skill in this context implies the ability (and proficiency) to learn and adapt. It's one of the cornerstones of what makes humans, human.

AI in general encompasses a ton of algorithms and designs that I would not consider skill. Which qualifies more as skill - the human who studies his opponent's tendencies, attitude, and behavior to predict the best move, or the computer script that blindly calculates all possible future moves and their subsequent percentages for likelihood of a win? I vote for the human's method. Unfortunately processing power trumps human skill in a vast majority of lesser complexity games. I've yet to see a mob script on any game that could trump a player without relying on cheating (unfair stats, doing things the player isn't able to do, etc.), though.

In what we're proposing, everybody has even odds in the global scheme. You have your strengths and your weaknesses. It's not as if somebody is particularly weak and somebody else is particularly strong. Rather, you simply need adapt your playing style to match your strengths and avoid your weaknesses. If the game is good, you will have equal opportunities for reward no matter what style of play you choose.


If criticizing a parody of an idea is a valid argument . . . (Hey, actually, there's a name for that: "straw man" argument. :wink:)


I'm not sure what exactly your point is. I'm also not sure why the proficient human gets to be considered "skilled" but the program that applies machine learning is unskilled. Perhaps one is exhibiting a form of intelligence but the other isn't really, but the computer can certainly be more skilled at a task than the human is.

I think it's rather interesting how easy it is to deduce who, in this conversation, is a fan of PvP and who is not. If you aren't a PvP fan, you simply aren't going to see the situation through that point-of-view. I understand where you're coming from (and I'm actually a fan of systems like this in games where PvP does not exist), but such a view on balance really only applies to games with no PvP.



First off, machines don't learn (they can be conditioned, but that's an entirely different concept). If they did, they'd most likely have taken over the world by now and this conversation wouldn't be happening.

Secondly, being skilled at a task, such as mathematical computation, is different than being "skilled at PvP in a MUD". One is a very concrete concept, while the other reaches outside the bounds of the formal definition for skill. It's quite easy to quantify skill for concrete concepts, but you can't do so for things like we're talking about. What qualifies as "skilled" in terms of PvP? Adaptability? Surprise? Unpredictability? Improvisation? The list goes on…

Such deductions can also be tenuous at best… :shrug:

Even assuming you were right about who likes PvP and who doesn't, the rest of your statement doesn't follow anyhow. Just look at existing PvP systems and see how obvious it is that some classes on their own simply will not defeat other classes on their own. I have no idea why you think that PvP somehow excludes the model we have been discussing.


I don't really care to get into an argument of what constitutes learning and why some learning is "learning" and other learning is "conditioning". In part because I doubt that anybody here (myself included) can actually provide satisfactory definitions of what exactly learning is in the first place. Suffice it to say that computers can become better at tasks over time and for all intents and purposes can develop skills at a given task, be it recognizing digits from an image or flying a remote-controlled helicopter upside-down. (And yes, the latter example is a real one.)


Why do you get to decide what the formal definition for skill is, and what is easily measured? It seems to me that there is a very simple definition for "skill" in PvP in a MUD. Skill is what allows you to establish your success rate. A creature is highly skilled if it has a high success rate, and not skilled if it has a low success rate.

Let's look at the definition for skill:



Seems that all of these are very easily translated to everything we've been talking about. PvP is not outside the bounds of any of these definitions.

Frodo & Sam wtfpwned a ton of orcs when they destroyed the ring. don't you watch movies :p

I very specifically said "based purely on race and/or class". In other words, if the only thing you know about the characters is their race or class, you shouldn't be able to predict a likely outcome of the fight.

The fact that a human player may be able to compensate for his disadvantage is a separate issue entirely. That same player would obviously have an even greater advantage if he were playing the "stronger" race or class.



No, that is absolutely not what I'm suggesting. What I'm suggesting is that you shouldn't be able to predict a likely outcome if the only thing you know is race and/or class. Instead, advantages and disadvantages should be based on things that can be adjusted. You don't play "Scissors" - instead, every race/class has a selection of Rock, Paper and Scissors abilities to choose from.

Can you predict who will win a game of Warcraft III if the only thing you know is the race they've chosen? Can you predict who will win a game of Magic the Gathering if the only thing you know is the colour combination they've chosen? There are far more interesting ways to make the options different than a +1 strength bonus.

Right… so no matter what race/class is chosen, everybody has the same chance of success modulo player experience. I suppose there is equipment as well but I assumed that was clear from previous remarks.


Gee, you're right, I should have clarified earlier that we're not talking about a completely simplistic Rock-Paper-Scissors approach. :rolleyes:


Comparisons to RTS and card games are not entirely appropriate given that the whole paradigm is quite different…

No, I'm saying that other factors should be used to determine the chance of success. When two warriors of the same level meet, it's perfectly reasonable for one to have an advantage over the other - but that advantage shouldn't be simply because one is an orc and the other a halfling.



You are talking about an approach where race X has an advantage over race Y, who has an advantage over race Z, who has an advantage over race X. That's still just a watered down version of RPS.



In a non-PK mud where grouping is a requirement, the emphasis would be different. But if it's a competitive PK mud where players are expected to be able to play solo, then exactly the same balancing principle applies as it would with games such as Warcraft III, Age of Mythology, Magic the Gathering, etc.

A common form of RPS balancing in RTS games is spearmen > cavalry > archers > spearmen. Victory isn't guaranteed, but (as with your race suggestion) certain units do have a clear advantage over others. However the reason that works is because all players can train all of the unit types - if each player could only select one unit type (spearmen, cavalry or archers) for the entire game, two thirds of the games would reduce the value of player skill (because that skill would need to overcome the unit dis/advantages before it became a factor).

In a mud, each PC has only one race, so if you provide that race with inherent advantages over other races it will be comparable with an RTS game where each player can only pick one unit type. What you should instead do is base those advantages on things the players can change - just as players in an RTS game can start training spearmen to counter their opponent's cavalry units, so in the mud you might go and buy a spear to counter your opponent who always fights from horseback.

As I've said before, there are plenty of ways to make races interesting and completely different from each other. If the only difference between them is the degree of bonus they give, then there's a serious problem with the game design.

I've yet to come across a decent game where any given class is intentionally more powerful than any other given class in this respect. I'd be interested to see what PvP systems you're referring to when you make this claim.



It's quite easy to differentiate; computers only do what you tell them to do. That's, by and large, the defining difference between learning and conditioning. Whether or not you believe humans were "programmed" by some higher power is getting way more philosophical, but it's definitely the most popular counter-argument (that I don't subscribe to, but whatever).



I didn't decide anything (which is why I posed those questions); I merely recognized the significant difference between skill in the context of concrete and abstract concepts. Success rate is not equatable to skill (though it can be related to it). In our particular discussion, if an extremely skilled player is playing a race that is vulnerable to fire and tries to fight some other arbitrary player who has uber-mega-fire-spell-of-ownage, his success rate is going to be practically nil. Does this make him less skilled?

It's not that a given class is more powerful than any other class. But if you want an example, look up Dark Age of Camelot for just one.


I could tell you that you're just doing what your brain chemistry does. No higher powers. Besides, how do you know that humans aren't also learning by conditioning? You've entered very shaky ground that even expert researchers disagree on…

Anyhow, I don't see what point this serves. The only relevant point is that a computer can perform a given task, and get better at it over time. It doesn't matter how or why…


I see no difference whatsoever. You posited the existence of a difference but did not explain it. You also didn't explain what makes a concept abstract or concrete in the first place.


Well, I did say success rate. In order to be a non-trivial "rate" in the first place you have to be considering multiple instances.

If that's not what you were talking about, then the only other thing I can think of is a multi-classing (or remorting) type of system, but even on these systems, the end result always strives for balance, so that no one "path" is better than any other "path". Whether or not a single class is balanced against any other class in these systems only relevant based on implementation. In such games where PvP is possible/encouraged while you only have 1 class, all classes tend to be balanced. The same goes for 2-class combinations, 3-class, and so forth. In such games where PvP is only viable end-game, class vs. class balance is irrelevant, and only "path" vs. "path" balance is considered.



Yes, and you'd be correct (in my opinion). This is called learning, since there is nobody conditioning you.



I never said humans can't be conditioned (they can); merely that humans are capable of learning, while computers are only capable of conditioning.



The point is that a computer is incapable of "skill" in the abstract context of PvP. It's only doing what a human told it to do.



Quantifiability. I gave the example of mathematical computation; it's easily quantifiable. A human can solve problem X in Y seconds utilizing method Z, whereas a computer can solve it in Y - n seconds utilizing the same method. It's therefor accurate to say that a computer is more skilled at solving problem X utilizing method Z than a human. You can't quantify skill in the same way as it pertains to PvP, however.



…and I said rate as well, and was considering multiple instances.

Uhh, fine, if you say so. I gave you a counter-example. :shrug:


I'm not sure why we care. The computer can do something, do it well, and get better over time. I'm not sure why we care at all about anything else. Why is this point even relevant, drrck? Even if I grant all of your points on this issue, what is gained by even thinking about whether or not a computer is exhibiting "skill"?


I already gave you a trivially easy way to do it: it suffices to look at success rate. If you insist, then assure that you have sufficient samples to be statistically significant, which you need in a human's case anyhow…


It rather seemed to me that you were considering a specific instance of uber-whatever-resistance-it-was.

But even then, if somebody always loses, then it seems that they do lack skill; if anything, skill in choosing battles intelligently…

What about DAoC, exactly? What example does it fit that supports your claim?

DAoC is a team vs. team game. Its competitive side can more accurately be thought of as an RTS in which, instead of one player controlling all the units, multiple players control them. This "counter-example" really has nothing to do with what we're talking about.



From what I remember, this tangent began as someone earlier trying to demonstrate that just because an RPS-style system works for PvE, it must also work for PvP. This is not true because of the fact that computers don't exhibit skill. It's much easier to overcome disadvantages against mobs using skill than it is against humans.



And I subsequently gave you an example (which you ignored) of how success rate can not be used to determine skill. Just because I happen to have better equipment, be a higher level, do more damage, exploit your weaknesses, etc. doesn't mean that I am more skilled than you. It does, however, mean that my success rate will go up.



See above for why the degree of advantage/disadvantage doesn't matter; the success rate can/does still go up independent of player skill.



Now you've come full circle, back to whether or not your choice in opponent is skill/strategy. I'm going to leave this topic alone, since it was rather clear earlier what my stance on it was.

Oh, sure, redefine PvP, while you're at it. :rolleyes:


Your definition of "skill" is a little too nebulous for my tastes to make this statement useful. An AI-controlled mob can get better over time and react to new situations. It might not have that "human touch" but it's not necessarily as stupid as you make it out to be. You keep claiming that computers have no skill and yet never address the fact that they can in fact learn to get better at tasks. (Substitute "learn" for whatever term floats your boat.)


I would have assumed that equipment and level would have to be normalized for any comparisons to be worthwhile in the first place, so I'm not sure why you even bring those up.

As for exploiting weaknesses: how can that not be an indication of skill? If I let you exploit my weaknesses, then clearly I have made a mistake somewhere!

Besides, like I said, a non-trivial rate involves a lot more than just the individual me vs. the individual you…

DAoC is TvT, not PvP. There's quite a difference (especially in terms of balance).



That "human touch" is exactly what differentiates skill vs. no skill. Also, I've yet to see adaptive AI in a MUD. Assuming it exists, somewhere, it's still never a match for human skill, and that's why coders have to compensate by allowing mobs to cheat.



Why would you normalize these factors when trying to determine if success rate == skill? That makes no sense, whatsoever.



If your weakness is built-in (read: part of your race), then you can't exactly do anything about me exploiting it now can you? Thanks for proving my point ;)



It involves the arbitrary individual player vs. any other arbitrary individual player, but is not global. You can't say that you're "balanced" because you can't beat me, but you can beat player X instead.

Let's not use our terms willy-nilly. DAoC most certainly is player vs. player. It does however tend not to be individual vs. individual. Nonetheless, even in team vs. team, there are clear cases where some teams are better than others "just because". So it's still an answer to your question several posts back.


Tell me why I care! You keep talking about skill, no skill, and you keep failing to consider the crucial point, which is that there is nothing at all that prevents an AI from learning. Sure, maybe you haven't seen it in a MUD, but that doesn't say anything about what's possible: besides, I was under the impression that we were talking about design issues for new MUDs, not old ones…


Why, for exactly the same reason why one normalizes class levels of students when judging their intelligence. "Oh, look, the 12th grader beat the pants off of the 5th grader, clearly the 12th grader is a whole lot smarter!"


So, uh, did you intentionally leave out huge swaths of the conversation in that reply, or did you just forget that we've mentioned over and over that being strategic overcomes this kind of issue?


And why not? Let's return to the simplistic RPS game for a moment. It is obviously balanced: everybody has precisely the same odds of success against a random opponent.

You're going to have to define balance very precisely if you want to claim that this is not balanced. What exactly does balance mean for you?

Ok, how about EVE-Online? I think EVE qualifies as about as hardcore PvP as any MMO out there. There are no classes at all in EVE, just races and skills. The race you pick determines the skills you start with, and that in turn determines the type of ship you're likely to fly and the kinds of weapons you're likely to use. If you play an Amarr, you are almost certain to be using lasers. If you play a Caldari, you're a fool to not use missiles or railguns.

If you pit a punisher (Amarr frigate) against a kestrel (Caldari frigate), the kestrel will probably win unless the Amarr pilot is very good or has spent a LOT of cash to trick out their punisher. In fact, Amarr will typically do poorly against Caldari unless they have superior equipment or numbers. Why? Because Caldari defenses are shield based, and lasers do poorly against shields (but very good against armor and hull).

Does that mean Amarr are unbalanced? Not if they fight the Gallente or Minmatar. Not if they fight a prolonged battle where the Caldari will run out of ammo (lasers don't take ammo). But in the specific one-on-one duel, yes.



So, where do neural nets fall? How about expert systems which self-reinforce or diminish action weights based on past outcomes? If you attack my NPC and win by using shield bash, and my NPC notes the fact and starts using a counter-measure when it sees you, because it "knows" you like shield bash, is that learning? Black and white are both just extreme shades of grey.



Yes it does. If he continues to fight the same opponent without finding a way of mitigating the disadvantage, I would definately call him unskilled, or extremely bored. Perhaps he is a computer program, since he hasn't learned that he can't win unless he changes something about the encounter.

Regardless of term clarification, that situation is not equatable to our conversation, because teams can be altered on the fly to better match an opponent. Race cannot.



I said, assuming it exists (or is to be created for a new MUD), it's still no match for human skill.



Which, in turn, would prove that intelligence is a factor of age, not just some inherent attribute. Likewise, success rate is a factor of <level, equipment, etc.>, not just player skill.



You were just arguing that exploiting a weakness is to be considered skill because there exists the ability to counter it (you said "let" me exploit it). If the ability to counter it is removed, then so is the skill, and the outcome is now no longer who is necessarily "better", but who has what race.



What if 98 out of 99 players are rocks? Ask the 1 scissors player if he thinks the game is "obviously balanced". Your comment makes the assumption that race distribution is completely even at any given time, and that's, for all intents and purposes, impossible. It also assumes the opponents a player has advantages over are "equal" to the opponents some other player has an advantage over. What if 33 players are rocks, 33 are scissors, and 33 are paper, but all the paper players are level 1, while everyone else is level 10?



The actual definition of balance is rather universal. What isn't universal is determining what context of balance should be considered for a particular situation.



Edit: Intelligence is not a factor of age; rather, age is a factor of intelligence. Likewise for the other similar comment I made. I guess I was writing faster than I was thinking!

I don't know anything about EVE, but you mentioned that you can have "numbers", which implies that you can construct fleets of different vessels, no? If this is so, then it's the same situation as DAoC or any RTS.



Neural nets only work properly when there are goals to work towards, or "bad things" to avoid. Supplying these conditions is a low-level form of conditioning. Without them, a computer doesn't know what it should or should not do and can't analyze past outcomes because it doesn't know what is favorable and what is not. It would just be the most complicated random number generator ever.



As I've already told DavidHaley, I'm not going to revisit whether or not choosing not to fight can be considered strategy/skill.