Does Online Multiplayer Always Make Sense?

[gardian06]

Fighting games have strategy. To imply there is none seems to me to be trolling. And I cannot believe I just signed up to this website (and got my password back in a plaintext email no less, a big security no-no), just to try and address this!

In fighting game's there is mix up. As a player is "waking up" (getting off the ground), they have options; block high, low, try to tech a throw, reversal, back dash, etc. All of which are beaten or lose to another matrix of options (but the player still standing up will have the natural advantage). Playing real people online and you have to adapt to their tendencies in order to win (do they like to do this on my wake up? What are they likely to try next and what is the best option against the two options he is most likely to try?)

You can randomise that with a bot, but it will in no way be the same.

And don't get me started on the AI in fighting games. I have never played a good one! They are all ridiculously easy to read and hence, beat! The only difficult AI's to beat are those where they have an unfair advantage, and that is in no way the same as playing another player using a character of equal (but different) strengths and trying to deconstruct their game to a point where you have them completely read and know what they are going to do before they do (one of the best aspects of fighting games, doing something your opponent thinks is in some way "psychic", when they thought I would never see it coming).

I can use a lot more examples than just wake up options, trying to zone and play footsies with the opponent, etc. etc. But, hopefully the above is enough to shed at least some light on why the entire premise of this article is just flat out wrong!

"you're against a complete expert who has memorized the most efficient possible move and counter for every situation"

Again, mix up. There is no right answer for every situation. There are options, and experienced players are more likely to pick options that are more likely to produce more favourable outcomes (like setting up a combo), but there is by no means a "I win" button in fighting games that pro players can constantly press to beat up lesser skilled fighting game players who haven't found that one button yet! There are risks/rewards, and the management of what you use and when (such as super meter, or a reversal dragon punch, which can be hugely punished on the way back down if baited with a block).

OK, seriously, I'm done now! XD

that was actually one of the points that he was trying to make that if the AI is designed in a good enough way that you can actually make it more difficult then a human. take this for example. If I created a fighting game with a pure learning AI assuming a rule set that only allows for 50 possible moves (yes I know this is a small number)then it is possible for the AI to completely learn each possible action the player can do within approximately 10 fights, and by nature win (no this level of learning is not cheating, but a lot of players think it is) though what actually happens in many fighting games is that they will start to dumb down the learning algorithm, or insert a far higher number of possible outcomes/moves like was done with Dead or Alive they actually introduce a good learning AI, but they dumb it down a bit to make it "fare". what you talk about mostly just boils down to call-response. (yes I know the limitations to an actual learning AI though many people don't) though it is completely possible for an AI to out perform a human, but there have many players would call it unfair. you make some valid points, but you do lack some level of understanding on how AI works, and how AI can mimic a player. in fact Yahtzee is talking about introducing an Alice for a fighting AI which by your attack indicates that you do not understand what that is.

 

[Mr.Pandah]

This is the WORST article from Yahtzee that I've seen to date. Yahtzee yields that he may be ignorant on the topic, but he is the most ignorant person I've ever heard trying to spew some ridiculous nonsense about something he has no right talking about.

I will never watch or read anything from Yahtzee again because this is about the worst thing I've ever read about video games. This is probably the only thing I've ever read that has actually offended me.

I actually feel this way, but to a lesser extent. Not that I'll NEVER watch anything from him again, but to say it's a waste of time? Please. Like others have said, just because you don't like it (Yahtzee), doesn't mean there isn't a point to it. Throw away your hipster glasses Yahtzee.

 

[hermes]

Well, that's my outsider's perspective, and as I say, probably very ignorant.

For once, I totally agree... that was pretty ignorant.

I am not a great example of online gamer (having a PS3 and a slow connection in a 3rd world country doesn't help), but to say fighting games are not complex and therefore could be replaced with fairly straightforward AI is pretty simplistic, and completely inaccurate. Fighting games (also FPS and many other genres too) have dozens of variables and are full of responses to react to them, making decisions on the fly to be organic. Simply put, you can make an AI that presses random buttons (the equivalent or a button-mashing rookie) or an AI that has the best response to every move, but that doesn't provide a meaningful experience.

A perfectly informed AI opponent is not challenging because its good, its challenging because its cheap; and cheap as it is, once you discover his pattern, it can be beaten easily. I have beaten Shao Khan in the hardest difficulty because once you find a move or a sequence that falls out of its programming, you only need to execute it, with minimum skill required. That didn't make me better... Still 4 out of 5 online fights ended with my defeat. MvsC3 tried to move in that direction by having AIs based on famous players to compete with, but those didn't work either... Why? Because decent human players (not even experts) don't stick to a strategy if they see it doesn't work. Human players learn, adapt and constantly try new things. Things that might be suboptimal to the eyes of an expert, but because of that are not easily predictable.

As you said, a middling AI is the hardest challenge for a developer; but middling players constitute 95% of the population of any game, so you better have some decent middling AI programmed. The beauty of it is that, while rookies are similar to each other (based mostly on random button-mashing) and experts are similar to each other (based on acquired, common knowledge), most of us are all middling in different ways. That makes human-to-human competition much more interesting. But, if you play enough with the same couple people, you tend to homogenize the field. You become aware of their favorite tactics and moves, and the end result is that you reach a stalemate (a local maximum). Online playing, even against random people, gives you the chance to improve outside of your comfort zone.

 

[gardian06]

This is the WORST article from Yahtzee that I've seen to date. Yahtzee yields that he may be ignorant on the topic, but he is the most ignorant person I've ever heard trying to spew some ridiculous nonsense about something he has no right talking about.

I will never watch or read anything from Yahtzee again because this is about the worst thing I've ever read about video games. This is probably the only thing I've ever read that has actually offended me.

how? why? I see words, but you make no actual point. at least his article attempts to spear a conversation, and actually does make some logical conclusions. his statement about not knowing something was a testament of admitting a want to know, and then he attempts to interject a thought process that is actually what many AI programmers would aspire to create.

I see no merit in your post so far.

 

[Steve the Pocket]

Heck, while we're bringing up conspiracy theories, have you considered that there's an ulterior motive behind the proliferation of online modes (including the patently pointless ones) in the first place? Like giving people an incentive to keep their consoles connected to the internet so that Microsoft and Sony can monitor everything they do and force unwanted system updates on them.

Also, I was disappointed that Yahtzee the king of cynics missed the most obvious reason that people would want to play a fighting game over the internet rather than in person: That a significant portion of gamers have no friends. >:^D

 

[kagecrush]

This is the WORST article from Yahtzee that I've seen to date. Yahtzee yields that he may be ignorant on the topic, but he is the most ignorant person I've ever heard trying to spew some ridiculous nonsense about something he has no right talking about.

I will never watch or read anything from Yahtzee again because this is about the worst thing I've ever read about video games. This is probably the only thing I've ever read that has actually offended me.

how? why? I see words, but you make no actual point. at least his article attempts to spear a conversation, and actually does make some logical conclusions. his statement about not knowing something was a testament of admitting a want to know, and then he attempts to interject a thought process that is actually what many AI programmers would aspire to create.

I see no merit in your post so far.

I apologize, I'm well aware of the fact that I posted no argument. I believe the other posters (FoolKiller and Thujal) made the same points that I would have made, so I did not see the point of me writing them again.

 

[hermes]

In terms of genetic algorithms (because we are talking about AI here), you can't improve if your population is the same handful of people. At most, you will reach a local maximum fairly quickly and stick into it. Expanding the population enables it to move to a different (better) local maximum.

 

[hermes]

Fighting games have strategy. To imply there is none seems to me to be trolling. And I cannot believe I just signed up to this website (and got my password back in a plaintext email no less, a big security no-no), just to try and address this!

In fighting game's there is mix up. As a player is "waking up" (getting off the ground), they have options; block high, low, try to tech a throw, reversal, back dash, etc. All of which are beaten or lose to another matrix of options (but the player still standing up will have the natural advantage). Playing real people online and you have to adapt to their tendencies in order to win (do they like to do this on my wake up? What are they likely to try next and what is the best option against the two options he is most likely to try?)

You can randomise that with a bot, but it will in no way be the same.

And don't get me started on the AI in fighting games. I have never played a good one! They are all ridiculously easy to read and hence, beat! The only difficult AI's to beat are those where they have an unfair advantage, and that is in no way the same as playing another player using a character of equal (but different) strengths and trying to deconstruct their game to a point where you have them completely read and know what they are going to do before they do (one of the best aspects of fighting games, doing something your opponent thinks is in some way "psychic", when they thought I would never see it coming).

I can use a lot more examples than just wake up options, trying to zone and play footsies with the opponent, etc. etc. But, hopefully the above is enough to shed at least some light on why the entire premise of this article is just flat out wrong!

"you're against a complete expert who has memorized the most efficient possible move and counter for every situation"

Again, mix up. There is no right answer for every situation. There are options, and experienced players are more likely to pick options that are more likely to produce more favourable outcomes (like setting up a combo), but there is by no means a "I win" button in fighting games that pro players can constantly press to beat up lesser skilled fighting game players who haven't found that one button yet! There are risks/rewards, and the management of what you use and when (such as super meter, or a reversal dragon punch, which can be hugely punished on the way back down if baited with a block).

OK, seriously, I'm done now! XD

that was actually one of the points that he was trying to make that if the AI is designed in a good enough way that you can actually make it more difficult then a human. take this for example. If I created a fighting game with a pure learning AI assuming a rule set that only allows for 50 possible moves (yes I know this is a small number)then it is possible for the AI to completely learn each possible action the player can do within approximately 10 fights, and by nature win (no this level of learning is not cheating, but a lot of players think it is) though what actually happens in many fighting games is that they will start to dumb down the learning algorithm, or insert a far higher number of possible outcomes/moves like was done with Dead or Alive they actually introduce a good learning AI, but they dumb it down a bit to make it "fare". what you talk about mostly just boils down to call-response. (yes I know the limitations to an actual learning AI though many people don't) though it is completely possible for an AI to out perform a human, but there have many players would call it unfair. you make some valid points, but you do lack some level of understanding on how AI works, and how AI can mimic a player. in fact Yahtzee is talking about introducing an Alice for a fighting AI which by your attack indicates that you do not understand what that is.

I think his attacks hold well. I think the main problem with Yahtzee's argument is that he greatly underestimates the variables involved in a fairly well designed fighting game, with several levels of recursivity; to the point that creating a deterministic expert AI is not feasible.

For any unexperienced player, any fighting game AI is an advanced Alice; but for someone more experienced, most fighting games AI are deterministic (and for an extremely experienced player, most AI are too predictable)... But human players are not that predictable, specially when they are not known. You can fool some people so that a machine passes the test of Turing of fighting games, but that becomes increasingly more difficult the higher the level of the opponent, because he is more aware of the variables. In your analogy, an Alice would be pretty impressive to a 4 years old, since he doesn't have enough mastery of the language to really challenge her, but it would be a lot less impressive to a 12, 15 or 20 years old, because they can construct phrases that she can't respond to. Add to that the fact a fighting game AI needs to answer in the span of miliseconds and it will likely fail the test even against a medium level player.

 

[Canadamus Prime]

If that's true then they're doing a lousy job of satisfying this market of 1.

 

[sonpansatan]

that was actually one of the points that he was trying to make that if the AI is designed in a good enough way that you can actually make it more difficult then a human. take this for example. If I created a fighting game with a pure learning AI assuming a rule set that only allows for 50 possible moves (yes I know this is a small number)then it is possible for the AI to completely learn each possible action the player can do within approximately 10 fights, and by nature win (no this level of learning is not cheating, but a lot of players think it is)

It's a bit more complicated than that, though it depends on the game.

Third Strike? Sure, write a bot that parries and techs everything, win the game.

Other games, things get more complicated than that. If your AI is set to be able to autoblock every attack, there are some moves that deal damage even when blocked, so just blocking forever won't work. If they attack, the opponent can counterattack certain moves, so they's be restricted to only using moves that they know can't be counter-attacked, which can be a small list and limits their offense. Also, there are certain moves that can't be blocked or teched, and they'd have to know all the possible ways the opponent can set up situations where they find themselves on the wrong end of that move.

In addition, the timing of the move, the relative position when the move is initiated, how much Super meter the opponent has when making the move, are all major factors on how to respond. If you're playing a game like Ultimate Marvel vs Capcom 3, you have to factor in the possibility of assists (10,584 possible assist combinations per character) team synergy (19,600 possible teams, with six different ways each team can be ordered) and very creative uses of game mechanics.

Is it possible? Probably, but it's not very easy.


Edit: Here's a more concrete, though extreme, example from UMVC3. Iron Fist has a Super with 2 Frames of Pre-flash. Storm has a Super with 2 Frames of Post-Flash. So if Iron Fist is first and Storm is second, they have two bars of Super Meter, and they do a team super, then then two frames will pass, there will be the super flash, then two frames later Hailstorm will come out. Hailstorm hits almost the entire screen.

Now what does the AI do? It knows that anything it does that leaves it vulnerable for more than 4 frames (1/15 of a second) can potentially be punished, and that involves most of its moves. The only way for it to guarantee its safety is to not take any risks, but without any risks it has no chance of winning. However, if it always takes the "best" move then it becomes predictable, and that's suicide.

 

[srpilha]

... to the point that creating a deterministic expert AI is not feasible.

I totally agree with this. That's the central argument in Yahtzee's text: if everyone playing a game can be fooled by an AI into thinking they're playing against a human, then multiplayer is pointless. *IF* this were true, *THEN* all the rest of his points would seem to stand.
*BUT* it is not; he very clearly overestimated what AIs can do (or what AIs we can reasonably make now). I think it's an honest mistake for someone admittedly uninformed and distant from the topic at hand. As a professional, he should've just avoided it, maybe.

 

[hermes]

... to the point that creating a deterministic expert AI is not feasible.

I totally agree with this. That's the central argument in Yahtzee's text: if everyone playing a game can be fooled by an AI into thinking they're playing against a human, then multiplayer is pointless. *IF* this were true, *THEN* all the rest of his points would seem to stand.
*BUT* it is not; he very clearly overestimated what AIs can do (or what AIs we can reasonably make now). I think it's an honest mistake for someone admittedly uninformed and distant from the topic at hand. As a professional, he should've just avoided it, maybe.

There is also the fact that fighting games are build around competition with friends (even those that live on the other side of the country) and very little attention is given to the single player campaign; so having some PR said: "We removed online support because we focused our resources into making a kick-ass AI. Who wants to compete with other people when the arcade is fairly decent?" would be a dead sentence for the game.

And, as a developer, I know a decent online support is much easier to build than a kick-ass AI.

 

[Kenjitsuka]

Hehehe, funny as always.
You know what though? I've always liked good bots way better than humans that I couldn't slap in the face after a match, in shooters too.

Unreal, Unreal Tournament etc. were just BLASTS to play against the AI, just as much as with 4 bots and a friend on my second PC (yeah, that was long before broadband).
Mind you the play with friends always ended up being just for laughs, just using the Redeemer cam and sniping.

My wish: EVERY game with co-op and online multiplay would offer the exact same modes, maps and thus experience with offline bots. It's easy; there's 99% already AI in extremely cut down singleplayer anyway.
Would make playing for a long time much more appealing, even if no community forms and the DAMN publisher drops the servers after a few months.
But obviously this would make you buy less games and cut down on the money XBOX Live etc. makes.

 

[Fuzzed]

I got all the respect in the world for Yahtzee. He's such an asset to the gaming world it's ridiculous. But I know, and a lot of other people also know, this article is a pile of dung.

 

[Naeras]

AIs in fighting games are horribly hard to make properly, because the result is always one of the following;

1. The AI has one or more inherent flaws that can be easily abused, thus all you need is to abuse that precise weakness
2. The AI uses every move perfectly and is unbeatable. The exception is if there's a parry option, in which case you can usually figure out the pattern of the AI and beat it every time by parrying all its attacks, turning it into scenario 1.
3. The AI reads your inputs. This is done way too often, and it's annoying as fuck, as the AI is unbeatable.

You need that human factor to add guessing games and reads to the equation.

 

[aegix drakan_v1legacy]

and rather than letting the game screech to a halt every time a player drops out, adds a feature that secretly replaces the lost human player with a bot bearing the same name? Because these days we'd be none the wiser.

I'm sure someone else has said this by now, but Smash Bros Brawl already did this.

The only time I noticed the change was when someone I was totally destroying actually started playing decently. Well...He may have just passed the controller to his older brother. That's still a possibility.

Also, I'm all for this kind of thing.

I hate random disconnects, and I hate my match being interrupted. If the game can swap in a bot right on the spot and I don't even know, that's fine by me.

 

[Shia-Neko-Chan]

I couldn't finish this article. Here I was, ready for more of Yahtzee's usual thoughtful insight, but this article was just so wrong on so many levels.

 

[Covarr]

I almost exclusively play online multiplayer games. Mostly Left 4 Dead 2, Counter-Strike: Global Offensive, and Natural Selection 2, as of late. Here's a few reasons for online multiplayer that this article fails to consider:

1. I play with friends. My best friend lives about sixty miles away from me, and it's just not feasible to play local multiplayer together on a daily basis. Online alleviates this.

2. I play with voice chat. Whether with friends or complete strangers, voice chat makes for a completely different experience than playing with bots, especially in these games where communication is so important to success. But even if we didn't need it to win, it's still more fun to joke around.

3. I play on PC, for the most part. Ever try to local multiplayer on PC? It ain't pretty. Depending on the game, players may get stuck with unequal controls (gamepad vs mouse+keyboard) or cramped together on a small keyboard You Don't Know Jack-style. (As an aside, PC + voice chat is a lot more fun than 360 + voice chat, because there's so much less of the 10-14 crowd).

4. Split-screen kills framerates. Ever try local multiplayer on the 360 version of Left 4 Dead 2? It's near unplayable, it's so choppy. A lot of games have this problem, even if not to the same extent. Sometimes, even if local multiplayer is an option, it's still preferable to play online just to avoid this.

Now, that's not to say that online multiplayer is always the best choice. For games like Worms and Mario Kart, half the fun is being right there with your opponent(s). But to just say "it's pointless" because it happens not to be the best in a few hand-picked scenarios is lunacy.

P.S. Thanks