What ever happened to AI/Bots in multiplayer games? Also why no spectator modes?

[CCountZero]

Can't say for certain, but Spectator always seemed to have massive potential for abuse.

8v8 match, one team TKs one guy at round start, after which said guy can relay EVERYTHING to his mates via out-of-game comms.

Maybe that has something to do with it? Just a guess.

 

[shrekfan246]

That's one of the reasons I was so into Super Smash Bros. Melee, actually. I liked the feeling of working my way up through the AI difficulty levels and all of the special event matches, and every now and then when I just wanted to sit back and relax it was pretty fun to just set four random AI characters on each other and watch how it played out.

Luigi almost always won.

I think bots have fallen by the wayside because the single-player has started being considered the "vs. Bots" mode or something, and multi-player is supposed to be with multiple people. I think Killzone had a bots mode.

 

[josemlopes]

Fair enough, but for games like Battlefield 3 and Call of Duty I can't possibly see a problem. They already have AI in the singleplayer part of the game! Surely it wouldn't damand that much effort to enable bots on the multiplayer maps too

I'm not so sure. The AI in these SP games is actually not only utterly retarded, it's literally programmed to be like that. If you watch the behavior of the enemies, most of the time they only stick to cover and pop up their heads from time to time. And they need to either stay on the same place, or run directly towards you so you'd have an easy time shooting them with a console controller. Also note they don't ever need to go very far from their starting point (you either shoot them quickly or they stay where they are) so there's no complicated wayfinding in place. It's a far, far faaaaaaaaaaaaaar cry from true bots like in Quake 3 or Unreal Tournament. In fact, it's pretty far even from decent monster AI. (Monster is an old-school mapping/AI term for enemies. E.g. in Half-Life, the Apache helicopter is called monster_apache.)

Of course it kinda begs the question why don't hey just use altered AI from those older games (e.g. COD series is based on Quake 3 engine, so maybe something could be reused), but it's probably not so simple due to more complex map geometry, physics etc.

Ages ago when I was mapping for Quake 3, it actually took some effort to make the map bot-compatible, even though the bot waypoints were created automatically during the map-building process. There was still a a lot of stuff to tweak, otherwise the bots felt very robotic. I can imagine it can be a nightmare to make a massive Battlefield-like map good for bots. Or even a COD map with loads of rocks and debris lying around. Also, there's stuff like vegetation, or small holes - stuff that you can see through, but not very well. Probably not easy to program this into bot behavior. Back in the day, maps were just blocks of geometry and stuff like this was easier.

I think it's not a coincidence that the only current game with bots is CS:GO - Source engine is still very much based on 90's era tech, maps are simple and cut into chunks by default so it's easier to make bot waypoints. In fact I wouldn't be surprised if CS:GO uses pretty much the same bot tech as the old pre-1.0 CS. Yea I remember making waypoints for that too... Not fun.

I'm sure this stuff can be solved - there are some really nice AI middlewares available, but it's obviously not a priority for the developers, since everyone just wants user to stick with multiplayer (and preferably pay for it).

Like it was already said, COD BLOPS 2 has bots, and they arent bad, average to say the least.
There is also the Gears Of War series although those games have some very defined and clean map layouts without much stuff in the way.

 

[Schmeiser]

DOTA 2 has all this, you can spectate any game you want and i mean ANY game that is currently going on. Plus you can watch PRO games with you moving the camera and the over the map and tune in to any caster aswell. You have bots ranging from passive to unfair, granted they are all pretty easy for a seasoned player but they will give you a challenger while you learn.

 

[rob_simple]

I lost interest in all multiplayer games as soon as they started dropping bots/local multiplayer.

No hyper-realistic military wanking simulator will ever beat the excitement of playing Timesplitters 3 with a friend and 14 bots, one-hit-kills and only bricks as a weapon.

 

[Risingblade]

LoL has this, some of the bots will even mess you up better than real players

 

[The Enquirer]

I loved playing multiplayer with the bots in Perfect Dark.
So very customizable with the relatively large amount of different bots, you can create any kind of hectic hell your little mind can come up with.

Or just leave the 64 on all night and farm AI kills in the Facility.
*snickers*

god I miss perfect dark. It was an amazing sequel to Goldeneye.

I think timesplitters would be worth mentioning here as well. Less programable, but fun and perfect for added cannon fodder during a multiplayer match.

I like bots, but only when fairly spread out. A team of people v bots is cheap, but a team of half and half can be delightful fun.

Actually in timesplitters you can in fact "program" the bots using glitches. This concept has been mastered by AdmiralHowdy. http://www.youtube.com/watch?v=F8ySwdjYUyo (actually inspired by a multiplayer map I made.

 

[CrystalShadow]

Ah... Bots. Did you know many of them were designed with 'artificial stupidity'?

No, I'm not even kidding. A developer of Unreal Tournament spent some time explaining.

The trick is, creating a bot that can defeat humans is easy. very easy.

You might not think so, but it's true.

What's difficult to do, is to create one that is flawed enough so a human stands some chance of winning, but not so flawed that they're obviously doing stupid things.

Another major point is that you want the mistakes the bots do make to be 'human'. That is, if something isn't being done perfectly, a good bot will do it badly in the same way that a human being will.

Again, an explicit example from Unreal Tournament involves aiming;

Humans rarely have perfect aim.
But... For a bot, coding it to always hit is trivial.
If you want it to miss sometimes, there are several different ways you can do this, but some are much more obviously fake than others.

A human that misses actually tends to miss in a predictable manner though; So... You can make your bot more human-like in it's aim by paying attention to what humans do wrong when they miss.
And usually, the most common mistake is not leading your shot properly. - Which means you've just fired at where your target used to be, rather than where they are now.

But that's just one example of many.

A good bot is not one that plays perfectly, but rather one that fails in a way that is as human-like as possible.

And that is pretty hard to get right, because all the odd behaviours that break this stand out like a sore thumb, yet actually figuring out how humans behave in practice is incredibly difficult, and requires some serious observational skills.

Then again, to an AI researcher, the 'AI' in most games is a sad joke. And keeping in mind how limited the state of the art in AI research is, that tells you something about just how primitive game AI must be...

I guess bots just aren't a priority for modern game developers though...

 

[Roxas1359]

Just play TF2 man. it has it all, even hats.

It also has Spy Crabs, but sadly they are endangered due to Pyros and F2P people killing them.
Remember, kill a Spy Crab, get 7 years bad crate luck.

OT: I have to agree with you that Bots in multiplayer really are a rarity nowadays. I know that Rainbow Six Vegas 2 has some in the multiplayer if you do online Terrorist Hunt and I had fun with that myself. Other than that TF2 has online bots.

 

[Souplex]

It's a lost relic of local multiplayer.
I miss local multiplayer.

 

[Pink Gregory]

Well, consoles happened. I guess bots never were a big thing there.

There were plenty of games from the Xbox/Ps2/Gamecube generation that had bots; it was pretty much Halo that was the first big title not to have them, to my memory.

It hurts me to do this, SNIP

...uh, yes. What they said.

Considering the availability of relatively stable multiplayer, and a fairly reliable international playerbase, I can understand why bots might not be a priority. Then again, this is reliant on the game's multiplayer not becoming a dusty wasteland after a year.

 

[silver wolf009]

You mean like Black Ops 1 and presumably 2, which had both of these things?

 

[Voltano]

Programming good AI is tough, and since the selling point of these games is the online multi-player, doing good AI has been put aside, possibly.

As for the spectator mode: It could be to detract cheaters. A friend who is spectating could give intel to another friend about the location of enemies or power ups. This could imbalance the game for everyone involved.

 

[Owyn_Merrilin]

Programming good AI is tough, and since the selling point of these games is the online multi-player, doing good AI has been put aside, possibly.

As for the spectator mode: It could be to detract cheaters. A friend who is spectating could give intel to another friend about the location of enemies or power ups. This could imbalance the game for everyone involved.

Funnily enough, spectating is one of the things you're generally supposed to do if you think someone else is cheating -- pop into spectator mode for a while and follow them around, see if they're doing anything that shouldn't be possible, like walking through walls or something. But then again, that's a holdover from before voice chat was common, so using spectate to cheat wasn't exactly something you could easily do.

Ah... Bots. Did you know many of them were designed with 'artificial stupidity'?

No, I'm not even kidding. A developer of Unreal Tournament spent some time explaining.

The trick is, creating a bot that can defeat humans is easy. very easy.

You might not think so, but it's true.

What's difficult to do, is to create one that is flawed enough so a human stands some chance of winning, but not so flawed that they're obviously doing stupid things.

Another major point is that you want the mistakes the bots do make to be 'human'. That is, if something isn't being done perfectly, a good bot will do it badly in the same way that a human being will.

Again, an explicit example from Unreal Tournament involves aiming;

Humans rarely have perfect aim.
But... For a bot, coding it to always hit is trivial.
If you want it to miss sometimes, there are several different ways you can do this, but some are much more obviously fake than others.

A human that misses actually tends to miss in a predictable manner though; So... You can make your bot more human-like in it's aim by paying attention to what humans do wrong when they miss.
And usually, the most common mistake is not leading your shot properly. - Which means you've just fired at where your target used to be, rather than where they are now.

But that's just one example of many.

A good bot is not one that plays perfectly, but rather one that fails in a way that is as human-like as possible.

And that is pretty hard to get right, because all the odd behaviours that break this stand out like a sore thumb, yet actually figuring out how humans behave in practice is incredibly difficult, and requires some serious observational skills.

Then again, to an AI researcher, the 'AI' in most games is a sad joke. And keeping in mind how limited the state of the art in AI research is, that tells you something about just how primitive game AI must be...

I guess bots just aren't a priority for modern game developers though...

The problem is, it's really easy to develop a bot that has perfect aim, or for fighting games, that reads your inputs and immediately uses the perfect counter. However, it's pretty much impossible to make the bots creative enough to devise their own strategies, which means they only know the strategies the programmer gave them. This almost always leaves holes in the way they act that a good enough player will be able to effortlessly exploit. Once you're at that point, playing against bots actually makes you a /worse/ player, because you get used to using strategies that go through a hole in the AI's programming, rather than strategies that counter other strategies -- and you can guarantee that human players will have a strategy that counters that hole in the AI you found.

Basically, bots have perfect reflexes, but they can't think. Humans in general have bad reflexes and natural weapons, but we're at the top of the food chain because we're so danged smart.

 

[CrystalShadow]

Spoiler: long stuff

Ah... Bots. Did you know many of them were designed with 'artificial stupidity'?

No, I'm not even kidding. A developer of Unreal Tournament spent some time explaining.

The trick is, creating a bot that can defeat humans is easy. very easy.

You might not think so, but it's true.

What's difficult to do, is to create one that is flawed enough so a human stands some chance of winning, but not so flawed that they're obviously doing stupid things.

Another major point is that you want the mistakes the bots do make to be 'human'. That is, if something isn't being done perfectly, a good bot will do it badly in the same way that a human being will.

Again, an explicit example from Unreal Tournament involves aiming;

Humans rarely have perfect aim.
But... For a bot, coding it to always hit is trivial.
If you want it to miss sometimes, there are several different ways you can do this, but some are much more obviously fake than others.

A human that misses actually tends to miss in a predictable manner though; So... You can make your bot more human-like in it's aim by paying attention to what humans do wrong when they miss.
And usually, the most common mistake is not leading your shot properly. - Which means you've just fired at where your target used to be, rather than where they are now.

But that's just one example of many.

A good bot is not one that plays perfectly, but rather one that fails in a way that is as human-like as possible.

And that is pretty hard to get right, because all the odd behaviours that break this stand out like a sore thumb, yet actually figuring out how humans behave in practice is incredibly difficult, and requires some serious observational skills.

Then again, to an AI researcher, the 'AI' in most games is a sad joke. And keeping in mind how limited the state of the art in AI research is, that tells you something about just how primitive game AI must be...

I guess bots just aren't a priority for modern game developers though...

The problem is, it's really easy to develop a bot that has perfect aim, or for fighting games, that reads your inputs and immediately uses the perfect counter. However, it's pretty much impossible to make the bots creative enough to devise their own strategies, which means they only know the strategies the programmer gave them. This almost always leaves holes in the way they act that a good enough player will be able to effortlessly exploit. Once you're at that point, playing against bots actually makes you a /worse/ player, because you get used to using strategies that go through a hole in the AI's programming, rather than strategies that counter other strategies -- and you can guarantee that human players will have a strategy that counters that hole in the AI you found.

Basically, bots have perfect reflexes, but they can't think. Humans in general have bad reflexes and natural weapons, but we're at the top of the food chain because we're so danged smart.

That's true up to a point, but depending on the game, there may only be so many optimal strategies that even exist. Holes in an AI's strategy may exist, but depending on what else the AI is good at, you simply may not even be good enough to exploit them even if you can figure out what the weaknesses actually are.

This has been proven for certain kinds of games (chess for instance - although someone did beat a near invincible AI by going against all basic logic of how you're supposed to play chess...), but many games are pretty complex.

In any event, AI that can learn and adapt does exist in labratory settings. But again, it's in a fairly primitive state, and even that rarely finds it's way into games.

A curious example of just what an AI can learn comes from a game called Stars! Supernova. It was the third game in a series of 4x style space strategy games, but it ended up never being released because the company collapsed.

During development though, they implemented a learning AI, that observed player behaviour, and figured out new strategies.
Unfortunately, at some point during early testing of it, they found out that the AI had too much power; It could observe and mimic behaviours it really shouldn't have been able to.

As a result... It locked up the game in and endless loop.
Why? Because it had learned that an effective strategy to deal with making a mistake while playing a game, is to reload an earlier save game...

I think you can see where that would lead to an endless loop... XD

That AI was clearly to smart for it's own good.

In any event, AI's can certainly be created that can learn and adapt, and create new strategies to counter what a human does.
It's rarely been done in a game though. - As I said, game AI is, on the whole, fairly primitive, and in many cases can barely be considered AI at all relative to the research in the area...

 

[piinyouri]

I loved playing multiplayer with the bots in Perfect Dark.
So very customizable with the relatively large amount of different bots, you can create any kind of hectic hell your little mind can come up with.

Or just leave the 64 on all night and farm AI kills in the Facility.
*snickers*

I loved the multiplayer with the bots in Perfect Dark! My brother and I played that more than any other multiplayer since. You're absolutely right about the customization too. We were able to create enemies that almost seemed to have a personality. I wish we could see another game like that.

An army of turtle slow, but golitah armored bots with rocket launchers?
The suspense kills me just thinking about it.

Does anyone remember the bots in Conkers Bad Fur Day?
Christ those things were eerily accurate on Einstein difficulty.
Allow me to personify "eerie" in this case.
They can hit you in the head from the other side of the map with one bullet while you're in the middle of the katana jump. (When you had a katana weapon, your normal jump turned into a pretty high somersault. They could nail you in the middle of that animation)

 

[Roxor]

In any event, AI that can learn and adapt does exist in labratory settings. But again, it's in a fairly primitive state, and even that rarely finds it's way into games.

A curious example of just what an AI can learn comes from a game called Stars! Supernova. It was the third game in a series of 4x style space strategy games, but it ended up never being released because the company collapsed.

During development though, they implemented a learning AI, that observed player behaviour, and figured out new strategies.
Unfortunately, at some point during early testing of it, they found out that the AI had too much power; It could observe and mimic behaviours it really shouldn't have been able to.

As a result... It locked up the game in and endless loop.
Why? Because it had learned that an effective strategy to deal with making a mistake while playing a game, is to reload an earlier save game...

I think you can see where that would lead to an endless loop... XD

That AI was clearly to smart for it's own good.

In any event, AI's can certainly be created that can learn and adapt, and create new strategies to counter what a human does.
It's rarely been done in a game though. - As I said, game AI is, on the whole, fairly primitive, and in many cases can barely be considered AI at all relative to the research in the area...

Ha ha! That's brilliant! We need more developers to try stuff like that.

 

[The White Hunter]

Well, consoles happened. I guess bots never were a big thing there.

I also guess AI isn't that simple to do and it's not worth the development costs. CS:GO has bots and they feel really unnatural.

Call of Duty has had bots for a while, Killzone has had bots in every game (actually has pretty fiendish AI, the sad thing is that you can't splitscreen the bot mode), CS:GO and TF2 have bots too but you mentioned that.

Bots are a tonne of fun and provide a good place to learn a bit before you dive into competitive.

 

[Colt47]

I kind of miss Unreal Tournament 2004's monster madness. It was like the precursor to Left 4 Dead in some regards, though the only real difference is there's no end to the waves in UT2k4. Anyone know if Serious Sam 3 is comparable?