Brawl CPUs Learn?

[SCOTU]

When are people going to learn that you can never prove either side of this argument by saying "well, I've seen the CPU do X; I've done X a bunch", even if you're the ONLY PERSON IN THE WORLD THAT DOES X. Speculative observation does not count as proof. If it did, we'd already have proven that Extra-Terrestrials frequently visit our planet. There are only a few people in this thread who have even put a half second's worth of thought into even trying to deal with this debate in a reasonable manner. Step it up guys. I don't want to see another post about seeing their computer doing something and relating that to what they do. Evidence DOES include CPUs being able to do something or not, not situational evidence being compared against your play.

For instance, a VALID piece of evidence:
If CPUs can store input to "learn", and a DAC is a valid input that gets used, why can't they DAC?

an INVALID piece of evidence:
I DAC all the time, if a CPU learned, it would pick up DACing easily, but doesn't.

Please try to see the difference in these two claims. Note how the one is Valid and the other isn't and WHY. Please only use valid arguments. If you're just giving a bunch of things that don't actually contribute to the discussion, you're just spamming. Step it up.

 

[Oshtoby]

Well, I did the test that I came up with before. I erased my game file after backing it up, and I made Mario and Pikachu fight, both being level 9. The settings were Two life, Final Destination, no Items. Then, after recording those three, I took control of Pikachu for 10 matches, and recorded the replay of them all. So, it was me as Pikachu, versus Mario level 9, on Final destination, two lives each, and no items. Here are the results.

PRE-TRAINING
Match 1: Mario wins, by a wide margin.
Match 2: Mario wins, by a wide margin
Match 3: Pikachu wins, by a slight margin.

10 matches and a soft-reset of the system later.

POST-TRAINING
Match 1: Pikachu wins, by a wide margin.
Match 2: Pikachu wins, by a wide margin.
Match 3: Pikachu wins, by a VERY wide margin.

As you can see by the results, after I recorded the replays of 10 matches of me playing as Pikachu, Pikachu becomes a destructive force. This is not definitive proof, but it is something to ponder. Why did Pikachu, all of a sudden, know how to beat Mario? Why was it a struggle for the little mouse before, whereas afterward it was a breeze?

I took video of the 6 CPU-only matches, and I will post them at my leisure. I have to convert them to .AVI first, then upload them to Youtube.

 

[Dominic]

lol ever heard of too small sample size there there oshtoby ?

i am not sure how one is supose to believe all that learning crap. Or maybe i just think that way because my CPU still suck...

that video of some human snake geting beat by cpu link ? no ****, that snake suck.

Plese someone show me a video of CPU falco or DDD chain grabing, who cares about CPU randomly presing neutral B with Falcon 3 times in a row

 

[Adetque]

lol ever heard of too small sample size there there oshtoby ?

i am not sure how one is supose to believe all that learning crap. Or maybe i just think that way because my CPU still suck...

that video of some human snake geting beat by cpu link ? no ****, that snake suck.

Plese someone show me a video of CPU falco or DDD chain grabing, who cares about CPU randomly presing neutral B with Falcon 3 times in a row

First of all, that person was an above average Snake.
Second, the Snake still won.

And, I've had a cpu Lucario chain grab me, after I did it on cpus.

 

[Alus]

I think that the CPUs just have priority levels... if you hit them with the same move like 5 times they will use one of their pre programed moves to avoid it...i dont think that they record inputs.

 

[ibd]

Well, I did the test that I came up with before. I erased my game file after backing it up, and I made Mario and Pikachu fight, both being level 9. The settings were Two life, Final Destination, no Items. Then, after recording those three, I took control of Pikachu for 10 matches, and recorded the replay of them all. So, it was me as Pikachu, versus Mario level 9, on Final destination, two lives each, and no items. Here are the results.

This is an excellent test, but you played too few matches to really come to a conclusion. You should try to at least play 20-30 matches before and after training.

Still, if a developer already stated that the computer AIs learn, why are we debating over it?

 

[bobson]

a developer already stated that the computer AIs learn

I'm gonna need a source on that.

stuff

This isn't going anywhere. You're saying "AIs cannot learn as per this evidence, end of discussion." We're saying "AIs can learn as per this evidence, end of discussion."

Oshtogirl's tests have far too small a sample size to be considered anything conclusive, so I'm going to conduct some of my own.

 

[Levitas]

The difference between the research I proposed (checking to make sure there aren't null values on "blank" saves and verifying that replays do not grow in size ever) and the research you proposed (many trials on usage of moves) is that mine is really easy to test and yours is really tedious. If you feel like doing this research, do it right and record your exact research logs so that we know exactly what's going on. Also, use a control group.

While your methods have indicated trends, mine has been logical deduction. If my argument cannot be disproven, it doesn't matter what you say, the conclusions I make are true and valid. Note that the testing I propose is to disprove my claim, so if you want to cut down that argument, be conclusive and test that.

 

[SCOTU]

Still, if a developer already stated that the computer AIs learn, why are we debating over it?

because one of the developers (Sakurai) said that brawl's not a fighting game, but rather a sports game.

I don't trust a flying **** about what the developers say.

 

[Oshtoby]

Oshtoboy

Why does everyone call me that?

 

[KO M]

lol ever heard of too small sample size there there oshtoby ?

i am not sure how one is supose to believe all that learning crap. Or maybe i just think that way because my CPU still suck...

Maybe because you do? o_o

 

[bobson]

The difference between the research I proposed (checking to make sure there aren't null values on "blank" saves and verifying that replays do not grow in size ever) and the research you proposed (many trials on usage of moves) is that mine is really easy to test and yours is really tedious. If you feel like doing this research, do it right and record your exact research logs so that we know exactly what's going on. Also, use a control group.

If I prove that replays store null values or that replays grow in size, I'll have accomplished little more than disproving your claim. There still won't be any conclusive evidence of the AI learning or not, just that replays grow/store nulls. The other way will take longer, but will shed much more light on the issue.

Why does everyone call me that?

¯\(º_o)/¯

 

[RyanPF]

because one of the developers (Sakurai) said that brawl's not a fighting game, but rather a sports game.

I don't trust a flying **** about what the developers say.

Misquote. http://www.smashbros.com/en_us/howto/basic/basic01.htm
I think he meant that it's like a sport to the characters in the game. (Hence the trophies, crowds, etc.)

If they said the AI "learns" (I'm not saying they have, but I'd like to know), then they must have put in some functionality for it to "learn." You can't tell the developers that they don't know anything about the game they developed.

 

[Frown]

I'm just wondering... has any CPU DDD ever downgrabbed you twice?

 

[KO M]

The difference between the research I proposed (checking to make sure there aren't null values on "blank" saves and verifying that replays do not grow in size ever) and the research you proposed (many trials on usage of moves) is that mine is really easy to test and yours is really tedious. If you feel like doing this research, do it right and record your exact research logs so that we know exactly what's going on. Also, use a control group.

While your methods have indicated trends, mine has been logical deduction. If my argument cannot be disproven, it doesn't matter what you say, the conclusions I make are true and valid. Note that the testing I propose is to disprove my claim, so if you want to cut down that argument, be conclusive and test that.

Honestly I would ignore Levitas if I were you. Just like the characters were true to their games thread,
Levitas is just hell bent on believing he is right, no matter what anyone says. He tries to act smart, but really
your opinions are not the only ones that matter.


As for the CPU learns. I said before they learn in a DER DUR DUR way, meaning it is not really learning on their own. It just taking data from us players and compiling that into who knows what.

Ok I think we can safely say, CPUS can learn and do learn in a way.
But they cannot learn on there own.

By fighting humans , they can copy behavior, a build there own style of play.. but they are still basically choosing a random style.

This is what happend when I tested this.

Fight #1 I used Marth , CPU is Falco lvl 9/ The Falco was very agressive and always was staying on my ***, tryed spking me a couple of times, it was very odd. When I got an early kill on it, it was almost as if.
It was thinking "How could that happen" Because it stayed there for about 10 seconds not doing anything.(Could also be it was searching its data banks for a counter messure).
Basic line is 1st fight Falco was agressive.

#2 Sheik, CPU Falco/ The Falco this was very campy... After I ledge gimped it... It started PLANKING
Kept jumping from the edge to double lazer me. I went over to stop it, and it stage spiked me.
Basic line-Falco was planking, very campy


#3 I used Snake, CPU Falco/ He started off by chain grabbing me ....( he saw I was Snake... it knows matchups) Spiked me , 1st stock goes to CPU. The 2nd Stock I kept him at bay with grenades and kept him off the stage. When I planted on mine, and threw him into it. he DI'd out of the way came back and SH lazer'd me and came in witha boost smash /UP smash) Now the Falco is still 3 stocks 180%
I get him with a an uptilt and he loses his stock.He Starts now wave bouncing like crazy, and SH DL'ing
I gets me to 30% grabs me and starts chain grabbing me to the Spike / 2 Stocking me
Basic Line- He played like a good Falco/ A Human

#4 I used ROB, CPU Falco /I have never played a ROB with my Falco on this Wii, So the match starts he really isnt sure what to do. Never had this kind of match up. I 3 stocked him didnt get hit once.
I camped and ledge stalled him. He tryed SH Dl and he even once tryed a boost smash. When I went under the stage. He tryed following to maybe stage spike me(Iv done that alot to Pits when they go under)
Basic Line- Was like a player who was not sure of the match up.

CPUS can learn , it justs not really our kind of learning. It is basically taking data from us, and using it to see if it will pay off.
If im Falco alot and Marth, and I do alot of match ups with Snake and win those. The CPU version will know the match up and beat the hell out of a Snake too.

As I stated, CPUs are just looking at the character data of what we are doing , just a bunch of 0s and 1s
directing we are doing. CPUs are like infants and will copy , nothing more.

Really end this discussion now, they do not learn on their own, they are just infants that copy what we do.

 

[bobson]

Really end this discussion now, they do not learn on their own, they are just infants that copy what we do.

The discussion is whether or not they "learn" based on what the players do.

I don't think anyone in the thread is suggesting that they have artificial brains which secretly train in the background while we're not fighting them.

 

[Oshtoby]

they are just infants that copy what we do.

That's... kind of what "learning to use our fighting style" is.

 

[Levitas]

Honestly I would ignore Levitas if I were you. Just like the characters were true to their games thread,
Levitas is just hell bent on believing he is right, no matter what anyone says. He tries to act smart, but really
your opinions are not the only ones that matter.

I don't care what you do. I only listen to arguments that make sense. As such, I'm done here. Prolonging this doesn't mean anything.

You're mistaking me for one of the 3 or so other people on that thread that actually wouldn't listen to arguments. I responded to everything that addressed me there.

 

[Excellence]

No. I wrote up a pretty decent argument for why they don't:
http://allisbrawl.com/forum/topic.aspx?pid=569180#p569180

Is your avatar of Lucy from Elfin Lied?

 

[RyanPF]

Hmm. I think SCOTU is right. Perhaps CPUs just change playstyle from time to time like someone previously said.

 

[t3h n00b]

Lol I got Falcon Punched and dtaunted. The CPU copied my best move >_>

 

[SCOTU]

Is your avatar of Lucy from Elfin Lied?

Izayoi Sakuya from Touhou.

 

[XienZo]

I don't care what you do. I only listen to arguments that make sense. As such, I'm done here. Prolonging this doesn't mean anything.

You're mistaking me for one of the 3 or so other people on that thread that actually wouldn't listen to arguments. I responded to everything that addressed me there.

You missed me...

I know fully well you "can". But no one has conclusive scientific-method-worthy process of proving or not. See, the hard part about this is that you have to remove ALL OTHER VARIABLES, regardless whether you're proving or disproving it.



That my CPU plays more often? Its more of "your replays should be bigger for a same length replay on a character that the CPU uses seemingly "learned" behavior than one where it doesn't." Simply "plays often" is too general to cover only "learned" behavior. You can't just play a CPU Falco 20 times and test it. Your Falco needs to be doing something we would claim to be "learned", or you're not testing anything.

Anyhow, we have no actual evidence as of now, so making arguments won't do much, regardless of whether people listen to them or not. Someone needs to go out and get the data. At the very least, the replay-file would be nice.

 

[sasook]

They do learn.

Proof: http://www.youtube.com/watch?v=pXNQo71shCQ

 

[noradseven]

The learn I don't think the level 9 computers came out arrow/laser canceling non stop. Or using pikachu's up-B cancel all the time. If you don't believe it get a fresh copy of brawl play its Toon Link/ falco, play 20 rounds as TL or Falco only laser/arrow canceling. Play that TL or Falco again watch the computer laser/arrow cancel all the time.

 

[DVDA]

I'm not quick to believe in something as amazing as the CPUs actually learning, but i remember something from the early days of brawl that makes me think they do. I made a stage with two main platforms(like 2 small final destinations in one) I was fighting the CPU luigi with samus I think and he would keep trying to jump over to my platform and I'd keep hitting him away. Then after about the 5 time he would double jump back to the other platform and he consistently did it to avoid the same edgeguard i was doing. it definitely looked like he started picking up on my edgeguard. Also I've been playing brawl+ recently and i notice that the CPUs actually edgeguard you in a way that would only work without auto sweet spotting. They would never stand near the edge and just forward smash when i used to play them before, but they have been recently. I dunno. It does seem like alot to program for just a game and that's why I'm kinda skeptical still, but it really seems like they do learn. either way though weather they learn or not, they do force you to play smarter than melee CPUs and that helps.

Oh Yeah I got alot of vids saved that i saved just because i thought the CPUs were smart or learned. i'll upload them to my youtube account tonight hopefully. i got some good stuff like obvious edgehog mindgames that I'd use on them being done to me lol.

 

[PsytricSword]

Ummm, I saw a squirtle shell shift >>Usmash. I never did that before. I don't even know how to shell shift @_@. I don't use squirtle...That happend earlier today on my Brawl. This brings me to some other topic: Can CPUs learn from wifi? Friends wifi? Basic/team brall wifi? Because I played a friend online who shell shifts.

 

[t3h n00b]

Ummm, I saw a squirtle shell shift >>Usmash. I never did that before. I don't even know how to shell shift @_@. I don't use squirtle...That happend earlier today on my Brawl. This brings me to some other topic: Can CPUs learn from wifi? Friends wifi? Basic/team brall wifi? Because I played a friend online who shell shifts.

I don't know why the person before you posted here now... but I guess this thread is active again? The answer to your question is yes. CPU Falcos chaingrab, Pikachus QAC, and Snakes mortar slide if it is done enough in games. My Captain Falcon even jab cancels on me. I'm not sure exactly how the AI works, but there are many opinions(/facts?) in other pages of this thread. CPUs can learn from wifi, and regular offline brawls as well.

 

[PhatyCHONG]

I can DISPROVE it all
PHATY:
- I play DIDDY ALOT for the past year
- I use ALOT of Bananas
- I use ALOT of peanut

CPU:
- Plays Diddy
- SIGNIFICANT less Bananas [there will be times will he won't use it at all!]
- less peanuts action

Case CLOSED

 

[t3h n00b]

I can DISPROVE it all
PHATY:
- I play DIDDY ALOT for the past year
- I use ALOT of Bananas
- I use ALOT of peanut

CPU:
- Plays Diddy
- SIGNIFICANT less Bananas [there will be times will he won't use it at all!]
- less peanuts action

Case CLOSED

Words aren't proof. Below is proof. Also, CPUs don't learn exactly. A better word for me to have used would be imitate. Things that don't involve a lot of thinking or time, but have specific inputs (SHDL/SHDA, QAC, jab cancels) are not only often inputted more frequently and consistently by a player than arbitrarily using peanuts and bananas, they are easier to mimic. I use Diddy. Bananas are very dynamic. Glide toss right>dashgrab>fthrow is nowhere near as common as SHDL. There are many different inputs you can do for banana combos, whereas in the things I mentioned, there are only a few inputs. These things are more easily replicated. Play as a character you use least and spam their neutral B move, no matter what it is, for 5 matches straight, if you want to see what I mean. Then play a CPU [whatever character you spammed B with] and they will use their neutral B move a lot of the time, much more than anyone normally would.

They do learn.

Proof: http://www.youtube.com/watch?v=pXNQo71shCQ

 

[Likeness101]

My CPU glide tossed before, my DDD sometimes does 2 or 3 straight down throws before using another resorting to another throw and he does it on walls sometimes also. Also my ike jab cancels to down smash sometimes cause I do it a lot.

http://www.youtube.com/watch?v=yybogS2U_8o - CPU Glide tossing, just bringing it up i guess.

 

[Dual Kirby]

hmmm i think they can adapt to your fighting style.... my kirby cpu can gonzo combo and my mk is now almost unbeatable i cant think of an explination maybe its the cpus or maybe im just going crazy

 

[Lythium]

A friend of mine was beating up sandbags with Pikachu in attempt to get all the music tracks. Unfortunately, all he used was D-smash. He fought a Pikachu CPU, and it D-smashed for the entire match. It was rather amusing, to say the least.

 

[Ha2On]

well, i think it could be possible, because of the fact, that my main chars are much harder to beat than other charackters(if they are cpus^^)

but, i cant tell if it is true

 

[OutlawStar]

I think they react to what you do, but I think they do learn things,
they forget them after the match.

 

[JOE!]

hah! my bowser klaw-hopped around Smashville for a few seconds today

 

[Orange_Soda_Man]

I've had several cases where a kirby has "crouch-taunted" me after I've died. you know, the wakawaka sound. I do it whenever I don't want to lose a power.

I've seen the CPU do it with I think game&watch too, though idk why it did.

 

[SCOTU]

When are people going to learn that you can never prove either side of this argument by saying "well, I've seen the CPU do X; I've done X a bunch", even if you're the ONLY PERSON IN THE WORLD THAT DOES X. Speculative observation does not count as proof. If it did, we'd already have proven that Extra-Terrestrials frequently visit our planet. There are only a few people in this thread who have even put a half second's worth of thought into even trying to deal with this debate in a reasonable manner. Step it up guys. I don't want to see another post about seeing their computer doing something and relating that to what they do. Evidence DOES include CPUs being able to do something or not, not situational evidence being compared against your play.

For instance, a VALID piece of evidence:
If CPUs can store input to "learn", and a DAC is a valid input that gets used, why can't they DAC?

an INVALID piece of evidence:
I DAC all the time, if a CPU learned, it would pick up DACing easily, but doesn't.

Please try to see the difference in these two claims. Note how the one is Valid and the other isn't and WHY. Please only use valid arguments. If you're just giving a bunch of things that don't actually contribute to the discussion, you're just spamming. Step it up.

To the people who've bumped this thread and are now starting new discussion on the matter, please read this quote.

 

[volbound1700]

If they do, then they do a poor job of it. CPUS are too easy to figure out and beat.

 

[jahkzheng]

CPUs have difficulty in Brawl+ since the physics they're programed for have been altered. Ex: failed recoveries.
In my experience there is no evidence of CPUs being able to adjust. There's my two cents...