• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Dota 2 |OT13| 6.86, our Pit Lord and savior

Status
Not open for further replies.
S

Shuggananas

Banned
yarden24 said:
eh, we are still so far away from anything close to AI though, this is the same as what deep blue did for chess, except now with stronger computers we can brute force Go as well.
Click to expand...

they didn't brute force though, they just learned to play to win
It's like if somebody made a perfect meepo or invoker bot and farmed 7k mmr games
 
H

Hylian7

Member
Shuggananas said:
they didn't brute force though, they just learned to play to win
It's like if somebody made a perfect meepo or invoker bot and farmed 7k mmr games
Click to expand...
That would imply they would have to make any bot (other than Viper) play well.

The "Unfair" bots are still a joke. Even Viper bot is garbage after the laning phase.
 
S

Shuggananas

Banned
Unfair bots are shit they just cheat with frame perfect last hitting/denying/spell reactions, reason why whenever I play bots (twice a year ?) I don't bother doing higher than medium difficulty, it adds nothing pretty much
 
Y

yarden24

Member
Shuggananas said:
they didn't brute force though, they just learned to play to win
It's like if somebody made a perfect meepo or invoker bot and farmed 7k mmr games
Click to expand...

it is brute force in a way though, im sure its a neural network algorithm where they just fed it millions of moves where it found the ones with the highest possibility of working, its not thinking, its just using past info to wager on the best outcome, and while that is certainly something humans do as well, its not intelligence, just experience
 
N

NaDannMaGoGo

Member
Just lost my 12 games win streak on phoenix. This is the game.



Emoticon_facepalm.gif
Emoticon_facepalm.gif
Emoticon_facepalm.gif


Of course with the classic FV and NP dual jungle against BH.
 
K

kionedrik

Member
yarden24 said:
it is brute force in a way though, im sure its a neural network algorithm where they just fed it hundred of millions of moves where it found the ones with the highest possibility of working, its not thinking, its just using past info to wager on the best outcome, and while that is certainly something humans do as well, its not intelligence, just experience
Click to expand...

What you described was CBR AI. It is the most simplistic AI form but considered AI nonetheless and it has been around for 30 years.

I can't watch the video right now but I doubt that's the algorithm that machine is using. My bet would be some sort of neuro-fuzzy hybrid with 19x19^3 input neurons, which is a very demanding and complex algorithm.
 
Y

yarden24

Member
kionedrik said:
What you described was CBR AI. It is the most simplistic AI form but considered AI nonetheless and it has been around for 30 years.

I can't watch the video right now but I doubt that's the algorithm that machine is using. My bet would be some sort of neuro-fuzzy hybrid with 19x19^3 input neurons, which is a very demanding and complex algorithm.
Click to expand...


yah, I dunno, im not saying its not complex, but as far as i know every neural network works because of the massive amount of data pored in, which while obviously difficult to do well, still works because of the sheer volume of data rather then anything else.


am i wrong?
 
K

kionedrik

Member
yarden24 said:
yah, I dunno, im not saying its not complex, but as far as i know every neural network works because of the massive amount of data pored in, which while obviously difficult to do well, still works because of the sheer volume of data rather then anything else.


am i wrong?
Click to expand...

Your reasoning is only faulty because they don't work by sheer volume of data pored in them but rather by finding some sort of mathematical correlation between simple variables, within a certain threshold of uncertainty. For instance, in the go game a CBR algorithm would teach the machine all the possible "blobs" and color combinations and it would try to maximize its color in the next move (rinse and repeat every turn), a neuro network would analyse what color is in each square and look for the best free square. In one algorithm you force-feed it cases and outcomes, in the other you train it to adapt the weights of each neuron and synapse.
 
Y

yarden24

Member
kionedrik said:
Your reasoning is only faulty because they don't work by sheer volume of data pored in them but rather by finding some sort of mathematical correlation between simple variables, within a certain threshold of uncertainty. For instance, in the go game a CBR algorithm would teach the machine all the possible "blobs" and color combinations and it would try to maximize its color in the next move (rinse and repeat every turn), a neuro network would analyse what color is in each square and look for the best free square. In one algorithm you force-feed it cases and outcomes, in the other you train it to adapt the weights of each neuron and synapse.
Click to expand...

its the same thing though, the algorithm improves the weights or the variables (whatever it chooses them to be) by volume, and it requires a huge volume of examples to do so.

I mean, would you consider the google search engine to be AI?
 
K

kionedrik

Member
yarden24 said:
its the same thing though, the algorithm improves the weights or the variables (whatever it chooses them to be) by volume, and it requires a huge volume of examples to do so.

I mean, would you consider the google search engine to be AI?
Click to expand...

It's not the same because CBR (Case Based Reasoning) will forever need the control cases to generate an output, which in turn, and only after human validation, will become a part of the control group. That's how it "learns" and that's why it needs an exponentially growing volume of data. Neural Networks' output is a polynomial function, it's math (out1 = k1*x1(t) + k2*x2(t) +... ). In the beginning it needs a lot of data to adapt and find the best values for k1,k2,k3... for each neuron and the correlative values of each synapse, but after that training phase it's not much more demanding than a current gen game.

Google search is not AI because: It doesn't do anything; It needs human validation. It's AA (artificial assistance). Although whatever data it is collecting might be being used to feed an AI system *tinfoil*....


Monochromaticx said:
Team No Diggity (DiG
Click to expand...

Wasn't synd a 3 player?
 
R

Red UFO

Member
put into a match with a 2-stack on each team, the low mmr (about 500 mmr lower than the average) on our two stack instapicks invoker mid and goes 1 for 15

why
 
M

Monochromaticx

Member
if you are bored like me check out

Kaipi vs DiG
singsing-----syndren
bone7 -------zai
cancel--------qojva
CMW --------yapzor
?---------------era

watch_server "[A:1:2190892033:6487]"
 
S

Shuggananas

Banned
Will you please stop being toxic, I'm getting triggered right now

drow maxing aura last
leshrac going brown boots octarine
this will go well

1457648457-epicdrow2.png


epic
 
Wok

Wok

Member
kionedrik said:
It's not the same because CBR (Case Based Reasoning) will forever need the control cases to generate an output, which in turn, and only after human validation, will become a part of the control group. That's how it "learns" and that's why it needs an exponentially growing volume of data. Neural Networks' output is a polynomial function, it's math (out1 = k1*x1(t) + k2*x2(t) +... ). In the beginning it needs a lot of data to adapt and find the best values for k1,k2,k3... for each neuron and the correlative values of each synapse, but after that training phase it's not much more demanding than a current gen game.

Google search is not AI because: It doesn't do anything; It needs human validation. It's AA (artificial assistance). Although whatever data it is collecting might be being used to feed an AI system *tinfoil*....




Wasn't synd a 3 player?
Click to expand...

It is not polynomial though. It is lots of layers of non-linear functions applied to weighted sums of outputs of the previous layer.

Also, as you said, this has nothing to do with feeding the AI with lots of data: the AI is simulating outcomes and evaluating the different policies to make a move. This builds on the benefit of high computing power but is rather close to our way of playing games like Chess or Go.
 
S

Shuggananas

Banned
Shuggananas said:
Will you please stop being toxic, I'm getting triggered right now

drow maxing aura last
leshrac going brown boots octarine
this will go well

1457648457-epicdrow2.png


epic
Click to expand...

somehow managed to win this crap, typical enemy getting overconfident and throwing like shit without getting any objective
sigh
I don't see how lich can stay relevant in the mid/late game without blink tbh, he does absolutely nothing without it
 
Status
Not open for further replies.

Similar threads

Rumor: First Details revealed for Valve's next Game, Deadlock
Rumor 2 3
142
10K
Puscifer replied
The Verge: We played Valve’s secret new shooter: Deadlock
Opinion News 2 3 4
199
11K
near replied
Deadlock new anti-cheat system turns cheaters into frogs
News Community Platform
4
380
darrylgorn replied
Romancing SaGa 2: Revenge of the Seven demo now available
News
25
1K
sigmaZ replied
Deadlock |OT|
|OT| Community 2
73
4K
Robot Carnival replied
Top Bottom