Self learning AI beats pro Dota2 players

File: 1499298869185.jpg (118KB, 844x590px) Image search: [Google]

118KB, 844x590px

https://youtu.be/wiOopO9jTZw

File: CSwY0fsXIAA50G-.jpg (20KB, 600x338px) Image search: [Google]

20KB, 600x338px

Let's wait how the progress for the Starcraft 2 AI turns out. That's the actual challenge, it's infinitely more complex than midlane 1v1 DotA with SF only. If an AI manages to take down a player like Innovation I'll get scared.

>>387032632
>If an AI manages to take down a player like Innovation I'll get scared.
not if, when.

>>387032632
It will. It has access to all the variables and can click far more accurately than a human can.

Its fucking retarded to really think bots will replace humans in tourney play though. It'd be boring and clinical, there'd be no fuck-ups, styling or anything.

>>387032632
>bot beating Innovation
I'm pretty sure that's possible, but it's just adding in enough "reactions" for the CPU to execute in the program. Look at the hardest difficulty in SC2 now, it's a joke, it barely reacts to anything and the only thing it nails is the economy. I think the entire concept of A.I. is just a meme, right now there is no such thing as a complex learning computer. Just programs that some neckbeard took fucking forever to add in the optimal solutions to problems so it seems "complex and alive".

>>387033001
>right now there is no such thing as a complex learning computer.

Then you know jackshit about machine learning
Since there are infact already AIs that teach themselves complete concepts without any scripted programming, the only thing they get told is the goal they need to achieve and a previously defined input of which data they get to learn from.

>>387033001
>Just programs that some neckbeard took fucking forever to add in the optimal solutions to problems so it seems "complex and alive".

Hello grandpa, 70s called, they want you to fix the mainframe again.
Machine learning is a technology from the 80s, clearly way after your prime.

>>387033187
You are full-on retarded and gay. That shit is and will always only be - MATHEMATICS. Have you ever coded anything? You are ALWAYS asking the computer to compute a solution. See, that's why the are called "computers", did you see what I did there?

>>387033545
You are retarded.

>That shit is and will always only be - MATHEMATICS

Yeah we humans are only mathematics too.
Mathematics can literally do anything that is possible and describe any concept that is possible.

If you dont get that its you who is the dummy.

>>387031958
It was massively beaten by cheesing scrub tier players, nice try OP.

An AI will always be superior in terms of reflexes and repetitive behavior, they already had to excessively limit the people playing through a set of arbitrary rules that gave the bot an advantage too.

It's an interesting concept but it's nowhere near finished, barely a proof of concept at this point.

>>387031958
>out of 100+ heroes pick the 1vs1 matchup where pure mechanic skill is everything because of souls

gee i wonder why the computer won, putting it in a scenario with almost zero variables

File: 1440567013337.png (192KB, 352x375px) Image search: [Google]

192KB, 352x375px

I bet it can't beat League players

>>387033545
People don't want to understand that its just a algorithm that has been programmed to remember other algorithms. Its nowhere near a consciousness which is what most people believe when they think about artificial intelligence. You still have to edit the algorithm what exactly it should learn. You are completely right that its not really impressive from the view that its an artificial consciousness on the other end. Its just a language that a computer develops because it was told to compute the language.

>>387031958
>>387033001
It is possible.
Players just want the illusion of smart AI though and not an actual intelligent AI
>http://askagamedev.tumblr.com/post/76972636953/game-development-myths-players-want-smart

>>387033187
>Since there are infact already AIs that teach themselves complete concepts without any scripted programming
wrong

>>387033642
>Mathematics can literally do anything that is possible and describe any concept that is possible
unproven

>>387031958
>Self-learning
>Got beaten 50 times by attendees. Not even pros, just literally who attendees.
Nice bait you fucking faggot

>>387034005
>not an argument
>not an argument
nice (you) farm you've got going cunt

>>387034041
'self learning' doesnt mean 'good at the game'

>>387034005
>>Mathematics can literally do anything that is possible and describe any concept that is possible

>unproven

What do you believe in the concept of a soul or some magical shit that makes humans better because reasons?

Then i cant keep talking to you either way, since thats maximum dummy tier.

>>387033001
>Just programs that some neckbeard took fucking forever to add in the optimal solutions to problems so it seems "complex and alive".
that is the opposite of general purpose learning AI's you retard. the whole deal here is that the AI didn't know wtf Dota2 was or how it plays before it started learning it by trial and error.

>>387034041
it was a dead end script running with all they had up to that point.
It needs to be plugged into amazons AWS to have the gpu and cpu grunt to learn more.

>>387034073
I don't need to provide an argument for facts, to make the claim that mathematics can describe any concept possible you first need to understand every concept possible, which we don't

Didn't an Amiibo come third in a global Smash bros tournament?

>>387034254
>taking it literally

Well it can atleast describe any concept we as humans understand.

So mathematics can easily represent human tier intelligence or above.

assfaggots is a complete joke and this bot isn't anything special at all, it's piss easy to make a bot that can win for any game but the challenge for AI programmers is to make a bot that can lose and still yield a challenge for the player

>>387034218
>the whole deal here is that the AI didn't know wtf Dota2 was or how it plays before it started learning it by trial and error.
wrong, it needs to know the actions it can complete in the game at the very least
stop spouting conjecture about things you know nothing about

File: 1501909948199.webm (1MB, 480x476px) Image search: [Google]

1MB, 480x476px

if anyone wants an in detph look at how these things are setup and how they 'learn':

https://www.youtube.com/watch?v=ILsA4nyG7I0

doggo might be your face if you don't follow it for a while like I didnt at first

>>387034340
>Well it can atleast describe any concept we as humans understand.
it can't actually, it can't completely describe aesthetics, for example

>>387034381
>it needs to know the actions it can complete in the game at the very least
no it doesn't

learn2generalpurposeAI, you scrub

>>387034478
it does, because it needs to communicate with the Dota API to perform the actions at the very least, and it's not a general purpose AI, it was specifically designed to play shadow fiend 1v1 mid

>>387034448
>aesthetics

What?

You can easily describe any form or color with numbers.

How is that not the case?

>>387034596
yes but it can't presently describe what makes something inherently beautiful or aesthetic

File: 1500403958611.jpg (32KB, 547x328px) Image search: [Google]

32KB, 547x328px

STOP POSTING WITHOUT LEGACY CAPTCHA
STOP GIVING CORPORATIONS YOUR DATA
WEAR TINFOIL HATS
STOP GOOGLE NOOOWWWW

>>387031958

of COURSE they had to pick SF.. it should have been something mechanically interesting at least.

1v1 SF mid is the lamest example of the game and the easiest way to abuse "self learning AI". the stock bots in DOTA can lasthit with above average-accuracy and this little pony show was hardly any different.

This isn't anything new, hell reaperbot for quake learned by copying the other players movements and could learn maps without fancy waypoints all over the place

>>387034556
no retard. you have to gtfo and read about general purpose AI learning.

File: 1472290156727.png (242KB, 356x492px) Image search: [Google]

242KB, 356x492px

>>387034726
>x game did this years ago when the programmers built a bot into their game, its totally the same as a general AI that has no involved with icefrog whatsoever episode

>>387034691
But thats entirely subjective and directly related to all the previous inputs your brain collected and processed in its life.

And how your brain comes to that decision can most likely be made into a albeit very complex mathmatical formula of inputs.

>>387034721
SF 1v1 mid the canonical 1v1 mid dick measuring contest though.

>>387031958
>self-learning
>it's actually just scripts with maphack

>>387034925
So i'd argue yes mathematics can describe that.

You are thinking in very limited dimensions here.

>>387034925
>But thats entirely subjective and directly related to all the previous inputs your brain collected and processed in its life.
if it was entirely subjective, then nobody would agree on anything, and there wouldn't be great artistic minds who could produce something everyone agrees is much greater than other art
>most likely
is an assumption, which is not scientific or mathematic

>>387035024
That it cannot describe it is an assumption you cannot prove too.

So what now?

>>387035105
I said 'cannot presently describe'
you're making an assumption by assuming that it can in the future

>>387034926
that's my point.

in a matchup like that, it doesn't matter what you show up with the machine will always scan your dick and add 25mm

>>387031958
it beat pros in a 1v1 SF only

creating an entire team of AI is totaly different

>>387031958
this is worthless tbqh, once and AI can adapt by itself to multiple games, that's when it will be an actual full fledged AI

>people are legitimately worried that procedural AI will eventually result in a Skynet situation

>>387031958
>all 50 prizes for beating it have already been claimed yesterday (less than a day after this whole thing started)
Why do you keep posting this?

>>387031958
it's not like there's much to accomplish skill-wise in 1v1 in dota

>>387035213
t. zuckerberg

We cannnot program that which we don't understand ourselves.
We can.
One of those statements is true or false, depending on who you ask. I'm not going to fence sit, I believe in the former.
We do not understand consciousness.

>>387034721
>>387035178
You know very well that 1vs1 SF is more than just lasthitting.
If you're making an AI for the game it makes perfect sense to start with things that would be easy for AI to learn.

>>387035529
Do you have any actual reason to believe the former? Seems to me you just picked what is most palatable without any real understanding.

We've already made numerous inventions that greatly exceeded their designs and expectations, that were more than we made them to be, we definitely can program something that will turn to something we cannot understand, in fact this has already happened several times, a recent example being a chat AI that started performing enigmatic behavior and was shut down because nobody could figure out what it was doing and it kept changing.

>We do not understand consciousness
Rather, we do not even know if it's something relevant or even tangible, what we understand of consciousness implies that it is merely a play-back of our memories upon which we reflect and adapt in greater depth than our subconscious mind does, and this segregated layering of mental processes allows for separation between vital and non-vital activities in the brain, nothing more than that.

People are quick to allude value to "consciousness" that it simply doesn't tangibly have, and such beliefs are naive and foolish.

>>387031958
God this is starting to get scary

https://www.youtube.com/watch?v=kpJrmOcRBcM

>>387036168
the positioning and zoning of the bot is impressive but 1vs1 SF is still so mechanical it's not really surprising an AI can do it
I'd be impressed if they played a real game

>>387036368
>Do you have any actual reason to believe the former? Seems to me you just picked what is most palatable without any real understanding.
Transistor counts, and neuron counts. There is a large discrepancy.
Even if we isolate everything that makes us smart, excluding large sections of useless motor cortex and such.
I don't think we have enough processing power yet. Not to reach our level.
>People are quick to allude value to "consciousness" that it simply doesn't tangibly have, and such beliefs are naive and foolish.
If it doesn't have value, why are people so afraid of AI achieving it?

>>387035529
https://www.ted.com/talks/maurice_conti_the_incredible_inventions_of_intuitive_ai#t-216815

This is a video about AI that came up with designs more efficient than what humans could achieve.

Wouldn't this work as a proof that latter is actually the case already and not former?

>>387036642
The thing here is a proof of concept more than an actual result, and the "result" here being a bot that can play 1v1 SF mid isn't spectacular or notable, but the way it got there is, simply because it's a learning AI rather than something manually programmed.

Technology like this is also being implemented in things like surgery robots, and being able to learn like an actual surgeon rather than having to manually program for every eventuality is far more viable. The benefits of this tech is also enormous, since such a robot could perform surgery that no human ever could, or would (since many surgeons will simply opt not to do risky operations for the sake of reputation etc.)

>>387036642
I'm not impressed either, but I think they are very well on the right path to making a proper strong bot team.
The single roadblock that I see is the bot would need to learn to plan ahead, and none of current popular ML libraries (theano, tensorflow, caffe, torch) have support for that.

>>387036856
>Why are people afraid
People are as afraid of having their beliefs proven irrefutably wrong as they are of tangible "threats" anon, if not more so, this in part due to the human brain's inability to re-calculate all the things we have thought and determined based on such beliefs.

>I don't think we have enough processing power yet
We are decades past that, anon, we have more than enough and we can expand it since computers aren't as limited by volume as the human brain. Our ability right now is inefficient, that much is true, but the processing power we can achieve already greatly outstrips the limits of the human mind, especially since the human mind still has a huge amount of worthless artifacts, processes that no longer have value or merit but aren't entirely removed.

>>387036869
You misunderstand what I wrote, we can understand the concepts the machine designed.

>>387036869
Could the AI decide to lose a match on purpose because it
>felt sad or depressed, so it didn't feel like giving it all this round
>as a strategy to have its opponents think they are not really that good, con artist style
>because it is in love with its opponent and wants them to win more than itself
You will probably answer yes, that it can be done it just needs to be programmed to learn these things.
But that leads at the fact that it would be US wanting the AI to do these things, not the AI itself. The AI would only do what it was originally programmed to do, which is win. If we want more variables we have to program the AI to learn them.

>>387036931
>but the way it got there is
yeah but can that way be used to make something useful? They haven't actually divulged their methods so it's unclear. Winning a 1v1 last hitting competition is alot different to winning a game of dota, which involves long-term strategic thinking

As long as machine learning can bring us competent bots to play against then im all for it. At the moment no matter what bot in any game you play against, i never feels as satisfying as beating a real person.

>>387037206
So what is the meaning behind your post then? An AI is able to surprise its own creators. What more do you want from it?

What is the point of this? There are already machines that could more heavy lifting than a human could. AIs that are programmed to beat human players at chess.

>>387037290
Er, no, my answer is, when AI is complex enough, when it is not designed to play the game but to exist and enjoy existance, and playing the game is part of it, and outright refusing to play the game well makes its life more miserable, yes it would be able to do all those things naturally.

File: 1450276636378.png (103KB, 500x500px) Image search: [Google]

103KB, 500x500px

>>387031958
Dota players are so bad they can be beaten by an ia at hard difficulty.
I hope the ones that lose hanged or sudoku themselves in shame.

>>387037565
why would you program AI to mimic human emotions?

>>387037312
>Can that be used to make something useful
See the second half of my post. Dota is a good place to start with these things because it offers a good mix of repetitive (game mechanics) and varied (player interaction) factors, so it can learn in a controlled environment before being expanded upon.

The aforementioned surgery AI for example can't exactly train itself in actual surgery since that'd involve people dying going through the trial-and-error phase of learning, and while it's technically feasible to have an AI learn medicine the classical manner that'd require years of education which is difficult to implement and creates a lot of potential problems, so creating an isolated learning environment is important. Dota is such an isolated learning environment, and we can extrapolate the things we learn from using it into other fields eventually.

>>387037420
Chess is an information complete game that you can brute force/prune decision trees.

At least a game like dota, you would need the foresight to read the situation and pre position yourself to set up a good fissure as earthshaker, as one example.

>Dota is a very complicated game
AHAHAHAHAHAHAH
*breathes in*
BAWAHAHAHAHAHAHAH

Call me AI beats a human in Starcraft

>>387031958
Was it a single AI against a team or was it 5 AI against a team?
If it's the former, meh, if it's the latter, that's good

>>387037846
you can see the match here: >>387032056

>>387037789
Starcraft build orders and interactions are far more simple than 10 people making decisions and interacting, anon.

On a human level Starcraft definitely takes more skill, but on an AI level Dota 2 is far more sophisticated due to an excess in variables, hero pool is but one of many factors there.

>>387037674
You wouldn't program it to do it, they would appear as side effects (at least this is my assumption). Even now, with our primitive AIs, we only design their structure and letting them learn things by example; concepts that those AIs learn are not programmed into them.

>>387037754
well if you want to be that general, how is it any different to AI in any other field? Like the shit we spend years solving on captcha and google search? They haven't made some general AI breakthrough, they've made a bot that learns to play dota

>>387037789
AI can't beat humans in dota neither.

>>387037913
Are you a delusional dotard? Starcraft AI falls apart beyond executing the most basic zerg rushes, Dota 'strategy' is simplistic and can be understood by a toaster (lane-gank-lane-gank-go back for items-lane-gank rise repeat until u win)

>>387037945
You make breakthroughs by doing small things. You don't get breakthroughs by just sitting and wishing for it.

>>387037936
Sentinence doesn't appear as a side effect of programming an AI to complete a task

>>387037945
Again, it's not what it does that is interesting, it's how it does it, and while that's not unique as there are hundreds of ongoing projects attempting this shit, the results it achieved so far are at least a step in the right direction.

This is completely different from the shit we see in captchas, the fact that you even go there implies you don't really understand the concept at a basic level.

>>387038042
Mate please stop. You are misunderstanding. there is no AI that can win vs humans in dota yet.

File: 1967___mug_omniknight_reaction.png (161KB, 508x524px) Image search: [Google]

161KB, 508x524px

>>387031958
>1 hero
>1 lane
>not allowed to buy bottle
>not allowed to buy soul ring
>not allowed to buy infused raindrop
>not allowed to take runes
>not allowed to use neutrals
>not allowed to use shrine
>10 minute timer preventing even more strategies
Woooowww what a victory for the bot! It can right click and use a single spell faster than me!

>>387038042
Ah, just shitposting then, carry on lad.

>>387037565
But then what you are talking about is consciousness and not intelligence. You are talking about something that essentially is a soul, like in the movie Her. Not just an algorithm that is made to find patterns and the shortest way to those paths and patterns.

That's science fiction, not science.

>>387038071
An interesting theory, but, alas, you have 0 proof and 0 evidence to show for it.

File: 1488390836542.png (11KB, 307x462px) Image search: [Google]

11KB, 307x462px

>>387031958
>dota2
call me when it can actually play a real game not babbbys first video gaem.

>>387038143
>>387038128
I understand that senpai, the point still stands, when AI finally do manage to beat humans in Dota, it's still going to be babby-tier in SC

>>387038201
1) I'm a computer programmer
2) I'm an AI programmer
3) I'm not a fucking retard
computers don't reprogram themselves unless you tell them to. If you give a computer a goal it's not going to overrwrite that goal (unless you give it the ability to)

>>387038256
No, the amount of variables in a game of starcraft is absurdly smaller than the variables in a dota game. Count the amount of units, upgrades and actions possible in a game of starcraft, then count the amount of heroes, items, skills, decisions, etc. in dota and multiply that by the amount of players.

SC/SC2 is but a fraction of Dota2 in that regard.

>>387038195
And we are doing our best to implement that, playing with neural networks. If you want to say that neural networks are the wrong way to approach our goal, then I'd like to hear your reasoning why.

>>387038410
>unless you give it the ability to
Fascinating. So it is possible after all then?

>>387038256
winning at starcraft and winning at dota are almost the same task for a computer, the rulesets are basically the same, you're comparing how difficult you think the task is for humans because you think your favourite game is better than someone elses favourite game

>>387038461
>Fascinating. So it is possible after all then?
Does your computer decide to stop doing what you tell it to because it's bored? You could give a computer the ability to overwrite all it's instructions and evolve in any random chaotic matter, but there's literally no point unless you're trying to create artifical conciousness or something

>>387038461
Neural networks is a schematic approach to replicate how a brain works. That is give the AI points of connection.
For instance you can teach a AI a chess players every move, teach all moves available in chess and have the AI beat said chess player.
However, I think we will find that humans are not just the product of neural connections. We have yet to find out where our consciousness is or even what it is exactly.
And then we are back at what I originally asked. Would the AI forfeit the chess match because
>it had fallen in love with the cess player
>felt the chess player deserved to win
so on and so forth. It could if we also programmed it with those variables but then it would be a want in us, we wanted the AI to do those things, the AI would only want to win because that is what it was programmed to do.

>>387038410
>1) I'm a computer programmer
>2) I'm an AI programmer
Also, my dad works at Nintendo NASA division and he says aliens already created singularity

>>387038689
No, my computer doesn't do that, and I don't see why you are asking that in the first place. It's a computer, not an AI. The pipe dream of AI researcher is to have a system that can set goals for itself and learn, without you defining its internal structure and then providing sets of problems and solutions for it to learn from. Then you'd be able to give problems to it and it would be able to provide answers according to what it learned. Clearly we're not there yet, but to claim that NN is not the right path you need a bit more than "my dad works at Nintendo".

>>387038746
>However, I think we will find that humans are not just the product of neural connections. We have yet to find out where our consciousness is or even what it is exactly.
Then it's just a matter of belief. I don't think I will change your mind, but neither you will change mine.

>>387032790
I think they should add the AI into dota 2 after it gets better, though
Like, the bots get incrementally more difficult based on how good you play, so you'll have to get better in order to beat it

>>387038410
Then you must be a shit AI programmer

I also program AI
DOTA2 has a finite number of possible moves unlike, say, the real world
Given this, you can come up with a list of possible moves
You then program at AI to come up with the best move to make given a situation
Use a point system computed at parallel using FPGA boards
The AI can ALTER the point per move depending on how the enemy moves/plays

>>387039079
You're talking about monte carlo tree search and that's not what the demonstrated bot uses.

Not the person you're responding to.

File: 1483195051867.jpg (349KB, 1920x1080px) Image search: [Google]

349KB, 1920x1080px

>>387033001
Read up on Machine Learning right now.

>>387039058
Its not a matter of belief that we don't know what consciousness is though, or where exactly it is.
You can teach oil painting techniques perfectly to any individual, and they can have perfect technique but their work might never touch someone they way someone with less technique have. Some people don't even get that experience from art.

To me it seems like a pipedream, the same way people dreamed of an atomic age with flying cars, jetpacks and rocket trips to their office on the moon.

>>387039079
>>387038410

Ah shit I'm retarded
I didn't follow the argument
Basically, yeah, what the guy said

Using current tech, it is simply not possible for an AI program to have "sentience"
Too many possible options to take and do in the real world

DOTA allows a programmer to limit the choices to a finite set that an AI can process at an acceptable speed

>>387039069
Playing well against AI and against players is two entirely different things, unless the AI can turn on russian try-hard mode or italian shitter with to much of an ego to support as a support.

>>387039079
If you program AI and don't understand basic math you won't be getting very far
The amount of possible moves in Dota2 make it completely impractical to search over them all

>>387039321
Yes, but from a gameplay point of view it would be useful, if you were forced to use different tactics at right times, instead of just cheesing the fuck out of the game

>>387039238
>we don't know what consciousness is
Truth; not a matter of belief.

>what consciousness is
Matter of belief.

>>387039257
Yeah, you are retarded all right. No one mentioned using current tech in the discussion.

>>387038937
AI is about making programs than solve problems, not programs that suddenly wake up and start magically acting human, like that would make them better at problem-solving. You need to stop believing in all the pop-sci bullshit

>>387039406
We're not going to get any further in this conversation really, because we differ too much at this point. But it was interesting.

>>387039552
>like that would make them better at problem-solving
It would and you are a really dumb person if you think otherwise. Humans are amazing at problem solving. If we could make an AI that acts like human, we definitely would. Yes, we can't. Yet. For sure because we lack the computational power, and most likely because because our methods of using that computational power are not good enough. But we are doing our best to arrive at that goal. To claim that we're don't want to make a program that acts like a human is just false.

>>387039614
What, are you scared of talking to someone who holds a different opinion?

>>387037913
>Starcraft build orders and interactions are far more simple than 10 people making decisions and interacting, anon.

Except the AI was tested in a 1 vs 1 mid only SF only situation. The argument falls apart since this AI was just a glorified last-hitter,not that complicated. And of course a computer can last hit faster and with more precision than a human.

>>387039810
>Humans are amazing at problem solving
then use humans, not machines. Emotions don't help you solve problems. Trying to recreate human conciousness doesn't really serve a practical purpose, we already have humans and we already know how to make more of them
>For sure because we lack the computational power
no not 'for sure' because nobody has a fucking clue where conciousness comes from, what it is or how it works, to say anything otherwise is arrogance.

>>387039858
Not in any way, but to say that its a matter of belief that we don't know what consciousness is, when its the one question that we have tried to understood for centuries and no one. Not poets, philosophers, scientists or religious folks have managed to come up with anything on yet. Is to me just a useless afternoon. I have dinner to cook. Its ok that we got this far in the conversation to me, I'm not gonna fight.

>>387039962
>then use humans, not machines
>Trying to recreate human conciousness doesn't really serve a practical purpose, we already have humans and we already know how to make more of them
Costs. Humans who are good at problem solving are expensive.

>Emotions don't help you solve problems.
Debatable. You're using your whole life experience to solve problems. Who is to say emotions played no role in obtaining that?

>no not 'for sure' because nobody has a fucking clue where conciousness comes from, what it is or how it works, to say anything otherwise is arrogance.
We know about kinds of neurons in human brain and their amounts. We know enough to make an approximation, and it's normal to assume we need at least that amount of computational power to make an AI as smart as human.

>>387040026
>but to say that its a matter of belief that we don't know what consciousness is
Anon, I clearly agreed with you that we don't know what consciousness is. We don't know. But we can have beliefs. You believe human brain alone is not enough to demonstrate consciousness. I believe it is.

>>387040152
>We know about kinds of neurons in human brain and their amounts
we don't know if conciousness is a product of those neurons, we barely know about even the surface functionality of the brain

If you want an intelligence to solve problems quickly, without fail or error or getting bored, emotions are a huge drawback.

>>387040385
>we don't know if conciousness is a product of those neurons, we barely know about even the surface functionality of the brain
We know enough to make an approximation, and it's normal to assume we need at least that amount of computational power to make an AI as smart as human.

>If you want an intelligence to solve problems quickly, without fail or error or getting bored, emotions are a huge drawback.
The thing is there is no AI (or any other way for machine) for solving general problems, so you can't lecture me on what is good and what isn't for general problem solving.

of course machines can beat humans

you can program a machine to carry out a task with perfect precision every time

doesn't really matter because you still need a human to program it in the first place and in that sense its nothing more than a tool

>>387032632
Look at all the bullshit stuff you can do with AI in starcraft 2.

https://www.youtube.com/watch?v=3PLplRDSgpo

https://www.youtube.com/watch?v=IKVFZ28ybQs

>>387040540
>We know enough to make an approximation
no we don't actually, people have been able to do only the most basic of experiments with the brain and even then they usually result in inconclusive evidence. Assuming we're close to figuring out how the brain works is just the arrogance of modern scienists who want to act like they've got everything figured out when they haven't

>you can't lecture me on what is good and what isn't for general problem solving.
yeah I can because I can solve problems and I can tell you if I had no emotions and no intelligence and machine precision I'd be much better at it, and my computer would be alot less efficient if it got sad or bored when I told it to do specific things

>>387040832
lol starcraft

>>387040867
>yeah I can because I can solve problems and I can tell you if I had no emotions and no intelligence and machine precision I'd be much better at it
I will write it again, one more time, and if this time you don't in some way demonstrate that you made an effort to understand what I wrote, I will not respond to you. Write your answer carefully:

You're using your whole life experience to solve problems. Who is to say emotions played no role in obtaining that?

>>387034831
So the bot just learned image recognition and control over the mouse and keyboard too or what?

>>387041275
I'm pretty sure that's not the case. The person said bot needs to interact with Dota2 API is most likely right.

There is a doom bot that doesn't use an API and just works by looking at the screen, though.

>>387041114
I understand what you're saying, but your point is irrelevant. You're assuming humans are the golden standard and we should be making machines that emulate humans. If you want a machine that solves a specific task better than humans do, what use are human emotions?

>>387031958
AI was always better than humans.
There's AI in old as fuck chess games which are almost impossible to beat.

Just because the AI in almost all games is shit doesn't mean people weren't able to make a better one, they deliberately made stupid AI so as to not constantly piss off the players.

I swear, you kids these days know fucking nothing anymore.

>>387041558
You're claiming to know the answer, but we actually don't have a machine that solves general problems, so I don't see where this confidence is coming from.

I'm not assuming that humans are golden standard. Re-read my posts. I'm saying that if we could write an AI that would act like human, we definitely would, and then we would try and use it to solve all kinds of tasks using machines that we can't currently.

>>387041651
Nice bait.

>>387041558
Additionally:
>If you want a machine that solves a specific task better than humans do, what use are human emotions?
I don't want better, I want as good as, at a lower cost.

>CS 1.6 bots with perfect accuracy
Where were you when Artificial Intelligence made human vidya players obsolete?

>>387042037
Not like anyone's going to watch bots play, at least not until they develop personalities, which seems to me decades if not hundreds of years away.

>>387042037
>>CS 1.6 bots with perfect accuracy
=/= AI

>>387038746
>We have yet to find out where our consciousness is or even what it is exactly.
Of course you can't know what consciousness is, it's an invention, if you haven't made it clear what you're looking for, how can you explain it? You need to sort your shit out, then you can explain how that shit comes to be.

>>387041558
And finally, you keep harping on about human emotions being unnecessary, but here are my posts about human emotions:

>>387037565
>>387037936

I'm not suggesting to design AI that would exhibit human emotions, I am hypothesizing that when you make a general enough AI for it to learn things by himself, emotions would appear as side effect.

>>387042139
I don't think it would be that hard to develop bot personalities, if we take a strategy game for example you could weigh different bots differently, so you could make a coward bot that only attacks when victory is almost assured by weighing loss of manpower more heavily than for the other bots, or make bots with favorite tactic by implementing some sort of feedback mechanism, or a bot that acted aggressively by penalizing it the more time it took until victory, plus all the manual editing you could do

the strategy genre would heavily be improved by making it not only a strategical, but a psychological challenge too
grow a bunch of AI that you freeze, then make them modular too (so you can basically generate new AI by putting together a bunch of pieces), then you put them in the game.

development would probably be too bothersome before someone managed to invent a general strategy game AI

>>387043279
>I don't think it would be that hard to develop bot personalities
Maybe.

>I don't think it would be that hard to develop bot personalities that would make bots playing interesting to watch
No, no, no, no, no, no, no, no, no, no, no, no.

>>387043443
esports bots not being fun to watch is more of a matter of celebrity culture
like, aside from the shock of the reveal, if it was revealed that every professional dota 2 player was just an actor and the actual players were stored on a supercomputer interest would die down fast I bet

>>387043662
If bots could exhibit their personality the same way people do, the interest still be there.

>>387043662
>>387043752
Actually the fun depends entirely on the programmer's intentional/unintentional kinks and the reactions of it's human opponent/watcher.
Like Dendi being scared by Open AI's zoning.
Or this one: https://www.youtube.com/watch?v=8JwGYpRU6Ao

The fun is created. Unwittingly or not.

>>387044079
If it's programmed it, then it's not bot producing amusement, it's the human programmer producing amusement. For bot to be advanced enough to exhibit enough personality to create fun we need, as I wrote, decades at least.

>>387044079
Also,
>linking to a 42 minute youtube video
pls go, go, go, and stay go

>>387032790
>It has access to all the variables

No it doesn't. The entire point of these AI's are to make them as human like as possible.

All they can see are the pixels on the screen, same as a human.

The idea is to create an AI that can learn tasks in the real world using a set of human-like senses, eyesight, hearing etc. So to just plug in all the code variables in DOTA would defeat the point of the entire project.

>>387044450
>All they can see are the pixels on the screen, same as a human.
I'm pretty sure you're wrong here. They use APIs provided by dota.

>>387044263
No shit, Sherlock. That's what I said.

>>387044415
Oops. Match starts at 1:58 then goes on for 20 mins or so.

Where is the problem?

Same thing will eventually happen IRL. Robots created by humans will "replace" the humans.
It's inevitable. We will end up being the creators of a new race that conquers the universe. Let's just hope or rather make sure they will treat us like Gods.

>>387044582
You have been misled. I'm not Sherlock, even if there are similarities.

>>387044752
>Same thing will eventually happen IRL
>Robots created by humans will "replace" the humans

But no one got replaced.

File: 128015471936.jpg (30KB, 552x360px) Image search: [Google]

30KB, 552x360px

>>387031958
Did he really just say Dota is a complicated game?

>esport fatfuck plays against the CPU
>loses

Any more proofs needed?
Esport '''''''athletes''''''' can't even beat games on easy mode yet there are people worshipping those useless fucks.

>>387044854
Guess why I put the replace in quotation marks.
Succeed would be more accurate, I guess, but still misses the point.

>>387045042
So the same thing will happen eventually IRL? No one will actually get replaced?

>>387044940
HEEEH HEEEH HEEEH MOBAS AMIRITE

>>387044752
Do you treat your seniles like Gods? What if you can rejuvenate them as your equals? What if you rejuvenate them too much?

There's no "leave them humans alone". Either it takes us for the ride to infinity or we're doomed.

>>387031958
ASSFAGGOTS are all about being able to do your meta the fastest with the least amount of inputs so it's no surprise here. It's like an aimbot pretty much.

>>387045260
it depends entirely on how you build the AI
any sort of law of psychology that you apply to humans is thrown out the window when you talk about AI

File: uZVfM.jpg (157KB, 530x353px) Image search: [Google]

157KB, 530x353px

Sure it can beat a single lane Moba "pro gamer".

But it will not beat a Starcraft player for a very long time where the game is not won by mere reflex. Boxer and Flash on AI after google's AI blows a great GO player out of the water 3-1.

http://www.teamliquid.net/forum/starcraft-2/505707-boxer-alphago-wont-beat-humans-in-starcraft

>>387045517
By senile I meant dumber and slower, but yet can reach AGI given the right hardware. It got nothing to do with species or what we're familiar with.
True AI is not just a computer. It's us with better hardware.

OpenAI is just a computer. A learning one, but not an expanding unpredictability like us.

>>387045747
Not without crazy APM too. Not yet at least.

>>387034149
Why bring souls into the conversation? Without a soul that's still unproven. Also, please use math to explain the historical significance of war in the Middle East, or use math to explain
“Thirteen Ways of Looking at a Blackbird”

>>387046439
>but not an expanding unpredictability like us

We are most likely 100% predictable, but we are an extremely chaotic system, so it's simply difficult to make an accurate prediction about us.

>>387031958
1v1 isn't Dota though, still pretty cool
Coolest part is they did this for Dota and not League of Babbies.

>>387032632
I seems pretty easy, the real challenge would be a team of 5 AI playing DotA against real players, how they would handle interactions with each AI is gonna be way harder for them to manage than just being in a 1v1 situation like Starcraft or that 1v1 mid.

>>387047217
Science or math can't predict future discoveries. If it did it would discover it now. Hence not predicting anything.

>We are most likely 100% predictable
Odds is not the right term to talk about it. It's either true or false. We are the chaos, that can restore order indefinitely and without limit. Nothing else in nature can do that, that we know of.
Even biological evolution (order out of chaos) is limited.

>>387045004
>useless
They literally make millions bud, hope you can do better.

>>387047902
>spending millions on fat gooks who can't even win against AI

Thats even retarded than hiring a woman who cannot code as a programmer because you gave yourself some retarded diversity quota.

File: 1234423341792.jpg (3KB, 126x126px) Image search: [Google]

3KB, 126x126px

>Elon musk said that Dota 2 is the most complex game, more complex than Chess and Go where Alphago already beaten both Chess and Go world master.
Why /v/ like this Steve Jobs clone again?

>>387048095
chess and go use grids and turns
dota uses continuous space in real time

>>387048095
I don't like him, but Dota is definitely more difficult to write AI that beats humans for.

>>387048095
He's right, though. Go and chess were solved by throwing a shitload of processing power at monte carlo tree search. The problem presented by chess or by go is easier to mathematically represent than dota.

>>387048671
>>387048962
Chess and Go are all about reading your opponents mind and planning ahead 20 turns in advance

Dota2 is all about character management and less about the shit that happens on the map.

>>387049316
I won't even argue with you about what is needed for success in dota 2. Bot for chess was made a decade ago, bot for dota still doesn't exist.

>>387049316
> less about the shit that happens on the map.
>Got beaten by 50 Ti7 human attendee by agrooing the creeps to the base.

>>387047793
>Science or math can't predict future discoveries. If it did it would discover it now. Hence not predicting anything.
If we had a system that could predict the future, the ultimate problem would be that it would need to predict the results of itself. The problem then becomes to give a result that gives an identity, since the future depends on the machine's result.

>>387031958
No shit, even if the ai has access to only the pixels on the screen, they still have a huge advantage in mechanical response time. They will comprehend and see movement and attack cues better than any human ever can.
But who gives a shit.

I'm not really impressed since 90% of the skill involved in SF 1v1 can be automated pretty trivialy (like last hits and casting raze with perfect accuracy).

Lets see when they make an AI team that actually challenge pros

>>387050850
OpenAI team goal is to get 5v5 AIbots to be done and presented in the ti8.next year.
So far most AIbots only do 1vs1 game like chess and go, seeing them teaming up in a team based game where abilities combos and decision making are needed to defeat your opponent will be interesting to watch.

>>387031958
Is OpenAI able to access console variables? For example, knowing when you are EXACTLY the right distance away to harass the enemy mid.
If so, it's fucking cheating. Call me when it can run entirely by processing a video feed and not manipulating things behind the scenes.

>>387052586
Anon, the 50 people who won against the AI were the one who "cheating".

Can this AI work with melee?

>>387045747
>gook click
kys

>>387032632
the auto attacks on computer opponents in SC2 was dumbed down to make player input more of a factor, so just doing that is enough to beat people