Elon Musk and the dangers of AI

His AI just learned by itself to beat the best DOTA2 player.

>The bot kept playing until its skill level reached that of the world's best Dota 2 players, he added.
>"vastly more complex than chess & Go"

Holy Shit, he was right all along. Thoughts?

Other urls found in this thread:

twitter.com/riningear/status/896297256550252545
en.wikipedia.org/wiki/Millennium_Prize_Problems
twitter.com/SFWRedditVideos

Not hard to do so mechanically 1v1 with perfect reaction times and arithmetic. Regular people have already beaten it by exploiting quirks in its behaviour like diving for couriers or pulling its creep wave.
Pretty sure someone beat it regularlyjust by stacking insta-regen and manfighting it, the bot didnt take into his items into account.
We are still miles away from bots playing the real game competently. This was just a very small subset of 1v1 gameplay from a 5v5 game using only one character out of 110 unique ones.

He's a total faggot who's against freedom.

They're going to showcase 5v5 next year, likely going to be running simulations all year

>AI wins at a videogame
How is this news?

They taught it to outmaneuver, outposition, outjuke and even trick the human player.

The Bot wasn't just lasthitting perfectly, the thing was manipulating creep position, it's own position, dodging Dandy's skills and faking it's own.

It learned some very complex behaviour.

Musk was bullied and horribly beaten as a child because he was a faggot.

In order to cope with his terrible home life and school life, he retreated into Sci-Fi and computers.

Unfortunately, he never became a good programmer. Luckily for him though, he was able to spin his delusions of grandeur into a persona that he could market.

And market he did. Everyone's drunk Elon's kool-aid. Now it's time for the cyanide in the punch bowl to kick in.

>implying

Its a game meant to be played versus other humans. For example you could have a robot basketball player that could make every shot and every pass with mathematical accuracy from its position versus Joe Blow who obviously could not. There would be no point really, since again its a competition against other humans

what the fuck is this post

wow, pure AI is getting close to shitty algorithms

SKYNET WILL KILL US ALL

huh?

ur a fkin moron tbqh

The bot in Dota is no different than making a CSGO bot lock on to your head. It's not difficult.

it programmed itself from almost nothing m8

There were bots that were unkillable in Cs 1.6 like one and a half decade ago.

>the best DOTA2 player.
>Dendi

One is programming a bot to apply the answer to a puzzle and the other is programming a bot to discover the answer to a puzzle. This is pretty advanced stuff that has never been done before.

Yes, the neural network meme.
Let's spend a lot of computational power and time in order to let computers solve problems we already know the solutions for.

If you already know the correct answer, you have solved the problem. And you need to input the correct answer into the neural network for it to work.

Call me when neural networks can intuit what "winning" even is.

Boring, overhyped nonsense. They aren't close to having an AI that can beat a human 1v1 in most situations with arbitrary heros, let alone an AI that can compete 5v5. But even if they did, who fucking cares? It's a game for humans. Furthermore, the whole thing is a dumb publicity stunt for the new AI research project, which is of course going to fail within the next decade when it doesn't meet any of its outlandish promises.

Also musk is a hack.

>players

>They aren't close to having an AI that can beat a human 1v1 in most situations with arbitrary heros, let alone an AI that can compete 5v5. But even if they did, who fucking cares?
are you havin a giggle m8?

twitter.com/riningear/status/896297256550252545

The bot was beat at least 50 times.

your point?

Context. I keep seeing "bot beats pro dota player" headlines, but it's still very limited and a lot of random non-pros at a con beat it within 24 hours.

I hope you never get to benefit from anything Musk has done or will do in the next 40 years.

u like robot? u like battery car? u like spaceship?

too bad, i already make those, no u!!

Yes, goy, you'll be trekking on mars next year. You could be playing golf on the moon. You could be lounging on Phobos with your AI waifu.

You could, but, for all this to happen you

have to

DONATE

NOW,

GOYM

The word "projecting" is often overused in internet arguments, but if this isn't projecting nothing is. That's some high fucking level insecurity.

can anyone explain what he is talking about for regulating AI ? I have a degree in CS and I participate in AI programming competitions and I have no idea what he could even be referring to.

he also claimed it was a harder problem space than Go and Chess...which is patently false. He used to be a good guy, I dont know why he has gone full retard as of late.

stop with the hyperbole. it didn't program itself. someone wrote a program that executed as intended.

I mean, say what you want about the depth and complexity of all three games, but it's not hard to see why a real time video game with 10 players and enormous amount of hero combinations would be a harder problem space than a game with pieces on a board.

what you say is true. they wrote a program that didn't do that though. it was a gimped (features not allowed from the actual game), mirror match where first blood counted as winning.

Baby steps

it didn't even do that though. if you messed with the courier or their creep wave it broke it.

>lmao don't worry about it it's nothing
10 years later
>whoops someone has created skynet now we're all dead maybe we should have regulated things to prevent everybody dying
This is why you're all short-sighted idiots.

I'll be impressed when bare hardware can approach an arbitrary object it has no notion of, and accurately analyze it. The brain can do this, even insects can do this. Except this won't happen, because everything is in software, with strict rulesets and methods first derived by humans. Neural networks are not AI.

Dudes I fly big big pencils up your ass.

Also, I have no hair.

>basing your knowledge of AI on scifi films
Imagine being this retarded.

you can't prevent someone from purposefully creating the skynet you fucking idiot
everyone keeps thinking it'll be "accidental" for some stupid fucking reason
even though all the shitty things that have happened and will continue to happen were conscious decisions

If intelligence explosion is a real thing, then we're already far too late

its not being short sighted its called being knowledgeable in the field. We look towards the future, we have stuff like: en.wikipedia.org/wiki/Millennium_Prize_Problems we look at the time and space complexity of various algorithms and see how they can be better in the future.

If you take Elon seriously you would have to imagine that if every roadblock in theory of computation was magically solved (which you have no evidence that it ever will). Its like forcing your doctor to prepare for a world where all cancers and diseases are curable. its total nonsense.

>which is patently false
No, I'm quite certain that real-time DotA with its complete feature set would be much, much more difficult to play like a human than a turn-based, perfect information game with a large (but limited) problem space. If you want to take into account all the possible actions in any moment of a DotA game, that alone would essentially give you arbitrarily large numbers of possibilities. Many of them would be essentially equivalent, but I guess that's not necessarily a given for a "dumb" AI just trying shit until something sticks.

But anyway, comparing a real-time computer game with a perfect information, turn based board game doesn't really make any sense.

Dude, I advertise non-existent projects online.

Like - you say - pooping on mars next Christmas?

Sure!! Anything can happen!

i am talking about the problem that they actually attempted to solve and are gloating about on twitter. not actually programming good AI for dota 2.

Honestly, if humans manage to create a strong AI that creates a better one, etc, who's to say that they aren't the superior beings?
They probably won't straight up kill humans, but instead just ignore us. When you dig a foundation for a house, you don't care about the beetles and worms burrowing in the ground. They die, so what, there are billions more.

The machines will probably soon leave earth since there's no point on staying on a planet where corrosive water and sand is constantly being whipped around by weather, when you could just mine asteroids and have virtually unlimited resources.

guaranteed replies
also
>get off Sup Forums, Jeff Bezos

Yes, it's limited and doesn't actually have a grasp of the entire game, but it's still a start on a problem which should be way more difficult than Go.

Your post clearly shows your ignorance. First ask yourself: why are we making AI?

nice try Cuckberg

I don't think it's plausible that an AI can reliably create a smarter AI, but I'm still relatively uninformed about advanced AI like that

it's the same answer to why do you create any programs in the first place. people act like its magic, it is just a subset of algorithms.

Is it really more complex than Go?

Nice credentials, must have cost you a lot of money.

This is all super-hypothetical (in response to the tard's post about "muh skynet), but if we do crack the code of intelligence, and emulate it on a computer, it is possible that said computer will be able to run many times faster than our own brains, with perfect recall, etc.

Imagine being able to think 80x faster than your peers. It will quickly move beyond the pace human development.

A lot of R&D experiments come from fiction such as sci-fi.

To Psyonix this is revolutionary

you going to answer the question or no ? point being I am not a total layman in the field. was just trying to be polite in saying i believe musk is talking out his ass in this instance unless someone can show me something to the contrary.

I'm not sure if more intelligence leads to faster discovery past a certain point. I mean science is basically a series of random series of questions and experiments. It's interesting thoigh

what they actually programmed ? no.
actual dota ? yes.

>Musk warns about the potential dangers of an AI thats allowed to run amuck and unchecked in positions of power within human civilization.

t's not hard to imagine how things could go wrong. You're living in la-la land if you can't imagine even one negative scenario.

I would be impressed if the AI had the same apparatuses to interact with the game as a human player did (Mouse, keyboard, monitor) but that's not the case here. Instead the AI had direct access to the computer's memory and was a program running on the same machine as the game.

>a computer can play DOTA
>therefore human civilization will lay in ruins
Lol.

strong AI is a myth. All the predictions made by supposed experts since in the 1950's have been wrong every time.

>a server can take player inputs and play around them in real time
>WHOA DUDE WATCH OUT THE ROBOTS ARE GOING TO WHITE GENOCIDE US

Not the guy you replied to, but given most of his achievements either require big government money (spacex) or are selling to the rich (tesla), most people aren't benefiting from what he does anyway. You might one to take one step back from the bubble the internet puts you in.

>AI in its infant stage can only play video games therefore 50 years from now things will be the same

You're both fools, just saying. Nobody is making definitive statements like

>AI *will* become evil and try to kill us all

But we need to examine every possibility from every angle. We're treading in uncharted waters. Literally no humans have ever had to deal with this, and nobody knows what will happen or where it will lead

>YOU'RE ALL FOOLS I TELL YOU
>FOOOOOOOOOOOOOOOOOOOOOLS
>THE PROPHECY WILL COME TRUE
this is a hymn for the hymnless

Hydraulics are too rigid and have a shitty response time. That's why robots are so boring. I imagine the ANNs would have a blast if they could operate on human-like muscles. For example they could learn to twitch quickly until its arm is in correct position (which is how humans catch the ball) not just "calculate where the ball will fall and start linearly decreasing pressure".

>THE PROPHECY WILL COME TRUE

Lol...again, nobody is making definitive statements. Only you are.

Keep being stupid, it's making me laugh

DUDE I HAVE A FAIRYTALE FOR YOU

THIS FAIRYTALE IS BASED ON ASIMOV

ASIMOV AND ROBOTS

SO THE WORLD IS UNDER PERILS

DONATE NOW GOYM

This guy is right, everyone that disagrees should go back to reddít. Redditors (millennial drones) are absolutely infatuated with musk. Go on any tesla/spacex yt video, and scroll through the comments. Everyone is jerking each other off and acting like elon invented everything when in reality he didn't do shit, it was all the work of the engineers. They act like tesla will save the environment when in reality EVs are significantly more harmful for the environment.

At what point does "kill all humans" and AI interect?

Terminator wasn't based on actual predictions. It was designed to play on the fears of nuclear war and the fear of the unknown (technology), and to have an otherworldly enemy that's not just another alien movie. It's a movie, there's no deeper meaning behind this.

Bots have been beating people at vidya for decades.

Nothing new (yawn).

Cool. Now let's see it face a white player.

>Lol...again, nobody is making definitive statements.
Except, by even contemplating the possibility that AI will become as overtly powerful and menacing as you believe, you're partaking in schrodinger's AI.
But sure, keep calling others stupid when you can't even understand simple concepts. All that projection is lovely babe.

>At what point does "kill all humans" and AI interect?

I don't know, and neither does any other human being. As I keep repeating (to deaf ears apparently), no humans have ever had to deal with this before. We don't know what will happen or where it will lead, so we have to tread cautiously.

That's all I'm saying really. Not talking about killing humans and all the other crap you guys keep inferring.

>EVs are significantly more harmful for the environment
That's not the point.

The idea is to free our transportation infrastructure from fossil fuels, so we can move onto something cleaner.
Also, a coal burning plant making electricity is cleaner and more efficient than an variable speed ICE engine running directly on hydrocarbons.

No, you're right, it's just that those morons aren't capable of critical thinking.

>We dunt undastand it. Muh uncharted teratory. AI is ebul and will kill us

Just because you dumb nigger failed discrete potability theory 101 doesn't mean all of CS is on your level.

>hurr Musk didn't do shit, it was us! engineers!
Communist detected.
AI DRIVEN HELICOPTER RIDES WHEN.

DUDE I HAVE SOME STORIES FOR YOU

It's funny you say that, shill, because unserviceable vehicles are exactly that, communism. Especially since Tesla/spacex wouldn't exist without government funding.

DUDE JUST WAIT FOR MY NEXT ADS

IT WILL BE PHENOMENAL

DUDE
U
D
E

dude i have some bullshit

Imagine being this faggot

All me btw.

This isn't like a standard "levels of difficulty" vs CPU thing, nobody programmed tactics into the AI or taught it anything. In 2 weeks from straight neutral network learning the system learned how to "play" DOTA to the point where it could beat international tier professionals, albeit in a limited variant of the game

>uncharted territory
Yes and no.
FFNN have no "risks" whatsoever. They have no memory, aside from the actual coefficients.
Unsupervised learning + recursive NN is the future of artificial decision-making, and could pose a problem in a possible future (for with the technology of today very little may be done), but even then, unless idiots start to hook them up to the real world, the only cause for concern would be moral issues.

>EVs are significantly more harmful for the environment.
Only the batteries and only if the materials are mined by the Chinese.

>unless idiots start to hook them up to the real world
Who's gonna stop this?

>the only cause for concern would be moral issues.
What moral issues?

anyone with common sense who doesn't want to be held responsible. we already have had accidents like the Therac-25 actually killing and severely injuring people and securities trading bots fucking up and wasting millions of dollars. and those things are extremely simple in comparison.

>Here are 4 buttons you can press
>Don't die
There's a clear difference between setting the parameters for a neutral network and hard coding tactics and technique. Real time interaction with human players doesn't have a "solved" solution/answer, the same way control system optimization for complex embedded systems doesn't have an clear answer. NN isn't a meme, it's algorithmic optimization from the ground up.

>Who's gonna stop this?
It will depend. If such a contraption achieves some form of sentience (as opposed to sapience), who will it be under? Will it be a great breakthrough in the art, and come from a revolutionary idea? Will it originate purely from advancement in storage and computation speed? Who knows?
>What moral issues?
Suppose such machinery could learn, maybe with some help, to live as, say, a ladybug, or a cockroach, under some simulation. Would it be right to shut it down and remove its progress? Should the fact that it is a man-made sentient object weigh on such matters?

>Schrodinger’s AI
Hey, thats pretty good

>this is a dog
>we taught it to sit
>we taught it to fetch
>we taught it to lead the blind
>we taught it to smell for drugs
>we taught it how to be a police k9 unit
>OMG, dogs so intelligent they teach other dogs to be even more intelligence until they hit a singularity and have exponential intelligence, and they are already police so they will by extremely dangerous.
>Beware teaching dogs tricks

ROFL the bot couldn't even use a bottle lmao