A Dota 2 pro was beaten by Elon Musk’s OpenAI bot during this year’s International

Dota 2’s The International 7 tournament is over, having concluded this weekend by crowning EU group Team Liquid — which arrived to the finals by way of the loser’s bracket — champions. Liquid walks off with over $10M in earnings.

But perhaps the most interesting bit for folks who don’t follow Dota 2 closely came on Friday when the ubiquitous Elon Musk revealed his company’s Dota 2 bot, which he says learned a lifetime’s worth of game skills in just two weeks. To prove it, he pitted the bot against pro player Danylo “Dendi” Ishutin in a demo match… and the bot won easily twice.

Lest you think Musk has aims to take over the world, recall that his company OpenAI was founded to stop the AI from taking over the world, meaning this demo was ostensibly another high-profile attempt to provoke regulation.

Source: The Verge. Cheers, Sally!
SHARE THIS ARTICLE
Code of Conduct | Edit Your Profile | Commenting FAQ | Badge Reclamation | Badge Key

LEAVE A COMMENT

43 Comments on "A Dota 2 pro was beaten by Elon Musk’s OpenAI bot during this year’s International"

Subscribe to:
Sort by:   newest | oldest | most liked
Reader
Bryan Correll

I for one look forward to a post-human society.

Reader
shear

Do you think they’ll make it as toxic as the players too?

Reader
Herro Mongorian

There goes Elon fear mongering over AI again!

Reader
Sorenthaz

From what I heard during the stream it also beat Arteezy, Sumail, and some other big names that I forget about.

I don’t even think Dendi was at this year’s International as a competitor, and the last time he was even considered a top player was in The International 3 when Na’vi made second place vs Alliance. He just has a truckton of fame because he was one of the older pros, and was one of the world champions for The International 2 most famous for his Pudge plays, but his time as a top level player has steadily gone to the wayside as he hasn’t bothered moving to a different team like others have.

miol
Reader
miol

People feel so old so fast in these games, it’s baffling to me!

Even just watching last year’s EG captain ppd only as a commentator anymore in a suit in this year’s TI, made me feel so old myself! o.O

Reader
Sorenthaz

There’s just so much micro management and knowledge involved in a game like DotA 2. Fear was like 28 or so before he retired to become Evil Geniuses’ coach I think, but yeah a lot of it also just comes down to things steadily changing and pro players not being able to always do what they’re comfortable with. Same issue happens with LoL but probably even on a worse scale, just because in LoL the game drastically changes every few months whereas DotA 2 seems to not change things as drastically on such a frequent basis.

At least in DotA 2 it seems that pro players can stay more relevant on a regular basis as long as they’re able to keep up with the changes and whatnot. Feels like that game doesn’t have a hard falloff after ~22-24 like LoL seems to have.

Reader
Patreon Donor
Schlag Sweetleaf

DOTeA-BAGGED?

DOTeA BAGGINS.gif
Reader
Utakata

…which also reminds me: When bots start showing up at protests brandishing tiki torches, we should start worrying too. :(

Reader
Sally Bowls

Although, some non-competitive or casual players prefer the lower challenge of PvP instead of PvE.

Mewmew
Reader
Mewmew

“OpenAI was founded to stop the AI from taking over the world.” And to do so, they create powerful AI capable of taking over the world…

Reader
Kickstarter Donor
thalendor

I, for one, welcome our new MOBA-playing AI overlords…

…as long as they stay out of my MMOs. :-)

Reader
Minimalistway
Reader
Minimalistway

Sigh … i wrote so many comments about that last link and deleted them, the only good words i find is: WTH?!

Reader
Utakata

The question becomes did this bot have fun? If the answer is yes, maybe then we should start worrying! o.O

kjempff
Reader
kjempff

The problem is, at that time it is too late to worry or regulate. You can’t undo what is already there and much less reverse the motion. That is why people with foresight are trying to raise awareness now.

Reader
Utakata

…or paranoia. Just saying.

kjempff
Reader
kjempff

You are underestimating the rate of progress in AI technology.. just saying. It is very much time to start laying out some ground rules.. Just to get some kind of general consensus amongst 7bil people can take decades from now and “thinking robots” AI is not waiting several decades.

Reader
Utakata

I know a lot of folks like to throw that word around a lot, but I am not even sure AI is even viable or possible at this point. We still struggling with the term consciousness, let alone trying to define it in machine language.

But between you and me, I am currently more worried about humans in positions of authority with a few walls short of a house (and really bad hair) with their hands near the nuclear trigger switch. That will end this world quicker then I can type “artificial intelligence” if it all goes South. Just saying again.

Reader
Sally Bowls

The latest Tom Stoppard play is about consciousness https://en.wikipedia.org/wiki/The_Hard_Problem
and there are two camps: with better hardware/software, it can be successfully simulated and it is uniquely human.

There is a line I won’t do justice from memory along the lines of “It’s not impressive to come up with a program that wins; what would be impressive is to come up with one that minds losing.”
As a gamer, I only somewhat facetiously say that an AI that minded losing would spawn camp humans to extinction, probably shouting “git gud!”

Minimalistway
Reader
Minimalistway

So when (not if) bots replace large number of people, what will governments do? i’m hoping some are working on this problem before it happens, the idea of work should change, and no … not everyone can or want to be another programmer/designer.

Reader
Schmidt.Capela

A few of the more socialist ones (i.e., where the government is bound to redistribute wealth and support society and individual citizens) are, looking at ideas such as the basic universal income (the government pays everyone a wage that should be enough for a frugal lifestyle).

The ones where neoliberalism ideologies abound, though, seem to be ignoring the question. Which is quite understandable, because most potential solutions to this issue revolve around either strong regulation of automation or a state that provides for those that will lose their jobs, and neither of those options sit well with neoliberal supporters.

Reader
Malcolm Swoboda

“more socialist ones” – like Milton Friedman?

Reader
Schmidt.Capela

The troubling part is that the neural network-based bot learned to beat pros with just a week of training; it was able to beat regular players with just an hour of training.

Now think what this means for all the real jobs that don’t need the worker to be physically present. A “bot” might be able to do the work of a a person, with productivity and quality on par with that of an expert; a single human supervising those bots could do the work of dozens, perhaps hundreds of regular workers, and even that human is only needed until there is enough confidence in the bot to allow it to work unsupervised.

Whether that is scary or not depends on how humanity reacts, but one thing is certain: things will change a lot. The cat is already out of the bag.

Oh, and BTW: does anyone know how the bot interfaced with the game? I ask because if it was using computational vision, that alone is troubling for the efforts in fighting against MMO bots, as a bot using computational vision is nearly indistinguishable from a not very talkative human. And OpenAI is supposed to share the tech they develop with any and everyone, which means it could potentially be used by bot makers and gold sellers.

miol
Reader
miol

As long as you’re cheaper than a replacement bot… /dystopia

Reader
Schmidt.Capela

That is part of the issue, there is no way to be cheaper than the bot. Not when the bot could potentially be free software running on a $100 device, or even an advertisement-supported free service.

We already see this in certain areas. For example, how many translators do you think lost their jobs to machine translation? Machine translation is reliable enough you can let the machine do all the work when the text isn’t meant for customers, and even for text the customers will read you can make do with just an editing pass by an editor that doesn’t even need to understand the original language.

miol
Reader
miol

That’s why I’ve used /dystopia!

Where manufacturing jobs turn people basically into slaves in horrible working conditions, willing to do anything to keep their jobs, because of the constant threat of robots getting cheaper and cheaper! As it’s already a proven method to spur countries against each other in terms of living wage and taxes!

Reader
Kala Mona

As I read there are a lot of cheesy tactics that won’t fool a human opponent but completely wreck the AI. I don’t know DOTA, but it was something like leading the enemy minions through the jungle so your own minions can beat the tower, and dropping the same item repeatedly.

Reader
Annoyed badger

now its seen those tactics, it will learn, it may take a while, but now its been exposed to them, its will learn from it, and thats what it does, you can beat it with cheese, but it will learn the cheese and perfect it.

Reader
Schmidt.Capela

Not that AI, though. It’s a learning bot that will rarely, if ever, do the same mistake twice.

Reader
Kala Mona

Well it was beaten 50 times by mostly the same cheese, so it is definitely not a fast learner. Also the playing field was artificial (1v1, no bottles).

OpenAI bots were defeated atleast 50 times yesterday. from DotA2

I don’t want to belittle this AI, but it seems a bit overhyped to me.

Reader
Schmidt.Capela

Yeah, you’re right, for this version of the bot at least. Looks like the bot devs intentionally left machine learning disabled for those matches, making it unable to learn from its defeats. As a result it didn’t know how to deal with anything it hadn’t come up with before (as the bot had to that point only trained by playing against itself).

On the other hand, by the time info from those matches are added back to the bots and it resumes training, tactics that worked against them, particularly cheesy tactics that worked repeatedly, won’t be nearly as much effective.

Reader
Oleg Chebeneev

Beating a pro isnt a big achievement. AI will always be superior then human in games that require calculations and accuracy. And Dota2 isnt the most complex game either. But the way bot learned to beat pros (by playing against a copy of himself) is interesting

Reader
Armsbend

No one on this website would be able to beat a dota pro in any scenario in any game.

Reader
Patreon Donor
Kickstarter Donor
agemyth ๐Ÿ˜ฉ

Anything that allows Dendi to show up on stage at TI is fine with me.

Cyclone Jack
Reader
Kickstarter Donor
Cyclone Jack

While impressive, it would have been more impressive with a hero other than Shadow Fiend, just due to how Shadow Raze works. What’s really impressive is how well the players did vs the AI. That said, I look forward to the bot AI upgrade next year.

Reader
Armsbend

I am now completely open to the idea of corporations running the world. I used to fear Blade Runner – now it is a much better alternative. Open your robe Musk – I am coming in.

Reader
Melissa McDonald

putting a computer program in a computer sim is the definition of “god mode”. It doesn’t have to miss.

Capture.JPG
Reader
Armsbend

Pros in MOBAs miss in matches maybe 1 out of a 100 times so that isn’t why this is important. The time it took and the casual-ness of it is exciting. So many cool new AI leaps recently.

off-topic – i want your opinions on Rez PC in the next VR article!

wpDiscuz