|
- Mirror match of Necrophos, Sniper, Viper, Crystal Maiden, and Lich.
OK.
- No summons/illusions
OK.
- 5 invulnerable couriers, no exploiting them by scouting or tanking
OK, lol.
- No Scan
OK.
- No invisibility (consumables and relevant items)
Ehh, how about invis rune?
- No warding
- No Roshan
- No Divine Rapier, Bottle, Quelling Blade, Boots of Travel, Tome of Knowledge, Infused Raindrop
IS THIS EVEN DOTA LMAO
AI will win because it will train in these stupid conditions nonstop. While there is no single incentive for humans (especially, pro players) to do so - this is completely different from a 1v1 SF challenge (which is limited by a nature). Basically, in a race of retards it doesn't matter that you win.
If humans simply avoid fights and farm Burning style, they will eventually win too - bots suck both at last-hiting and item builds choosing. But it won't make it into newspapers headlines, so I bet on AI manhandling pure meatbags once again, so researchers can receive a proper funding for next year improvements.
|
How many times does it need to be said? This is why AI will win, because humans can't read.
"While today we play with restrictions, we aim to beat a team of top professionals at The International in August subject only to a limited set of heroes"
|
I think people should read the blog before commenting about the restrictions.
It is a great initiative and gives me chills to think what AIs can achieve in near future.
|
Even without this restrictions people are misinterpreting, I don't see the negative or positive extremes people have expressed. Humans constantly build things that can outperform ourselves in one way or another. Doing so here neither invalidates the efforts of human players nor represents some unimaginable feat.
|
On July 09 2018 08:21 polgas wrote:
How many times does it need to be said? This is why AI will win, because humans can't read.
"While today we play with restrictions, we aim to beat a team of top professionals at The International in August subject only to a limited set of heroes"
Yeah, humans really can't read.
Our team is focused on making our August goal. We don’t know if it will be achievable, but we believe that with hard work (and some luck) we have a real shot.
On July 10 2018 23:22 Kyir wrote:
Even without this restrictions people are misinterpreting, I don't see the negative or positive extremes people have expressed. Humans constantly build things that can outperform ourselves in one way or another. Doing so here neither invalidates the efforts of human players nor represents some unimaginable feat.
Personally, I just think it's boring having humans play by AI-centric rules, rather than forcing the AI to play in environments where they're handicapped.
If the vision game and Roshan are such a massive advantage, I want to see that in action and see how those advantages play out, and how far the AI still has to go.
|
Thank you for being explicit in your quotes.
|
Bot edit.
User was banned for this post.
|
On July 11 2018 03:12 WolfintheSheep wrote:Show nested quote +On July 09 2018 08:21 polgas wrote:
How many times does it need to be said? This is why AI will win, because humans can't read.
"While today we play with restrictions, we aim to beat a team of top professionals at The International in August subject only to a limited set of heroes" Yeah, humans really can't read. Show nested quote +Our team is focused on making our August goal. We don’t know if it will be achievable, but we believe that with hard work (and some luck) we have a real shot. Show nested quote +On July 10 2018 23:22 Kyir wrote:
Even without this restrictions people are misinterpreting, I don't see the negative or positive extremes people have expressed. Humans constantly build things that can outperform ourselves in one way or another. Doing so here neither invalidates the efforts of human players nor represents some unimaginable feat. Personally, I just think it's boring having humans play by AI-centric rules, rather than forcing the AI to play in environments where they're handicapped. If the vision game and Roshan are such a massive advantage, I want to see that in action and see how those advantages play out, and how far the AI still has to go.
Their goal is to have the bot team beat the humans. Meaning, when they play, the only restrictions will be limited set of heroes. The team's goal is to get the bots up to speed to be able to handle things like warding, Rosh, consumables, etc. by August so the AI wins. So unless the team goes back on their word and puts the restrictions back in place because they struggle to implement warding, Rosh, etc. by August, the only limitations are going to be the smaller hero pool.
|
Nobody is suggesting that openAI won't change the rules between now and august, the complaints revolve around the suggestion that what openAI have going right now is "dota".
Its merely a case of "Oh so you say you are going to do X? great, come back when you've actually done it instead of some facsimile thereof that is interesting but not the same"
It's annoying when you have stuff like bill gates tweeting that AI is beating humans at "dota". That's the kind of falsehoods that people are forced to decry.
|
On July 13 2018 02:26 Sn0_Man wrote:
Nobody is suggesting that openAI won't change the rules between now and august, the complaints revolve around the suggestion that what openAI have going right now is "dota".
Its merely a case of "Oh so you say you are going to do X? great, come back when you've actually done it instead of some facsimile thereof that is interesting but not the same"
It's annoying when you have stuff like bill gates tweeting that AI is beating humans at "dota". That's the kind of falsehoods that people are forced to decry.
That's perfectly fine. I even think the "AI bots beat professional Dota 2 players" stuff from last year is totally overblown and when looking below the surface, total bs. There are a lot of comments and people though, who seem to believe the AI bots' match coming up will have the same restrictions as the very first one (no warding, no rosh, etc.). That's who my post is trying to inform and correct, My post is not stating anything about whether openAI is playing "real Dota", just attempting to clarify misinterpretations of what the rules are supposed to be during the upcoming match.
|
Again, the exact quote from the devs is "We don’t know if it will be achievable, but we believe that with hard work (and some luck) we have a real shot."
If they can meet that goal then all the power to them, but they make it pretty clear that removing all but the hero restrictions is a very tough grind.
|
On July 13 2018 03:06 WolfintheSheep wrote:
Again, the exact quote from the devs is "We don’t know if it will be achievable, but we believe that with hard work (and some luck) we have a real shot."
If they can meet that goal then all the power to them, but they make it pretty clear that removing all but the hero restrictions is a very tough grind.
OK, so we're interpreting that quote completely differently. I'm reading the "While today we play with restrictions, we aim to beat a team of top professionals at The International in August subject only to a limited set of heroes." as a statement that they will, no matter what, play against a team of top pros in August with the only restrictions being a limited set of heroes. The "We don’t know if it will be achievable, but we believe that with hard work (and some luck) we have a real shot." to me means they don't know if winning against the pros will be achievable and that's what they're working toward.
Here's a quote from their blog; "While today we play with restrictions, we aim to beat a team of top professionals at The International in August subject only to a limited set of heroes. We may not succeed: Dota 2 is one of the most popular and complex esports games in the world, with creative and motivated professionals who train year-round to earn part of Dota’s annual $40M prize pool (the largest of any esports game)." I view the "not succeed" being "beat a team of top professionals". It's not about what restrictions they'll play with, but whether OpenAI will beat the pros at TI subject only to a limited pool of heroes.
My point is that the July 28 and August games are supposed to be done with only restrictions on the hero pool, but there are many posts who seem to be missing that detail and complaining about the restrictions that were in place when OpenAI first got notoriety. Now, could the devs back out of what they said and put more restrictions back in place (no warding, no Rosh, etc.) if they don't get the bots to a point they feel comfortable with? They could, but I don't think they will (and hope not).
|
You have to remember, this is not a competitive DotA project, it's a learning AI project. I think it was mentioned before, but beating pro players without restrictions and playing without restrictions are one and the same goal. As they say:
Most of the restrictions come from remaining aspects of the game we haven’t integrated yet. Some restrictions, in particular wards and Roshan, are central components of professional-level play. We’re working to add these as soon as possible.
Judging from the graphs they've posted, on average it takes a week for the game simulations to crunch the variables and reach close-to-peak performance. If the AI is still (today) unable to handle something like Roshan, the bottleneck is not because of CPU time, but because they haven't created the right conditions for the AI to learn the importance or value of Roshan yet.
|
Garbage test with so many restrictions. Waste of time. Show me an ACTUAL Dota game with AI vs. a top-tier team.
|
On July 14 2018 04:57 pNRG wrote:
Garbage test with so many restrictions. Waste of time. Show me an ACTUAL Dota game with AI vs. a top-tier team.
Why do you consider it to be garbage? It is standard to consider subproblems of a problem in many areas of mathematics to solve the original problem.
This is not really about dota I think; it's more about getting the experience of designing and implementing an AI in a competetive setting which has real time. It is quite likely that they can export much of it with some redesign to real robots.
|
Absolutely, OpenAI and Deepmind have far greater goals in mind than beating players at Dota or Go. Its training AI to solve our problems.
|
I wonder if they would react erratically if the human team decides to go 5 man deathpush and they cant figure how to deal with such an unorhodox strat
|
Guys we are witnessing the making of next-gen AI.
Emphasis, via italics, on making. It's not perfect yet, it's not even done, but it's still mucho better than last year. Just take it for what it is, imo. No it's not human level yet, and it still maybe needs some restrictions (rather, the team does need them), but it's getting closer.
The product ain't finish yet, but it's dam cool to see the progress they made. At least imo.
Cool and scary. AI should be scary,
|
On July 14 2018 04:49 WolfintheSheep wrote:You have to remember, this is not a competitive DotA project, it's a learning AI project. I think it was mentioned before, but beating pro players without restrictions and playing without restrictions are one and the same goal. As they say: Show nested quote +Most of the restrictions come from remaining aspects of the game we haven’t integrated yet. Some restrictions, in particular wards and Roshan, are central components of professional-level play. We’re working to add these as soon as possible. Judging from the graphs they've posted, on average it takes a week for the game simulations to crunch the variables and reach close-to-peak performance. If the AI is still (today) unable to handle something like Roshan, the bottleneck is not because of CPU time, but because they haven't created the right conditions for the AI to learn the importance or value of Roshan yet.
You mocked people for not reading correctly, when what was pointed out was that the August game against pros will be with only limited heroes as a restriction as opposed to all the people complaining about the upcoming game being a mirror match with a bunch of rules in place (which again, won't be the case during the August match).
They've set a goal for the August match to be able to beat pros with only a limited hero pool as a restriction. I take that to mean no matter what, they'll play that game with wards and Rosh to show where the AI is at, even if it's not operating at optimum capacity. Yes, their end goal no matter what will be for OpenAI to beat pros without any restrictions, no matter what benchmarks (like the August match with just a limited hero pool) they place along the way.
Maybe if they don't feel confident the AI can beat the pros they'll back out or put more restrictions in place. From what they've stated however, the only restrictions in August's match will be limited heroes, even if OpenAI isn't 100% ready for it.
|
On July 16 2018 06:18 Odawg27 wrote:Show nested quote +On July 14 2018 04:49 WolfintheSheep wrote:You have to remember, this is not a competitive DotA project, it's a learning AI project. I think it was mentioned before, but beating pro players without restrictions and playing without restrictions are one and the same goal. As they say: Most of the restrictions come from remaining aspects of the game we haven’t integrated yet. Some restrictions, in particular wards and Roshan, are central components of professional-level play. We’re working to add these as soon as possible. Judging from the graphs they've posted, on average it takes a week for the game simulations to crunch the variables and reach close-to-peak performance. If the AI is still (today) unable to handle something like Roshan, the bottleneck is not because of CPU time, but because they haven't created the right conditions for the AI to learn the importance or value of Roshan yet. You mocked people for not reading correctly, when what was pointed out was that the August game against pros will be with only limited heroes as a restriction as opposed to all the people complaining about the upcoming game being a mirror match with a bunch of rules in place (which again, won't be the case during the August match). They've set a goal for the August match to be able to beat pros with only a limited hero pool as a restriction. I take that to mean no matter what, they'll play that game with wards and Rosh to show where the AI is at, even if it's not operating at optimum capacity. Yes, their end goal no matter what will be for OpenAI to beat pros without any restrictions, no matter what benchmarks (like the August match with just a limited hero pool) they place along the way. Maybe if they don't feel confident the AI can beat the pros they'll back out or put more restrictions in place. From what they've stated however, the only restrictions in August's match will be limited heroes, even if OpenAI isn't 100% ready for it.
They're quite clear that those DotA mechanics are not implemented in the AI at all, not that they're not working optimally. And thus far humans have only been allowed to play the OpenAI with mirrored restrictions (for both the 1v1 bot or 5v5 bots).
But barring a definitive statement, or until August, it's just splitting hairs right now. You say beating pros is the goal, I say playing without restrictions is.
|
|
|
|
|
|
EPT
