|
Archive
+ Show Spoiler +We already saw from last year that the AI is much better 1-on-1 against humans, but what about 5v5? Dendi vs. OpenAI Note that they will reduce the hero pool during this challenge. Since people are complaining about current restrictions (no warding, no rosh, etc.), repeat: They will only restrict the heroes in this challenge in August.
"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" Poll: Will OpenAI beat 5 professionals in a team?Yes (122) 47% No (136) 53% 258 total votes Your vote: Will OpenAI beat 5 professionals in a team? (Vote): Yes
(Vote): No
https://blog.openai.com/openai-five/
Important dates:
August 5, 12:30pm Pacific: OpenAI Five Benchmark match, streamed on Twitch.
https://blog.openai.com/openai-five-benchmark/
 OpenAI Five Benchmark
https://blog.openai.com/openai-five-benchmark-results/
August 22, 23, 24: Official matches against professionals. https://www.twitch.tv/dota2ti
Will be held after the conclusion of the 1pm matches on Wed/Thu/Fri. Probably around 3:30-4pm, some variance.
Game 1: OpenAI vs. Pain Gaming. https://www.twitch.tv/videos/300704237
Game 2: OpenAI vs. Big God. https://www.twitch.tv/videos/300907006?t=07h31m30s
Game 3: Aug 24
The International/2018/OpenAI Showmatches
https://blog.openai.com/the-international-2018-results/
TBD: Match against TI8 winner
https://openai.com/five/
|
Quite a bit of DotA comes down to the draft. Much of the rest comes down to things like communication and team cohesion.
If they want this to be a true test, they should give the human team a better draft, of heros they are familiar with, and choose a current team that is performing well and will take the task seriously instead of an 'allstar' team that won't play well together and will treat it as a random pub.
I'm interested in seeing the results regardless.
|
I'm honestly not interested in this until these restrictions are gone:
- Mirror match of Necrophos, Sniper, Viper, Crystal Maiden, and Lich
- No warding
- No Roshan
- No invisibility (consumables and relevant items)
- No summons/illusions
- No Divine Rapier, Bottle, Quelling Blade, Boots of Travel, Tome of Knowledge, Infused Raindrop
- 5 invulnerable couriers, no exploiting them by scouting or tanking
- No Scan
I don't want a limited environment that the AI are equipped to handle. Far more interested in seeing the AI getting crushed until it's reached a point where they can excel at the real game.
|
I voted yes because I think the rules are in favor of the AI.
|
Even with those limitations, I don't think it's possible at the current moment, if it's an actual team (ie Liquid or Fnatic or something. If it's like a Na'vi reunion, I'll be kinda confused).
|
If they’re putting a restriction like a smaller hero pool I think OpenAI will win
|
My money's on no but it's hard to guess with these things.
On June 29 2018 08:33 WolfintheSheep wrote:I'm honestly not interested in this until these restrictions are gone: Show nested quote +
- Mirror match of Necrophos, Sniper, Viper, Crystal Maiden, and Lich
- No warding
- No Roshan
- No invisibility (consumables and relevant items)
- No summons/illusions
- No Divine Rapier, Bottle, Quelling Blade, Boots of Travel, Tome of Knowledge, Infused Raindrop
- 5 invulnerable couriers, no exploiting them by scouting or tanking
- No Scan
I don't want a limited environment that the AI are equipped to handle. Far more interested in seeing the AI getting crushed until it's reached a point where they can excel at the real game.
Yeah I'm in agreement. Especially when the heroes they picked are five of the lowest skill ceiling/highest skill floor heroes in Dota.
|
The laning phase will favor AI. Teamfight will likely be AI win as well with superior coordination. It's the midgame that may be a slight toss up. Let's see if human ingenuity can beat the odds.
|
thats a pretty set of huge restrictions for the real people
edit: like no warding? is this even dota?
|
Yeah too many restrictions for me to be that interested honestly. I'd rather see the bots be coerced towards a strategy that they can execute and see if human players can identify it and respond in time. No wards / rosh / smoke and fixed lineup really just glosses over a lot of what makes dota a beautiful game.
|
1001 YEARS KESPAJAIL22272 Posts
the restrictions are stupid
|
Yeah these rules are trying to change Dota into a pure mechanics game rather than a teamwork/strategy game.
|
"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"
https://blog.openai.com/openai-five/
Guys the plan is to only restrict the heroes in August.
|
I believe the humans will get absolutely crushed in lane, you guys talk about draft and restrictions and stuff but at the core level dota is about microing your one unit and the bots will be... overwhelming.
I think this can be a bit disappointing actually, in that the match will be "over" very fast, instead of showcasing "a great machine mastermind" level of decision making, coordination, and itemization, which would come later in the game and require a somewhat equal footing.
But still funny when the punny humans will be reduced to clowny strats to try to overcome the unfair and uncaring opposition, in the laning phase.
Well, my 2 cents, maybe I'm wrong. Very excited and curious to see how it plays out, at any rate :-)
edit : my deepest sympathies to the guy(s) who will lane vs an OpenAI Sniper. Yummy.
|
@OP
I don't know if it's worth mentionning, but Bill Gates actually took notice of the event :
https://twitter.com/BillGates/status/1011752221376036864
edit : Oh, and to further comment on the post above :
"OpenAI Five plays 180 years worth of games against itself every day, learning via self-play"
HahahahahahaHAHAHAHAHAH - laugh with me while we still can, boys
|
Laning is probably even more relevant because of the selected 5 heroes and it being a mirror match. These heroes don't exactly make it easy to outmaneuver the enemy if you are behind. They are not mobile, they are pretty bad farmers, they have a really hard time initiating fights and instead have to just walk in. All the skill and item builds are also hard coded, which isn't nearly as viable if you had to consider the strengths and weaknesses of your lineup vs the enemy.
What I suspect the bots will do is just a pretty 5 man heavy strategy. It would be quite enlightening to see the bots play against themselves to see what sort of strategical understanding they've gained or if they just basically do the same thing and whoever does it a bit better wins.
Some people who saw the bots live seemed to claim that the bots were not actually that good last hitters and so on, and instead they beat who they beat by grouping up and fighting together better than the humans. Having laning weaknesses would be pretty bad against top level players, but I'd guess the bots have learned to be better in a month. There may be something in the way their reward functions are shaped that they don't learn super efficient laning easily. But if they have trouble, I'd guess the research team will again try to introduce some sort of semi-manual solutions like they did with creep blocking last year. They explicitly made the bot to learn to do that rather than it realizing doing that brings an advantage in terms of getting the other rewards it wants 20 seconds later.
Another thing is that imo humans should have the ability to see the bots play before a benchmark happens. Dota can be very snowbally, and if you don't have any understanding of where the enemy is really good, it can be hard to come back from a screw up. Would have been great if they released footage of entire games with this press release rather than carefully selected clips.
I'm curious to see how this sort of approach scales for them and what sort of ways they will find to expand to the rest of the game. I saw some machine learning people estimate that in terms of computational resources it costs around 1.1 million dollars to run their training configuration for 19 days (the time they trained the bots to beat whoever they beat this far).
|
On June 29 2018 16:47 spudde123 wrote:I'm curious to see how this sort of approach scales for them and what sort of ways they will find to expand to the rest of the game. I saw some machine learning people estimate that in terms of computational resources it costs around 1.1 million dollars to run their training configuration for 19 days (the time they trained the bots to beat whoever they beat this far).
Wow.
Now THAT, at least in my mind, would make an actually compelling community goal for a future TI prizepool.
Like, if we get to 30M, then we can give X amount to openAI and help them run their trainings for that much longer, and hopefully get the program ready to the real deal, aka no restrictions.
Well... ok, the current reward/cost for Valve is handing out 10 virtual levels, so I guess... they might not love the idea. Would be compelling to me, anyway 
|
I voted "no" but then 2 posts down from OP's I saw the list of restrictions and I'm feeling hesitant now >_>;
And I fully agree that I am not that really interested in this match or its results until these restrictions will be gone. Pidgeon-holing players into that specific of a scenario and handicapping them seems to make this match pointless.
Why not make the human just play with their feet?
Why not pick from the attendees and audience 1K scrubs heralds, get them drunk, and let them play vs AI without restrictions?
edit: shit I'm sorry It's 2018 and 1K scrubs isn't politically correct anymore. It's heralds now.
|
The reason for the restrictions are there to help the AI to learn 5v5 faster. Unlike humans, the AI doesnt have foresight so in order to get better, it needs to try every little thing until it learns. Humans dont need 150+ years to learn Dota because we have foresight, we dont need to try stupid things like suicide on the tower (although some do for different reasons lol) to know that's not gonna give you an advantage. Last year it was 1v1, this year a heavily restricted 5v5, but I expect it to be on human level in a few years. The one thing I'm mostly interested in, how is the AI going to respond to major balance changes and how long will it take it to adjust.
|
I'd personally enjoy this a lot more if there were no restrictions. Yes, no one would expect the bots to win in a straight up 5v5, but it would show how far the AI has come from the previous year. Even ask the pros to go easy on the bots, or event some silly challenges like win mid as CM for funsies.
This just feels like monkeys and typewriters.
|
|
|
|
|
|
EPT
