OpenAI stomped first - but there is still hope for the human race

posted by Smyke,
Since The International 2017 we had been waiting for the ultimate fight between the human race and the robots - a 5v5. The OpenAI Five Benchmark match ended with an easy win for the robots, but Merlini and Co. didn't embarass themselves. It was a long way from the 1v1 win on Shadowfiend against Danil 'Dendi' Ishutin to the duel against the four talents. Ben 'Merlini' Wu, Ioannis 'Fogged' Loucas, William 'Blitz' Lee and Austin 'TheCapitalist' Walsh as well as pro player David 'Moonmeander' Tan faced off against OpenAI in a 5v5 in which the bots proved that they were able to improve their skills.

A few rules were set beforehand to give OpenAI a fair chance to win against the humans:

  • The hero pool included only 18 heroes
  • The items Divine Rapier and Bottle were not allowed
  • No summons or illusions
  • The game included five invulnerable couriers
  • No scans allowed


OpenAI pushes too hard



When the hype match between the human team and the OpenAI finally started the bots turned out to be relentless.



The humans did not have any time to prepare and adapt to the playstyle of OpenAI. OpenAI rushed through their enemies' bottom racks and won the fight on the midlane - the "gg" was called after only 21 minutes. Merlini and Co. were still able to get an impressive eight kills in the game. The bots pushed fast and their teamfight potential was, as expected, on point - every mistake the humans made next to the tower was punished harshly.

The human team used the short break between games one and two to switch up their strategy a bit. They picked Death Prophet and used Riki's invisible skills, because the warding of the OpenAI bots seemed to be the big weakness in their strategy - and it turned out to be the right approach: dewarding and wombo combos initiated by Riki became a problem for the bots.



However, the bots consistently punished every single mistake made by the humans. Game two became way closer than the first one, but in the end OpenAI's teamfight potential was too strong for the human lineup. OpenAI pushed the bottom racks in the 23rd minute and killed enemy after enemy - the bots won the series, but the humans got another chance to at least win one game.

The draft helped the human race to win



Cap and Co. switched up their hero roster again to find the real weakness of the bots. After dewarding and invisibility had not been enough to defeat them, the humans tried the lineup that OpenAI Five used in their first game, except they switched Sniper with Necrophos. The side of the bots was a surprisingly weak one.

No real support combined with mid to late game heroes, which need items to snowball their enemies - OpenAI finally showed a weak point in their draft (Thanks to the Twitch chat). Based on that pick and ban, OpenAI predicted a 2,9 % chance to win against the humans.



OpenAI Five got a quick start by killing a lot of their opponents, before the humans finally fought their way back into the game with great teamfights and huge picks against the underperforming carries. Although the humans had the way better draft, OpenAI brought them to the edge of losing - the game took more than 35 minutes.



However, in the end the humans finally did it! After analyzing the playstyle of the OpenAI Five, the humans were able to not just win some teamfights, but actually take down the opponent's Ancient!

They won against the OpenAI bots defending the honour of not just the entire Dota 2 community, but the human species as a whole.

What do you think will happen, when the real pro players will face against OpenAI next time?



Photo credit: Twitch.tv/openai

Comments