A Brave New OpenAI

It wasn’t that long ago that OpenAI, the company behind the scary Shadow Fiend bot that destroyed the best players in the world during The International 2017, had announced its preparations to take on an entire team of human players at The International 2018 in an attempt to push the limits of their project and break new ground.

Initial reaction

However, not everyone in the community was excited about the specific details of their stated goal. And some, including myself, were very dissatisfied with the limitations imposted upon the Humans vs. Machines match-up, feeling that it favored the AI way too much for the outcome to be of any relevance. Read this recent opinion piece for more details.

Plot twist

This week though, the story changed completely as OpenAI announced a massive and unexpected plot twist: the rules under which the Humans vs. Machines match will take place have been radically changed. In other words, the AI will no longer have the high ground right from the start and force the humans to fight against it while facing the Sun. The battle will be much fairer and its outcome will actually speak volumes about how the AI currently performs and what it can do against a solid team of human players.

The old rules and the new

Initially, the main limitation that was imposed upon the conditions in which the Humans vs. Machines confrontation was to take place had to do with the hero pool. Only 5 heroes had been included and the two teams would have had to play a mirror match. The fact that these 5 heroes were super basic (Necrophos, Sniper, Viper, Cryistal Maiden and Lich) and didn’t allow for flashy plays, meant that the AI had a huge advantage. Practically, the human team was likely going to play its first match ever using this draft while the AI would have played countless matches up until that point in the exact same formula.

Now, the hero pool has been greatly extended to include 18 heroes, and the confrontation will no longer be a mirror match-up. This means that humans will actually be allowed to exploit draft weaknesses in AI’s pre-match choices and come up with clever strategies of their own.

Furthermore, a number of other limitations have been lifted, including the use of wards and various other items, the existence of Roshan on the map and the possibility of heroes becoming invisible.

Team humanity and match details

For its next and likely final battle before The International 2018, OpenAI will throw its scary creation in a cage with a team of 5 solid players, 4 of which are high MMR (99.95 percentile), famous analysts: Blitz, Cap, Fogged and Merlini (who’s actually a former Dota 2 analyst).

The battle will be a Bo3 series scheduled to take place on August 5th at 12:30 PM (Pacific Time) in San Francisco. The games will be streamed on OpenAI’s Twitch channel and casted by Purge and ODPixel.

To be honest with you, I’m more hyped for this match than for TI itself, as its outcome could potentially be a game changer. If the AI loses, fine. And it should lose under these new conditions. But if it wins, then OMG! I mean, Dota 2 is a very, very, very complex game to master. And having an AI system kick our wise human asses after such a short period of training would be a scary result indeed.

Comments are closed.