clock menu more-arrow no yes

Filed under:

An advanced Shadow Fiend bot just 1v1’d Dendi—and decimated the esports celebrity

New, 1 comment

The match at The International 7 showed off a high-level AI bot that learned from endless hours of Dota 2.

Dendi before The International 7 OpenAI show match.

If you think bots are no match for pro Dota players, think again.

At The International 7, Valve presented a showmatch featuring well-respected and world-famous Natus Vincere mid Danil “Dendi” Ishutin. The game pitted him against a mid high-level Dota 2 bot project, created by artificial intelligence research company OpenAI. The match was a standard 1v1 Shadow Fiend game: two kills, no neutrals, item restrictions and such.

The first game ended with the OpenAI bot far, far ahead in last-hits: Dendi with 14 hits/2 denies and the AI bot with 34/15. The bot didn’t hesitate, either, to kill Dendi 2-0 when it had the opportunity.

Quickly into second game, Dendi gave in after the first kill.

How smart is this bot? According to an on-stage representative for OpenAI, the bot can reach this level of understanding complex tasks in two weeks; it can beat in-game bots in an hour.

“We've created an AI which beats the world's top professionals at 1v1 matches of Dota 2,” says the website. According to the segment beforehand, the bot was designed by studying “a lifetime of 1v1 matches.” It showed players getting upset after losing to a series of professional players.

We’re not actually sure of the full extent of the on-stage bot’s abilities. (Players were invited to challenge it at a nearby LAN event for a chance at Shadow Fiend arcanas.) The parameters were pretty strict, and they seem to be focused on Shadow Fiend games. Still, it’s a pretty fascinating concept, and the show match had everyone on the edges of their seats.

OpenAI describes itself as a “a non-profit AI research company, discovering and enacting the path to safe artificial general intelligence.” A representative on stage described how they wanted to use the technology to do “messy” tasks that humans usually find difficult.

UPDATE: The group has released a blog outlining the project a bit further.