Published in


Revenge of the Nerds! Humans Beat Bots at Dota 2 International

Last August at the Dota 2 International tournament in Seattle, OpenAI introduced an AI bot that upset the world’s top 1v1 human player. The San Francisco-based AI research institute is now at the International 2018 in Vancouver, where their team of state-of-the-art bots is battling professional human teams in a highly anticipated best-of-three 5v5 Dota 2 showdown.

Alas, the humans drew first blood: In a match that would make John Connor proud, Brazilian pro team “paiN” dispatched the “OpenAI Five” Bots yesterday in 52 minutes.

Dota 2 at Rogers Arena in Vancouver

OpenAI Five had more kills and a slight economy edge in the midgame, but did not push their advantage and kept losing their towers. The smart bots also made some dumb moves, such as warding in the wrong positions, bad item choice, and fewer gankings (leaving your lane to kill an enemy Hero in another lane).

OpenAI Research Scientist Jonathan Raiman told Synced he was not shocked by the results. “Before the game, most of the team were pretty conservative and believed we might have only a 30 percent to 40 percent chance to win.”

The OpenAI Five performance was nonetheless impressive: the AI bots lasted longer than in a usual game; had more kills than the human team (45: 41); and won most teamfights, this attributed to their error-free micro-level control.

OpenAI CTO Greg Brockman tweeted after the match, “Lots of extremely exciting plays by both teams. Has been a great showcase of what both humans and AIs can do.”

The Match rules were changed from normal human competitions as OpenAI has not yet integrated some aspects of Dota 2 gameplay into its bots. Certain items (divine rapier, bottle) were eliminated; the number of heroes was reduced from 115 to just 18; techniques such as scams and illusions were forbidden; and the number of couriers was increased from one to five. On the eve of the competition the courier rule change was reversed to make the contest more like a human game. Raiman speculates the AI bots might not have fully adapted to that change.

Raiman says there are a couple of things they could have done better. He is concerned for example that OpenAI Five’s objective prioritization — which puts long-term goals ahead of short-term reward — might have somehow sacrificed the bots’ acquisition of gold in the early game.

Launched in 2013, Dota 2 is a highly complex and wildly popular multiplayer online battle arena (MOBA) video game played between two teams of five players. The team that takes down their opponent’s center base “Ancient” wins the game.

Dota 2 player interface

AI has seen tremendous breakthroughs in board games such as Go and Chess, but complex strategy MOBA video games extend the challenges onto entire simulated worlds with highly detailed environments. Dota is an incomplete game where a large portion of the information is unknown. The game is variable and unpredictable with multiple components such as heroes, items, NPCs, towers, skills, etc. taken into account. Compared to other popular games like StarCraft or WarCraft, Dota 2 is more highly team-oriented: a single strong player cannot expect to carry their team to victory.

As Brockman told VentureBeat, “Dota is this really [complicated] task where you have to deal with these long time horizons in a very continuous state.”

OpenAI researchers trained their AI bots to play from scratch without any human data or gameplay guidance. Each bot’s brain is built on a Long Short-Term Memory (LSTM) neural network, which are extremely good at remembering information for extended periods of time — an essential capability for either humans or AI mastering Dota.

Dota 2 Interactive Demonstration: How AI bots “see” the game

The bot improves by playing against itself and its past iterations as Reinforcement Learning algorithms provide rewards for actions that lead to victory. The bot can train on the equivalent of 180 years of nonstop human gaming in a single day.

Earlier this month OpenAI Five swept a best-of-three series against a team of 99.95th percentile Dota players, four of whom had played professionally. The humans only lasted 20 to 25 minutes before calling GG (good game) in surrender.

Stanford AI Researcher Andrey Kurenkov tweeted, “OpenAI focused on showing that complex coordination and long horizon play was possible, and did that.”

OpenAI is going to play another two games this week at The International. OpenAI researchers told Synced they are also planning to have the bots play a fully unrestricted Dota 2 match later this year or early next year.

The eighth Dota 2 International features 18 teams and has attracted gamers from around the world. It runs to August 25 at the Rogers Arena in Vancouver. The total crowdfunded price purse is an eSports record US$25 million.

Will OpenAI Five fire back and snatch victory from the humans? Synced is covering the story and will update readers with the latest news from Vancouver.

Journalist: Tony Peng | Editor: Michael Sarazen

Follow us on Twitter @Synced_Global for more AI updates!

Subscribe to Synced Global AI Weekly to get insightful tech news, reviews and analysis! Click here !

We are honored to have Fisher Yu , the post-doctoral researcher at UC Berkeley, as our guest speaker in DTalk E4 “@UCBerkeley DeepDrive and Its Driving Data Efforts”. Sign up at and learn about their recent achievements of self-driving technologies!




We produce professional, authoritative, and thought-provoking content relating to artificial intelligence, machine intelligence, emerging technologies and industrial insights.

Recommended from Medium

Client Loyal AI

The Future of Portfolio Strategies

thinking, brainstorm, business-4035982.jpg

USC & Amazon ‘SLADE’ Self-Training Framework Uses Unlabelled Data to Improve Information Retrieval

“How AI is revolutionizing the Digital Publishing Industry?”

MHCAttnNet: Predicting MHC-Peptide Bindings for MHC Alleles Classes I and II Using an…

ReWork Deep Learning in Finance Summit

The benefits of chatbots in customer service roles.

Google on a Review Takedown Tear

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store


AI Technology & Industry Review — | Newsletter: | Share My Research | Twitter: @Synced_Global

More from Medium

Cash App Labs Modifies the Very Deep VAE to Achieve a 2.6x Speedup and 20x Memory Reduction

How Microsoft & OpenAI Are Squeezing the Best Out of GPT-3

Inside Meta’s New Architecture for Build AI Agents that Can Reason Like Humans and Animals

Multimodal AI Modeling is the Future, But It’s Also Pandora’s Box