A next-generation OpenAI Gym with adversarial environments and active competitions.
What we’re building over at Coder One to make AI/ML more accessible to all. Launching late 2021.
In March 2016, DeepMind’s AlphaGo beat Lee Sedol 4–1 in a televised match viewed by over 200 million people. There was a global shortage in Go boards, and AlphaGo’s victory is seen as a landmark moment for artificial intelligence.
Shortly after in April, OpenAI launched its Gym toolkit to help researchers develop and benchmark reinforcement learning algorithms. Reinforcement learning was a key technique used in the training of AlphaGo, along with deep learning.
Today, interest in artificial intelligence and machine learning is at an all-time high.
With this in mind, we started Coder One with the mission to help engineers, data scientists, and researchers interested in AI/ML access all the tools you need to build game-changing AI.
About us
We’re a global team with experience across cloud computing, startups, business strategy, esports, and game-playing AI research.
Over the past 7 months, we’ve launched two iterations of the AI Sports Challenge. This was an AI game competition where a total of 60+ teams across 20+ countries built the best agent to compete in a recreation of the classic Bomberman game for 10,000+ AUD in cash prizes.
Something we heard from participants of the challenge was:
“I’ve been wanting to apply reinforcement learning, and this competition seemed like a good place to do so.”
Participants wanted to learn and apply novel technologies. Some wanted to live out their AlphaGo dreams.
That insight is what led us here. We’re now focused on building the best place for engineers, data scientists, and researchers to experiment with cutting-edge techniques in the broad field of artificial intelligence and get recognized for pushing the boundaries.
Coder One is a place to experiment with novel AI algorithms in competitive multiplayer games
Here are our three main goals for the platform:
1. Have the most challenging and immersive environments that will encourage you to explore creative solutions
After running the AI Sports Challenge twice and speaking to 60+ participants, we’ve got a good idea of how to do this. Our philosophy on game environments:
- Entertaining and intuitive, with high ceilings.
- Quality, over quantity.
- Address real-world and research challenges (partial observability, sparse rewards, multi-agents, etc.).
- Adversarial. These bring back the ‘human’ element and give way to interesting metas. Unlike typical data science competitions where there’s a single accuracy score to optimize, it’s not always clear what you need to do to win in an adversarial environment.
2. Build the most user-friendly tools and frameworks
Start in minutes, not hours. Our platform is language-agnostic. We plan to provide starter kits provided for popular languages used across both industry and research. This includes Python, Java, and C++. Any other languages you’re interested in? Let us know in the comments.
We’ve already built out the infrastructure that will help us scale from hundreds of games to thousands and beyond, giving you plenty of data to train with and improve from.
3. Have year-round high-stakes competitions
We know it’s mostly about learning and the satisfaction of seeing your agents come to life. But that doesn’t mean it can’t be about prizes and prestige as well. From our previous experiences as competition organizers, we know how to raise funding for large prize pools, bring onboard exciting sponsors, and run entertaining Twitch broadcasts to showcase your creations.
What’s next?
We’re currently in private Beta and plan to launch in late 2021.
If you’re interested in a place to apply novel AI/ML algorithms and benchmark them against the competition, you can sign up for our waitlist (researchers can skip the queue).
Follow our journey on Twitter
Join our community on Discord
Got any questions or feedback? We’d love to hear from you.
