Back in 1996 IBM’s Deep Blue was competing and overtaking chess world champions. Later on, in 2011, IBM showcases an even better AI, Watson, who participated in Jeopardy and won. In 2016, Google rises to the “who has the best AI” challenge with AlphaGo. This time, the AI won a game before thought impossible for a machine to win. In fact, not only that AlphaGo won, but it did so against Lee Sedol, a legend of the Go game.
You might not think much of it. but in reality all these victories are key milestones for artificial intelligence and machine learning technologies. But even more impressively, AIs are now challenging humans in complex and dynamic computer games. You could say we are battling AIs in a home match. Exciting, isn’t it?
The future of AI
Today we dive in the future of AI, with OpenAI. Backed by Elon Musk, the OpenAI project is in fact a research initiative for “Discovering and enacting the path to safe artificial general intelligence.” In other words, OpenAI is a movement towards releasing research on AI technologies to the public. Thus, everyone has access to information, not just governments or big corporations. You could say that OpenAI is trying to create the new internet. A democratic place where people and businesses have equal rights and reach to the most complicated algorithms.
How complicated those algorithms can be? Well, the OpenAI team decide to surpase any expectations, so they created OpenAI Five. Much like its predecessors, OpenAI Five was put to the test by competing with humans in a mainstream game. Only this time, the artificial intelligence needed to compete in a much more complicated platform, namely Dona 2.
Practice makes perfect
This may sound like an incredible challenge for an AI, but OpenAI Five is not just an AI. It is a team of neural networks with more than 45.000 years of game play training in Dota 2. Clearly more training than any other Dota team could possibly do in a lifetime, and as a result, OpenAI Five defeated Team OG, the Dota 2 world champions.
The victory is an incredible achievement, especially since the technology uses a rather different algorithm. OpenAI did not learn how to play, instead it learned how to learn. This process is called deep reinforcement learning, and the reason why this is a huge win for AI technologies, is that games like Dota 2 are real-time based. This not only means that the game is not played in turns, but every player is unique and different, and basically impossible to predict with a hundred percent accuracy.
Smart or just fast?
Of course, there are some who think OpenAI won simply because its computing time is marginally faster than a human’s. However, in MOBA games like Dota 2 speed is not as much of a key strategic advantage. In fact, in games like this, your pre-defined strategy, the way you start the game, prepare your character and so on, are by far more important.
OpenAI Five is an undeniable step forward, and the team behind it is not planning on stopping anytime soon. In fact, later this week, they will open the “Arena”, a platform where everyone is welcome to join and interact with the technology. However, this may be your last chance as the OpenFive creators are ending the public demonstrations. Later on, the team plans on creating a software where humans and the AI can collaborate and not simply interact with each other. The endgame here is understanding and applying the technology in other fields like robotics, thus advancing our knowledge even faster.
What is next?
In the end, all we can say is that AI development is moving at a truly accelerated pace. Now more than ever as more people have access to advanced algorithms. Starting with OpenAI and other technologies like it, we are excited to see how companies big and small apply the knowledge to their own field. We are even more excited to see what incredible results we will witness.