- The second edition of the ‘AI World Cup’ was held in Korea last week with 24 teams from 12 countries participating
- The winning team used an algorithm based on Q-Learning, a reinforcement learning technique
- The team simulated three million different steps to design their final winning algorithm
24 teams. 12 countries. 1 trophy. This was the 2nd edition of the ‘AI World Cup’, a tournament that combines the power of artificial intelligence and football. And it was as intriguing as it sounds!
The tournament was hosted by the Korea Institute of Science and Technology (Kaist) last week and saw participation from around the globe. The hosts, Kaist, ran out winners in the end, seeing off competition from the big giants like Google and MIT (according to this report).
Below is a high-level overview of how the tournament was structured:
- Each team had 5 players
- Each player was cube shaped and had been programmed with algorithms
- Each half was of five minutes
This was reinforcement learning and neural networks being put to the team on a virtual football field, and it passed the intense examination with flying colors. According to the winning team’s leader, they used an algorithm based on Q-Learning. It was designed to maximize rewards if the ‘agent’ (in this case the virtual player) took the right action, and correct itself if it did not. If you want to learn more about reinforcement learning and and how to implement it, check out this beginner-friendly article.
In their particular algorithm, the player could take five steps in a second. The team simulated three million different steps to get the ‘agent’ to understand the environment and optimize the final results. The tournament even had journalists and commentators powered by AI!
Intrigued? Now watch the below video from last year’s event to understand how the virtual players were trained using neural networks. The video contains different cases which further illustrate how crucial these algorithms are to clinching the ‘AI World Cup’:
Our take on this
This is quite an entertaining use of AI, wouldn’t you agree? Not even single breakthrough or development has to be groundbreaking, it’s refreshing to see the light side of AI once in a while. Being a huge advocate from AI and ML in sports, it was a joy covering this world cup event here.
However, it adds to the belief that researchers are not quite able to find a practical use of reinforcement learning algorithms outside a simulation environment. We seem to still be quite a while away from that scenario. Watch this space for more!
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!