Artificial intelligence (AI) technology have made strides over the past decade. Though tech experts and entrepreneurs had varying opinions about the role of AI in the progress of mankind, this technology will be shaping the future. It has been challenging humans and beating them in the tasks that seem impossible for machines. Undoubtedly, the extensive research is involved in training the machines to think in the way humans think and making them capable of beating humans in various tasks. Gaming has become commonplace for research activities and to test whether it is possible for AI bots to beat humans. However, the underlying aim to determine if the AI system can be used to solve the real-life problems. Google’s AlphaGo AI had beaten the best human Go player in the world. Other companies have also been trying to reach such milestone. However, the games are different, but the objective is same: To test whether AI can solve real-life problems.
This time, OpenAI, the AI research firm co-founded by tech billionaire Elon Musk, has announced its recent milestone. It states that a team of top AI agents can defeat 1% of amateur Dota 2 players. From last August, the firm has been trying to capture the world of Dota 2 with the help of its trained systems. It released a system that could defeat top players in a 1v1 match. However, there is a less challenge in this type of game. Now, it has released the system that plays against humans in a 5v5 match. The level of algorithms and training required is complex as long-term planning and coordination are required. The system has not yet challenged the best players of Dota 2. It will challenge at The International Dota 2 tournament this year. This event is the biggest e-sports event organized annually.
The objective of this research was to determine if AI systems can solve the complex real-life problems that resemble video games. “This an exciting milestone, and it’s really because it’s about transitioning to real-life applications,” told Greg Brockman, OpenAI’s co-founder and CTO, to The Verge. “If you’ve got a simulation [of a problem] and you can run it large enough scale, there’s no barrier to what you can do with this.”
Open AI used the reinforced learning approach in which AI bots are faced with a virtual environment in which they need to fulfill their objectives with the help of trial and error method. Researchers offered “reward functions” in which they gave away points for AI bots for killing an enemy. Then, they left them to play with themselves again and again.
OpenAI Five operates on more than 100,000 CPUs. The neural networks installed in the system performs nearly 150–170 actions per minute. The mean reaction time is 80ms. This time is faster than human beings. In few hours, the bots could play more games than a human play in the lifetime. These statistics showed that the AI system plays nearly 180 years’ worth of games against itself each day. This amount of play offers a lot of opportunities to learn. Though the crucial aim is not to beat human players, it is to determine if AI can be employed to solve the real-world problems. It is interesting to see how the firm will proceed in developing its AI systems. For now, we will see if those AI bots can beat the best Dota 2 players.