Dota 2 is not an easy game to learn let alone master. But, for the hundreds if not thousands of professional Dota 2 players from all over the globe...

AW by Asura World in Apr 15, 2019
Asura World Highlights DOTA2
Dota 2 is not an easy game to learn let alone master. But, for the hundreds if not thousands of professional Dota 2 players from all over the globe, that is their lifelong goal. While no one has yet to fully master the game, it seems that today’s crop of professional Dota 2 players will have to make way for artificial intelligence as their latest competition.

On Saturday, April 13, OpenAI’s very own algorithm defeated the defending The International 2018 champions 2-0 at a showmatch in San Francisco. The first of the two games were rather close, as the artificial intelligence team developed by OpenAI took a little over 40 minutes to beat the humans. However, the second game was a bit more lopsided. The humans didn’t even take more than 20 minutes to concede in what was an absolute stomp right from the get-go.

Ultimately, however, trouncing humans in video games is not OpenAI’s goal. Instead, it’s a mere step in their quest to create an “artificial general intelligence.”

Otherwise abbreviated as AGI, the idea behind their goal is to create an algorithm that performs very similarly to human beings. This means that they want to create something that can learn to adjust to things on the fly. This is an ability that the AI of today does not possess. Or, at least, not yet anyway. According to Open AI, it took them 10 months, or roughly around 45,000 years worth of watches, to train their algorithm to beat OG.

To accomplish this, Open AI sources data centers from various cloud providers to help provide them with the necessary computing power to train their own algorithms.

In a way, Open AI’s win on Saturday was more of a proof of concept, and far from their ultimate goal.

OpenAI researcher Filip Wolski believes in putting the algorithm to work in games that are more complicated. This includes games that require players to control more than six characters. Wolski also believes that they can streamline the training process, shortening it from 10 months to a month, or even a week, or even less. That’s something they’d like to do as they work on building something that closely resembles AGI more.

In addition to making the algorithms capable of doing more with less preparation, OpenAI co-founder Sam Altman wants to work on making the AI require much less computing power and lower the costs of training. They also want to make the algorithms do just as well outside of a controlled environment. This is very important for various applications of the algorithm. Like, for example, in organic chemistry, where the simulations are far from perfect or ideal and the AI will have to learn to adapt to the many possible molecular reactions that can happen inside the human body.
Share Article:

Similar Articles


Leave a Reply