AI Defeats Human – Arago Achieves Groundbreaking Advancement in Machine Intelligence

“How can we show what AI is really capable of doing?” – An idea that came up randomly during one of Arago’s company-BBQs turned into an exponential advancement in machine intelligence. After months of preparation, long nights and endless coding, Arago’s CEO Chris was finally ready to make the big announcement during this year’s TechCrunch Disrupt Event in London.

HIRO™, the company’s autonomic problem-solving AI has now defeated human players after it became an unbeatable opponent for the game’s built-in AI.

Why this is extraordinary news

Well, a couple months back, Facebook claimed that it would take 5+ years before AIs would be capable of playing such strategy games – however, Arago was able to achieve it within six months!

A group of seven enthusiastic Aragons teamed up to train the AI and feed it with aggregated expert knowledge they had received from passionate Freeciv players worldwide. Step by step, they created a SuperHIRO with exceptional capabilities of making strategic decisions and reacting to situations autonomically to beat its opponents.

While Chris was speaking to TechCrunch’s editor John Biggs about his company’s development over the past years, footage of a live Freeciv game was screened onstage illustrating the complexity of the game. The challenging factors about playing Freeciv, regardless of whether the player is human or an AI, is to win the game with the limited information you have on hand with regard to your enemy and the immediate vicinity. You have no god-like view of everything around you – you can only do and react to whatever is visible within your range.

In other words, you have:

  • restricted information of your surroundings (compared to Go or Chess which reveals its complete environment to the player)
  • more than 1015.000 possible games imaginable in Freeciv (for comparison there are only 1080 atoms in the universe)
  • the choice of approx. 2400 moves each turn (that is almost 10x more than Go)
  • opponents acting unpredictably and changing parameters throughout the whole game

So, similar to how a human brain works, HIRO™ was taught to understand and interconnect words, therefore enabling the recombination of experience to grow exponentially. For an empty machine with initially zero knowledge to not only learn but also win such a complex game against respectable expert Freeciv players, is therefore an extraordinary progress in technology. As Chris puts it:

“This is the main difference between machine learning and machine reasoning. You could say machine learning is brute force trial and error while looking at something, and machine reasoning is only talking about it. So we combined the two to make it happen.”

Watch the entire panel discussion to learn more about Arago’s AI solution and how it learned to defeat human players in Freeciv.

Arago Redaktion 10. February 2017