Arago teaches an AI to play games, the better to manage IT systems
When Arago taught its Hiro IT automation system to play a Civilization-like strategy game, it learned how to make IT more fun
By Peter Sayer
If an AI could rule a world, would you trust it to manage your IT systems? German software company Arago is hoping you will.
The developer of IT automation system Hiro (short for Human Intelligence Robotically Optimized) has been teaching its software how to play Freeciv, an open source computer strategy game inspired by Sid Meier’s Civilization series of games, and in the process is learning to make IT management more fun.
Hiro is an AI-based automation system that usually sits on top of other IT service management tools. Unlike script-based systems, it learns from its users how best to manage a company’s IT systems.
And, it turns out, it can also learn to play strategy games. After besting Freeciv’s built-in computerized opponents, Hiro has now beaten its first human adversaries, its developers said Tuesday.
To do that, Hiro used a combination of rules-based and machine learning. That is, a human had to painstakingly teach it how to solve a series of tiny problems, and it then figured out how to combine many of those tiny steps into an efficient overall solution.
The Arago team came up with the idea of having Hiro play Freeciv over the summer and recruited a number of players from the gaming community to teach it their strategy. That meant sitting down with the computer and explaining everything from the first move on, teaching it the words necessary to comprehend their strategy.
“This is the main difference between machine learning and machine reasoning,” according to Arago founder and CEO Chris Boos. “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.”
Arago’s approach means that its AI isn’t quite the black box that others can be, as it’s possible to identify which rule was activated in a particular circumstance — and even who taught it that rule, and why.
The Freeciv player and the sysadmin versions of Hiro have a lot in common, according to Boos.
“It’s exactly the same software doing it,” he said. “We even store the machine learning data in exactly the same graph.”
Apart from creating a robot sysadmin that could play Freeciv with its workmates when off duty, the project does have another benefit for IT staff.
Teaching a computer everything you know about a task, one little rule at a time, can be pretty boring, something the Arago team had to consider when they began working with gamers.
“If it’s not fun to use the machine, they’ll quit,” Boos said.
That led the team to reconsider the feedback cycle on the learning interface, and look for ways to speed things up.
“We’ll keep this up just to make the machine more fun to interact with,” he said.
Making work more interesting is what AI is for, according to Boos. “I believe AI should be an augmentation of people. You should be able to teach the machine the stuff you have to do every day, and then go off and do new stuff,” he said.
As for the Freeciv players who helped with the project, there’s more fun in store for them, too.
Arago organized the training program for Hiro as a tournament, with the AIs playing against one another on a variety of maps. You can check out the leader board at the Hiro website.
“After we’re done with the tournament, we will set up a machine they can play against, and even choose which players’ knowledge they want HIRO to use,” Boos said.
There’s also the satisfaction of playing, and perhaps beating, a computer that has only the same access to the game that they do. “There’s no god-like access, no background information, only what a player can see and do,” he said.