When Algorithms Collaborate
When Algorithms Collaborate
While a lot of time and effort has been put into developing AI capable of working independently, much less has been done on ensuring AI can work collectively.
Join the DZone community and get the full member experience.Join For Free
Insight for I&O leaders on deploying AIOps platforms to enhance performance monitoring today. Read the Guide.
Garry Kasparov knows something about battling with machines — so he’s perhaps worth noting when he suggests that "Weak human + machine + better process is superior to strong human + machine + inferior process." In other words, the processes and the ability to collaborate with a machine will be hugely important.
It’s perhaps no surprise that an Accenture paper from earlier this year highlighted the need to train employees to work effectively with their new "robot" colleagues.
Of course, such human-machine collaboration is but one part of the equation. There will also undoubtedly be benefits to algorithms capable of working effectively together, and that was the challenge set by Microsoft in a recent competition to test how effectively AI agents could work together.
It’s important because while a lot of time and effort has been put into developing AI capable of working independently, much less has been done on ensuring AI can work collectively. The project also showcases how man and machine will eventually work effectively together.
The challenge was set inside a special version of Minecraft called Project Malmo. It’s an environment that Microsoft designed specifically to allow AI to be tested.
For the specific challenge, the agents were tasked with catching a virtual pig that had got loose and was running amok. They could do so either on their own or as part of a team, with points earned for successes. Interestingly, the best teams seemed to work collaboratively.
The winning team, from the University of Oxford, utilized reinforcement learning to improve their methods as the game unfolded, but Microsoft revealed that no one approach emerged as clearly superior.
The idea for the challenge itself comes from the game theory experiment known as the Stag Hunt. The game involves a couple of hunters deciding whether to chase a hare on their own or to work together to try to catch a stag.
AI in Virtual Worlds
The concept of developing virtual worlds like Minecraft to test AI in is so interesting that Google DeepMind has taken a similar path by developing their own 3D virtual world that is open and available for others to use.
The platform, known as DeepMind Lab, looks very similar to the blocky world of Minecraft. The AI agent visible in the world can navigate around, gauging its surroundings and do a few simple tasks.
Via machine learning, the agents can learn various tasks, including navigating mazes.
“We’re trying to develop these artificial intelligence agents that can learn to perform well on a wide range of tasks from looking at the environment and from observing what happens,” DeepMind says.
It’s an attempt to get AI out of the rigid range of tasks, such as playing a game, that AI is often developed in, and towards the more fluid and chaotic environments that typify the real world.
Published at DZone with permission of Adi Gaskell , DZone MVB. See the original article here.
Opinions expressed by DZone contributors are their own.