A competition which saw human players go up against artificial intelligence (AI) in the game Doom showed just how far AI learning techniques had come but at the same time became the subject of some controversy.
Two computer science students in the US, Devendra Chaplot and Guillaume Lample, who study at Carnegie Mellon University published a paper in which they reported in detail how their AI bot learned to kill human players in deathmatch scenarios. Using deep learning techniques theytrained their AI bot ‘Arnold’ to navigate the 3D environment of the game.
By repeatedly playing the game, Arnold eventually became an expert in fragging its Doom opponents artificial combatants and human players alike.
Deep learning was previously used to train AIs to master 2D video games and board games and now research shows that the same techniques can now be applied to 3D virtual environments.
“In this paper, we present the first architecture to tackle 3D environments in first-person shooter games,” the researchers write in their paper. “We show that the proposed architecture substantially outperforms built-in AI agents of the game as well as humans in deathmatch scenarios.”
As impressive as it may seem, the fact that AI researchers are training machines to view human opponents in the game as ‘enemies’ and kill them has drawn some criticism.
“The danger here isn’t that an AI will kill random characters in 23-year-old first-person shooter games, but because it is designed to navigate the world as humans do, it can easily be ported,” writes Scott Eric Kaufman at Salon.
“Given that it was trained via deep reinforcement learning which rewarded it for killing more people, the fear is that if ported into the real world, it wouldn’t be satisfied with a single kill, and that its appetite for death would only increase as time went on.”
While it’s virtually impossible that this particular Arnold AI will somehow end up in the real world and lead to the rise of the machines, we’re tinkering with some pretty scary stuff here.