Artificial intelligence is getting smarter and smarter, more intelligent every day. Robots are getting scarier and scarier every day. I suppose it’s only a matter of time before smart but cold-hearted robots will take hold.
While Japanese engineers rejoice at the birth of a robotic child with an artificial intelligence of a five-year-old boy, the Chinese police force is testing robot policemen. The United States Marine Corps has several robots whose intelligence controls machine gun turrets.
Of course, most robots of a military nature can only respond to preprogrammed commands, and must be controlled by a human overseer. However, advances in artificial intelligence could allow robots to wield weapons and act autonomously - even if that means using weapons against humans.
Recently, two Carnegie Mellon computer science students were announced for a new vision of science fiction. Scientists of the future have pushed deadly robots closer to reality. According to the press release, students Devendra Chaplot and Guillaume Lample have developed an artificial intelligence system that is vastly superior to past computer AI.
It uses deep learning and reward systems to encourage the AI to kill other players better. Testing was conducted in the classic first-person shooter game DOOM.
Researchers behind the potentially deadly development of AI have published their findings on arXiv.org, although their methods have not yet been peer reviewed.
According to the study, artificial intelligence is able to evaluate different aspects of the game simultaneously, which allows it to learn and develop optimal winning strategies much faster than humans and existing AIs:
One of the study's authors, Devendra Chaplot, states in a press release that the system was not designed to train artificial intelligence to kill people. We just trained him to play the game well.
Promotional video:
However, some computer magazines noted the system is easily portable and adaptable with other devices. This means that AI can be used not only in video game scenarios. It is enough to connect the visual inputs of the system to street cameras, and now the AI can easily track city streets, and can shoot real people, and not game avatars.
It gets scary when you imagine what scientists working for the military-industrial complex can do. After all, they are probably developing their own killer artificial intelligence systems, right? Will future descendants not have to decide the question of who will "saddle" whom and put to his service - a robot of the 20 … model year or a human?