Machines are becoming more intelligent every year thanks to advances being made by companies like Google, Facebook, Microsoft, and many others. AI agents, as they're sometimes known, can already beat us at complex board games like Go and they're becoming more competent in a range of other areas. Now a London AI research lab owned by Google has carried out a study to make sure we can pull the plug on self-learning machines when we want to. DeepMind, acquired by Google for a reported 400 million in 2014, teamed up with scientists at the University of Oxford to find a way to make sure AI agents don't learn to prevent, or seek to prevent humans, from taking control. The peer-reviewed paper -- titled "Safely Interruptible Agents [PDF]" and published on the website of the Machine Intelligence Research Institute (MIRI) -- was written by Laurent Orseau, a research scientist at Google DeepMind, Stuart Armstrong at Oxford University's Future of Humanity Institute, and several others.

Trying to understand what is _really_ going on in the universe.

Earlier this year, Google's DeepMind hopped a massive artificial intelligence hurdle when it ousted Lee Sedol, a 9-dan Go player. This was significant because Go has long been upheld as a distinctly human game. There are far more possible games that can be played on a 19 19 Go board than there are atoms in the known universe, and this has proven to be a particular challenge for non-human competitors. For years, even a mid-range Go player could trample the most sophisticated Go-playing programs, so AlphaGo's victory was a pretty big deal. However, although Lee Sedol is a grandmaster, he is not the world's top Go player.

Google's DeepMind AI division will face off against humanity's number one Go player, Ke Jie, according to Chinese state news agency Xinhua. Ke Jie is currently the top-ranked Go player in the world. From killer robots, to runaway sentience, there's a lot of FUD that needs clearing up. DeepMind's Go-playing program AlphaGo played against Go grandmaster Lee Se-dol in March. Lee lost to the computer 4-1.

TechRadar and other tech sites are reporting that Google is thinking up a kill switch for dangerous AI. As TechRadar put it, humans can keep the upper hand, for now. David Nield reported: "Google is working on an AI'kill switch' that allows human operators to turn off super intelligent systems no matter how big their egos get. It's called "safe interruptibility" and it's being developed as part of the DeepMind system." An open letter went out last year that caused as much of a stir as the worrisome statements preceding it over building superintelligent machines.

It's an interesting turn of events, particularly after Ke once said he didn't want to play Google's Deepmind computer because it would learn his playing style. "I don't want to play against AlphaGo because I can tell from its performance that it is weaker than me," he told Chinese media. "I don't want it to copy my patterns and learn from me." Those emotions changed when 33-year-old 9th dan professional Lee Sedol was finally defeated by AlphaGo. The win is considered a huge milestone for artificial intelligence given the complexity of the Chinese board game.

Google has suggested a "big red button" could be used to prevent artificial intelligence from a "harmful sequence of actions". A research paper from DeepMind and the University of Oxford says there should be a way to "repeatedly safely interrupt" an algorithm. "Safe interruptibility can be useful to take control of a robot that is misbehaving and may lead to irreversible consequences, or to take it out of a delicate situation, or even to temporarily use it to achieve a task it did not learn to perform or would not normally receive rewards for this," the paper, which proposes a framework to let humans stop an algorithms from continuing on a dangerous path, says. The paper's authors – Laurent Orseau, from DeepMind, and Stuart Armstrong from the The Future of Humanity Institute – explain that an "interruption policy" could be built into algorithms to safely stop a machine. Reinforcement learning algorithms, the paper continues, often work in complex environments, such as "the real world", and are unlikely to act as they are intended on every occasion.

The Future of Humanity Institute, University of OxfordStuart Armstrong is a philosopher at the University of Oxford and one of the paper's authors. Machines are becoming more intelligent every year thanks to advances being made by companies like Google, Facebook, Microsoft, and many others. AI agents, as they're sometimes known, can already beat us at complex board games like Go and they're becoming more competent in a range of other areas. Now a London AI research lab owned by Google has carried out a study to make sure we can pull the plug on self-learning machines when we want to. DeepMind, acquired by Google for a reported 400 million in 2014, teamed up with scientists at the University of Oxford to find a way to make sure AI agents don't learn to prevent, or seek to prevent humans from taking control.

Ever since the creation of artificial intelligence, we've worried about what the consequences would be if AI suddenly decided it knew better than we did and started making decisions of its own. This hypothetical scenario hardly ever ends well. It's comforting, then, to know that Google is working on an AI'kill switch' that allows human operators to turn off super intelligent systems no matter how big their egos get. It's called "safe interruptibility" and it's being developed as part of the DeepMind system that recently proved its prowess at Go. The team working on DeepMind has published a paper on the topic and set out a basic framework for a kill switch (or "big red button") that will wind down whatever robot army is currently marching on the major capitals of the world.

With artificial intelligence crossing milestones in its capability to learn rapidly from its environment and beat humans at tasks and games from Jeopardy to the ancient Chinese game Go, Alphabet's Google is taking proactive steps to ensure that the technology it is creating does not one day turn against humans. Google's AI research lab in London, DeepMind, teamed up with Oxford University's Future of Humanity Institute to explore ways to prevent an AI agent from going rogue. In their joint-study, "Safely Interruptible Agents," the DeepMind-Future of Humanity team proposed a framework to allow humans to repeatedly and safely interrupt an AI agent's reinforcement learning. But, more importantly, this can be done while simultaneously blocking an AI agent's ability to learn how to prevent a human operator from turning off its machine-learning capabilities or reinforcement learning. It's not a stretch to think AI agents can learn how to outthink humans. Earlier this year, Google's AI agent AlphaGo beat world champion Lee Sedol in Go, the ancient Chinese game of strategy.