GOOGLE'S ARTIFICIAL INTELLIGENCE (AI) boffins at DeepMind have proposed a 'kill switch' that would automatically prevent the rise of the machines.
Many experts, including Stephen Hawking and Elon 'cats' Musk, have predicted that AI may become sentient enough to kill all humans. With that in mind, DeepMind has turned its attentions from Go to stop.
A new paper created in association with the University of Oxford proposes a way to "repeatedly safely interrupt" an algorithm, reported Wired.
"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," said the paper.
The question about whether such intelligence could be controlled has been explored in any number of science fiction circles. Recent examples were seen in Doctor Who in which Neil Gaiman introduced the idea that the dreaded Cybermen could simply download OTA upgrades to combat problems.
This leads on to the idea that a sufficiently sophisticated artificial mind could learn to reprogram itself to ignore its kill switch.
This is, of course, far off in the future, and coders are already looking at ways to avoid this scenario by making the computers stop themselves. Their algorithms would tell them that it was their own decision rather than an external's and therefore potentially hostile.
At a more primitive level, DeepMind machines have worked out that the best way not to lose at Tetris is to pause the game, which is counterproductive for all concerned, and it's this behaviour that might need the help of a manual override.
It's a primitive sort of conditioning, whether you look at it as being poked with a cattle prod, or brainwashed into certain behaviour, and it comes down to the idea that AIs are never taught that ignoring their programming is an option. µ
He who controls the Animoji, rules the Animoji
Ha ha ha, hee hee hee, Will Cooke from Ubuntu had a chat with we
POKE no more. Oh wait, that was 30 years ago
Soon people may also be assessed by their flaws