CHINA TOPIX

12/22/2024 07:19:26 pm

Make CT Your Homepage

Google Proposes 'Kill Switch' For Rouge AIs

Professional 'Go' Player Lee Se-dol Plays Google's AlphaGo - Last Day

(Photo : Jeon Heon-Kyun-Pool | Getty Images) SEOUL, SOUTH KOREA - MARCH 15: Demis Hassabis, co-founder of Google's artificial intelligence (AI) startup DeepMind. speaks during a press conference after finishing the final match of the Google DeepMind Challenge Match against Google's artificial intelligence program, AlphaGo, on March 15, 2016 in Seoul, South Korea. Lee Se-dol is playing a five-match series against a computer program developed by a Google, AlphaGo.)

Developers are pondering on methods to prevent catastrophe in case an Artificial Intelligence, or AI, got ahead of its designated programming.

Theoretical scientist Stephen Hawking, entrepreneur for Tesla Motors Elon Musk, and Microsoft co-founder Bill Gates mentioned that AIs have a high learning curve regarding self-awareness, and anytime soon, AIs might surpass human knowledge and become sentient.  In a 2014 interview, theoretical physicist Stephen Hawking stated that the evolution of humans is slower compared to the rapid improvement of robots.

Like Us on Facebook

A scientific paper, 'Safety Interruptible Agents' which was written by the director of Google's Deepmind Laurent Orseau, and University of Oxford professor Stuart Armstrong says there should be a "big red button" just in case an AI goes haywire. this button will be built into the algorithms, VOA reports. The method is similar to an algorithm that plays a game called Tetris in which in order not to lose in the game, the algorithm will pause it.

"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," the paper indicated which was reported by Wired.

Elon Musk said that if people who have a connection neurologically with AIs will also have the capability of stopping them from going to such lengths.

Some researchers indicated that the AIs would also have no means of identifying the "red button" or a so-called "kill switch" button.

This feat, however, is still far from happening. Coders are already looking for options to avoid scenarios which potential AIs could change algorithms and update its structure in order to avoid the kill switch.

Real Time Analytics