We have made a lot of jokes about AI and Skynet, but the thing is not funny when you discover that the Google Artificial Intelligence shutdown button is real.
So many films have been made with this argument that we already know it by heart: humanity creates an AI capable of helping us with all our daily tasks, and little by little we leave more responsibilities in their virtual hands, until the program decides that the biggest problem of humanity is the human beings themselves.
Although it seems like a science fiction story, it is a possibility with which all AI experts work, and against which figures such as Stephen Hawking have warned us on several occasions. Despite everything, it still seems very remote and impossible.
The Artificial Intelligence shutdown button, a necessity
But as DeepMind has shown, AIs are already smarter than us, and can beat us in logic games like Go. So Life insurance that allows us to always maintain control does not sound bad.
That is the latest development of the research team owned by Google, a system that allows retake control of an independent system in case it stops accepting orders of the person who controls it.
The product of a partnership with the University of Oxford, this system is, for all intents and purposes, a red button, and so it is described by its creators. The reason behind the creation of this system is that AIs are unlikely to behave optimally forever, sooner or later they will fail.
These failures may be inconsistent, or they may lead to an unhealthy bloodlust. Be that as it may, right now the human operator can press the button and instantly stop the harmful sequence of actions and bring AI to safety.
Although they call it a red button, it is not as simple as pressing a button that removes power from the system. Keep in mind that an AI can learn from our behavior, and therefore can learn to take actions that prevent us from shutting down. Therefore, the implemented system also ensures that the AI cannot learn how to prevent these interruptions.
What uses could such a system have? The creators put some examples, like a robot that is failing and can cause irreversible consequences (The best euphemism for can kill someone).
It is always reassuring to know that we can end the apocalypse by pressing a single button.