Why Superintelligent AI Cannot Be Controlled?
Using theoretical calculations, a team of international researchers, including scientists from the Center for Humans and Machines at the Max Planck Institute for Human Development, shows that it would be impossible to control highly intelligent AI.
The study was published in the Journal of Artificial Intelligence Research.
Suppose someone could design an AI program with higher intelligence than humans, so it could learn independently. Connected to the Internet, AI may be able to access all personal details. It can replace all existing systems and control all online devices worldwide. Will this produce utopia or dystopia? Will AI cure cancer, bring peace to the world, and prevent climate catastrophe? Or can it destroy mankind and take over the earth?
Computer scientists and philosophers have wondered if we will ever be able to control the sophisticated AI, ensuring that it will not pose a threat to humanity. An international team of computer scientists has used theoretical calculations to show that it would be impossible to control highly intelligent AI.
“The ingenious machine that controls the earth sounds like a science fiction novel. But there are already machines that perform certain important tasks independently of the system to fully understand how to learn them. The question then arises as to whether this can sometimes be controlled and dangerous to humanity, ”said co-author Manuel Cebrian, Director of Digital Mobilization Group at the Center for Humans and Machines, Max Planck Institute for Human Development.
Scientists have explored two different ideas about how intelligent AI can be managed. On the other hand, the power of professional AI can be particularly limited, for example, by plugging it into walls on the Internet and all other technological devices so that there is no external contact with the outside world.
“The intelligent machine can controls the earth sounds like science fiction. But there are machines that perform certain important tasks independently of the system to fully understand how to learn them. So the question arises as to whether this can sometimes be controlled and dangerous to humanity,” said study author Manuel Cebrian. , Leader of Digital Mobilization Group at the Center for Humans and Machines, Max Planck Institute for Human Development.
Get the latest news and learn more about The Internet Of Things Report
Scientists have explored two different ideas about how intelligent AI can be managed. On the other hand, the power of professional AI can be significantly reduced, for example, by installing it on the Internet and all other technological devices so that it cannot communicate with the outside world — but this will make the dominant AI less powerful, unable to answer people’s questions. Lacking such an option, AI may be encouraged from the outset to pursue only the best interests of mankind, for example by setting ethical principles in it. However, researchers also point out that these ideas as well as other modern and historical theories of intelligent AI control have their limitations.
In their study, the team developed a content algorithm that ensures that the most powerful AI cannot harm people under any circumstances, by imitating the behavior of AI-first and stopping it when it is considered dangerous. But careful analysis shows that in our current computer concept, such an algorithm cannot be built.
“If you skip the problem with the basic principles from theoretical computer science, it turns out that the algorithm that can command AI not to destroy the earth may stop its operation unintentionally. contains harmful AI. In fact, this makes the content algorithm unusable, “said Iyad Rahwan, Director of the Center for Humans and Machines.
Based on these statistics the content problem is incomparable, e.g. No single algorithm can find a solution to determine whether AI can cause harm to the world. In addition, researchers point out that we may not even know when intelligent machines arrived, since determining whether a machine displays higher intelligence than humans is in the same area as the content problem.
