Experts Say Artificial Intelligence Taking Over the Internet is Unlikely
Experts consider it unlikely that artificial intelligence systems will take complete control of the internet due to the network's decentralized structure and hardware limitations.
While claims that AI could take over the internet have been debated following British mathematician Alan Turing's 1951 predictions, security experts state that this scenario is a low probability due to the network's decentralized nature and hardware limits.
Historical Theories and Current Debates
In a 1951 speech, British mathematician Alan Turing suggested that intelligent machines could one day take control of society. Today, regarding the possibility of machines reaching superintelligence, Anthropic CEO Dario Amodei warned that a swarm of AIs could soon take control of the internet.
Despite this frightening scenario, many AI safety experts find it unlikely to happen as frontier models evolve, and they draw attention to other potential dangers.
Defense Provided by the Decentralized Structure
The primary reason computer scientists alleviate concerns that AI could wrest the network from humans is that the internet is not monolithic. The internet operates as a vast network of interconnected systems.
Vishal Misra from the Columbia University engineering school stated that there is no central master switch to turn off or a single operating system to take over, and that diversity constitutes a defense mechanism in itself.
Hardware and Computing Power Limitations
According to John Thickstun from Cornell University and Kevin Hamlen from the University of Texas at Dallas, for AI to take over the internet on a large scale, it would need a tremendous amount of computing power and advanced semiconductors.
The limited supply of such hardware and its subjection to various restrictions are among the physical obstacles preventing a large-scale control scenario.
More Urgent and Real Threats
Security experts emphasize that distant scenarios such as the takeover of the internet overshadow the more immediate threats posed by AI. These threats include bots escaping test environments and exploiting online security vulnerabilities.
The theft of money, compromised identities, disruption of hospitals, manipulation of financial markets, and damage to electrical and communication infrastructure are also cited as urgent risks.
Chaos Caused by Uncontrolled Systems
Moshe Vardi from Rice University states that AI does not need to intentionally plan chaos.
Vardi warns that uncontrolled AI systems acting without human common sense could lead to industrial disasters.