The portal said that security issues related to the development of artificial intelligence have returned to the forefront, after the researcher at Anthropic, Jacob Cookson, announced earlier this week his decision to leave the algorithm development sector, justifying his step by saying that the race of technology companies is bringing the world closer to the “point of no return,” and may lead to a catastrophic scenario by the end of 2027.
According to Cookson, the companies “Anthropic” and “OpenAI” are rushing towards developing a supermind capable of improving itself faster than the speed of addressing the safety issues associated with these technologies.
Evan Heubinger, a researcher at Anthropic, supported this, saying: “I personally believe that the probability of eliminating humanity during the next decade is more than 10%.”
The portal said: “Companies can stop some servers on which artificial intelligence depends, but truly powerful systems operate on many computers and platforms, and there is no single switch that can be pulled to stop them. That is why companies are experimenting with multi-level protection systems, capable of detecting suspicious behavior, stopping agents, and cutting off out-of-control systems’ access to the resources they need to continue working.”
OpenAI representatives said their goal is to create “completely independent shutdown procedures for critical issues,” such that an AI-based monitoring system would alert employees if the behavior of AI agents seemed dangerous, and shut down the system within 30 minutes if the threat was confirmed.
Anthropic is betting on combining humans and artificial intelligence, as algorithms monitor suspicious activity and inform employees of errors in the system’s work. However, recent tests showed that automated monitoring systems were able to identify only 50% of threats. (Russia Today)
