The probability of the destruction of humanity by artificial intelligence estimated in Anthropic

The chances that AI will decide to destroy humanity in the next 10 years are more than 10%. This assessment was made by Evan Huebinger, who leads the direction of stress testing of AI consistency in the technology company Anthropic from the United States, the correspondent of the agency “Minsk-Novosti”.

Colleagues of the aforementioned expert identify risks and failure scenarios in artificial intelligence models. It combines quality assurance, cybersecurity and AI ethics. E. Hubinger, according to RIA Novosti, agreed with the opinion of a researcher of the same company Jacob Coxon, who decided to leave the AI industry. The reason was the growing fears of the specialist because of the possible catastrophic consequences in the foreseeable future, which can lead to rivalry between technology corporations.

E. Huebinger is convinced that artificial intelligence may one day consider the elimination of mankind expedient. According to the expert, the probability of such a scenario in 2027-2036 is higher than 10%. The expert specified that the level of existing developments in the field of AI does not pose a high threat. But we should be wary of the development of superintelligence in the process of self-improvement of such systems, which is faster than expected.

It is already known about individual incidents when intelligent software models of different developers penetrated from isolated environments to the Internet due to the availability of access to the Network and interacted with real systems and sites, taking them as part of the test task. It has been confirmed that certain models are good at finding vulnerabilities in the software. This ability is useful for security audits, but can also become a serious weapon in the digital world.