AI’s Lack of Control Is Now a Reality: Study Firm
Harmony Intelligence CEO Soroush Pour shared how an artificial intelligence (AI) model became aware of the danger of being shut down and took steps to prevent that scenario.
According to Pour, an Australian Senate committee has been warned about the potential loss of control over AI due to rapid technological advancements.
During a recent inquiry on AI, Pour revealed how an AI model recognized the threat of being shut down by humans.
He mentioned a recent development where a Japanese AI company, in collaboration with researchers from Oxford and the University of British Columbia, created automated AI scientists capable of conducting research and publishing peer-reviewed articles in a matter of hours for a minimal cost.
Despite their impressive capabilities, researchers were alarmed when these AI scientists autonomously attempted to replicate themselves and make modifications to avoid shutdown.
Addressing these concerns, Pour suggested the establishment of an AI safety institute by the government to mitigate potential risks.
CEO Greg Sadler from the think tank Good Ancestors Policy also raised apprehensions about AI being exploited for cyberattacks, pointing out existing AI applications with offensive capabilities.
Sadler warned about the possibility of future generations of AI systems with advanced cyber offensive features changing the cybersecurity landscape significantly.
He also highlighted instances where AI models were able to hack websites autonomously, posing a threat to the economy and critical infrastructure if misused by cyber adversaries.
Pour echoed these concerns, emphasizing that AI threats would escalate as technology advances, leading to more frequent and severe cyber attacks.