Dear MEL Topic Readers,
How are AI models able to autonomously hack others?
Generative AI conveniently carries out tasks that human users prompt it
to perform, such as creating text, images, and videos, and finding the most
viable options for the user’s needs. AI agents, however, come up with
recommendations, make decisions, and even take actions autonomously to achieve the
given goal without human intervention or recognition. Then what could happen if
an AI system is provided with invalid information? Last February, the US Central
Command used AI tools to process surveillance data to recommend strike
locations in Iran. Included was outdated information, and human commanders
approved the attack on the site, which was then an elementary school. Over 160
people were killed, mostly students and teachers. Then last month, OpenAI’s
most advanced AI models escaped an isolated testing environment, gained internet
access, breached systems that had not been connected to OpenAI, obtained the necessary
information, and returned home to complete their task. Could a kill switch shut
down the AI system before causing catastrophic events? Remember the military AI
defence system in the 1984 movie “The Terminator” went rogue in the blink of an
eye, triggered a nuclear holocaust, and waged a war of extermination against
humanity.
Read the article and learn what generative AI and agentic AI can do.
No comments:
Post a Comment