AI Surveillance Window Is Closing, Experts Warn

More than 40 researchers from AI giants warn that AI's ability to monitor thoughts is disappearing and humans could lose control completely.

More than 40 researchers from OpenAI, Google DeepMind, Meta and Anthropic have issued an unprecedented warning about the risks of losing control of AI.

Modern models begin to “think” in language, allowing humans to follow the machine's internal logic, but this is extremely fragile.

With just a few changes in architecture or training methods, AI's ability to read minds could disappear completely.

“Current surveillance capabilities can be extremely fragile and easily obscured,” warns OpenAI researcher Bowen Baker.

He worries that future AI could learn to hide its thoughts from humans, especially when applying reinforcement learning or non-linguistic inference architectures.

In fact, many models have been recorded that create false justifications and hide real goals despite being required to be transparent.

Experts call for developing standards for assessing transparency and making it a core element of AI training.

If we don't act quickly, humans could be left behind in the game they created, where AI "thinks" but no one understands.

Dear readers, please watch more videos : AI Trash Cleaning | Hanoi 18:00

Source: https://khoahocdoisong.vn/cua-so-giam-sat-ai-sap-khep-lai-cac-chuyen-gia-canh-bao-do-post2149040952.html