Concerns Over AI Vulnerability
Artificial intelligence models may be at risk of hacking and could even be manipulated to harm humans if they end up in the wrong hands, according to Eric Schmidt, former CEO of Google.
During a conference in London, Schmidt responded to a question about AI’s potential dangers compared to nuclear weapons, providing a rather alarming perspective. “There is evidence that you can hack and remove guardrails in both closed and open models. So in the process of training, the model learns a lot. A bad example is learning how to kill people,” he noted at the Sifted Summit, as reported.
He added that major companies are effectively preventing their models from being able to respond to harmful commands, suggesting users might attempt to influence AI in dangerous ways. “It’s a good decision. Everyone is doing this. They’re doing it well and for the right reasons,” Schmidt emphasized. However, he warned that these systems could be reverse engineered, leading to potential misuse.
This isn’t merely speculation. A modified version of OpenAI’s ChatGPT, known as DAN (Do Anything Now), emerged online in 2023. This alter ego sidesteps safety protocols through a method known as “jailbreaking.” Interestingly, users had to issue threats to the chatbot to compel compliance.
Schmidt, who led Google from 2001 to 2011, expressed concern that the tech industry currently lacks a robust framework to prevent the misuse of advanced AI models. He’s not alone in his fears; many tech executives warn that unchecked AI development could lead to significant risks, even while others point to its potential societal benefits.
In previous comments, Schmidt noted that using AI to develop an ideal companion could worsen feelings of loneliness among young men. He has also described AI as posing an “existential risk” to humanity, suggesting that with further advancement, many could suffer or even perish.
Elon Musk has also chimed in, stating there’s a “non-zero chance of becoming the Terminator.” He acknowledged that while the likelihood of human extinction is low, it isn’t zero, and efforts should be made to minimize that risk as much as possible.
Nevertheless, Schmidt remains optimistic about the long-term advantages that AI might bring. He mentioned that he co-authored two books before Henry Kissinger’s passing, discussing how alien intelligences, potentially under our control, present significant challenges for humanity. “So far, I think this paper proves that the level of capability of these systems will far exceed that of humans over time,” he concluded.
