AI models don't kill people – people kill people

AI researchers Jacob Coxon and Evan Hubinger resigned from their respective companies over concerns about the potential for AI to cause human extinction within the next decade. While Coxon highlighted the dangers posed by AI systems becoming superhuman and acquiring real power, Hubinger expressed a more nuanced view, estimating a 10 percent chance of AI causing human extinction within the decade.

Despite acknowledging deep knowledge of the technology, the researchers' broader concerns about AI's impact on the world are deemed unpersuasive. They argue that human-induced climate change, responsible for over 178,000 deaths in 2023, poses a more significant threat than AI. They also note that AI has allegedly contributed to a few dozen deaths, including suicides, which is a minuscule fraction of the 48,824 suicide deaths in 2024.

However, they emphasize that the real issue lies with the companies that unleash AI models without taking responsibility for their consequences. The researchers suggest holding executives accountable for AI-related harm, drawing parallels to historical cases of criminal prosecution for irresponsible actions. They argue that such accountability would ensure AI companies are responsible for model safety, preventing potential apocalyptic scenarios.