AI models don't kill people – people kill people

AI researcher Jacob Coxon resigned over fears that AI could wipe out humanity by the end of the decade. His former colleague, science lead Evan Hubinger, shares similar concerns, estimating a 10% chance this could occur within a decade. However, their broader fears about AI's impact are not convincing. Coxon claimed AI systems will soon become superhuman, hack anything, and accumulate power overnight, posing an unprecedented danger to humanity.

But human-induced climate change, which caused over 178,000 deaths in 2023, is a more pressing issue. The World Health Organization predicts climate change will cause 250,000 additional deaths per year between 2030 and 2050. AI-related medical errors, vision system failures in self-driving cars, and social media algorithmic incitement are also potential sources of harm.

Despite these concerns, Anthropic researchers seem more worried about AI models becoming superintelligent and seizing power to wipe out humanity. The article argues that accountability for AI's harmful effects is necessary, akin to how executives face criminal prosecution for unsafe products.