Diary of a "Doomer": 12+ years arguing about AI risk (part 1)

AI researcher David Scott Krueger has published the first part of a personal account detailing his 12+ years as an AI 'doomer'—someone concerned about existential risks from artificial intelligence. The narrative begins in late 2012 when Krueger, recovering from heartbreak and considering graduate school options, stumbled upon Geoffrey Hinton's neural networks course on Coursera. He had no idea Hinton was 'the Godfather of Deep Learning' or that AlexNet had just revolutionized computer vision, but the course revealed that deep learning actually worked, contradicting the conventional 2009 wisdom he'd heard that neural networks 'don't work.'

Krueger describes being 'blown away' by demonstrations of neural networks generating text and creating new words, seeing it as 'artificial creativity.' He recognized the implications immediately: deep learning's use of distributed, hierarchical, and learned representations meant Real AI (AGI) was no longer a century away but potentially just decades. This realization, occurring in a single afternoon, shifted his career path from aspiring musician to dedicated AI safety researcher. He joined the field when it was still fringe, with only a handful of research groups, but with clear evidence that scaling deep learning could be the path to transformative AI.