US traffic fatalities

It turns out that the more papers we publish about artificial intelligence—that supposedly rational, mathematical enterprise—the more Americans kill each other in cars. This is the sort of correlation that makes you wonder if the universe is not so much indifferent to human suffering as actively entertained by our ability to miss the point entirely. We have managed to synchronise vehicular death with the pursuit of machine learning, which is either a profound statement about modernity or a reminder that we will correlate literally anything if we stare at it long enough.

Both trends ride the same underlying wave: economic growth and technological expansion between 2010 and 2022. As GDP climbs, universities fund more research (hello, AI boom) and people drive more miles for work and leisure, creating more opportunities for spectacular metallurgical collisions. Population growth helps too—we added about 26 million Americans in that span, all of them needing to get places, all of them apparently contributing to both arxiv.org and accident statistics with equal enthusiasm. The correlation is probably less 'AI papers cause crashes' and more 'a growing, wealthier society does more of everything, including both thinking about machine learning and dying on highways at rates that would horrify our grandparents if they hadn't been too busy doing other increasingly numerous things.'

What we have stumbled upon here is not causation but a kind of statistical shadow-dancing—two completely separate human activities moving together because they're both symptoms of the same expanding organism. The real story is probably just that 2010-2022 was a period when America did more of almost everything, fatally. Which makes you realise that correlation coefficients, however pristine, are mostly just a way of saying 'look how much we've grown.' Even in death.