Bicyclist traffic fatalities

As the artificial intelligence research community has produced papers at an accelerating rate—each one promising, in its own modest way, to reshape the future of human civilization—more bicyclists have been dying on American roads. The correlation suggests either that AI researchers are terrible drivers, or that the universe maintains a strict equilibrium between intellectual progress and physical vulnerability. Between 2010 and 2022, both metrics climbed with the quiet determination of things that do not know they are being compared.

AI paper output on arXiv grew from a few thousand per year to over 30,000 during this period, fueled by the deep learning revolution that began around 2012 and the subsequent flood of venture capital, academic grants, and corporate research budgets. Bicyclist fatalities rose from about 620 in 2010 to over 1,000 by 2022, driven by the same urban factors plaguing pedestrians: heavier vehicles, faster traffic, inadequate cycling infrastructure, and the omnipresent smartphone. Both trends are products of an economy that was simultaneously investing in computational intelligence and failing to invest in protected bike lanes—a set of priorities that says something uncomfortable about what we value enough to fund.

We are building machines that can write poetry and recognize faces but cannot seem to build cities where riding a bicycle does not carry a meaningful risk of death. The AI papers and the cycling fatalities share a decade, not a mechanism, but they share it with the uncomfortable symmetry of a society that optimizes brilliantly for some things and catastrophically for others. The algorithms, at least, are improving.