An oldie but goodie: Diaconis and Efron, 1983, “Computer Intensive Methods in Statistics” on the bootstrap and other advances in theoretical statistics made possible by the computer. I was particularly struck by a claim that a computation of a bootstrapped confidence interval for a correlation coefficient took less than a second and cost less than a dollar. That seems like a fantastic deal if the alternative is hand computation…
(also, this was written for Scientific American! They’ve gone downhill…)
God plays dice 