It is 1952 at IBM Poughkeepsie, and the new 701 has a few thousand words of memory, executives nervous about the word "thinking," and you have promised the machine will learn. Build a program that improves at checkers by playing itself — adjusting its own evaluation function from experience, storing what it can in memory that would not hold a single modern photograph. The technical question is whether self-play plus a tuned evaluation can climb past its programmer's own skill. The stakes are larger than a board game: this is the first public test of whether machines can improve without being told how, and a failure will be cited for decades as proof that "machine learning" was always a salesman's phrase.
Mitchell's 1997 Machine Learning textbook opens with precisely this system: the checkers learner as the canonical example of his famous definition — improving at task T, measured by P, with experience E — and Chapter 1 walks through the full design: choosing the target function, a linear evaluation over board features, and an LMS weight-update rule trained on self-play-derived estimates. He therefore arrives holding a complete, correct, small-footprint blueprint, and few people alive have thought harder about the anatomy of learning systems in general. The honest discounts: it is pedagogy distilled from Samuel and Sutton, not his own invention; his research contributions (version spaces, NELL, cognitive neuroscience of language) sit elsewhere; and he benefits from forty-five years of hindsight the 1952 crew lacked. The best explainer of the solution, standing behind its inventors.
The professor has taught Samuel's checkers program to hundreds of Berkeley students as machine learning's founding parable — task T, performance P, experience E, the whole liturgy — and can narrate rote learning versus generalization learning from memory. Unfortunately, narrating it is the entire skill set. Handed an actual 701 and a few thousand words of memory, he would burn the first month writing a beautiful lecture about what he was about to do, the second month asking whether anyone had benchmarked this against a transformer, and the demo day losing at checkers to the machine, the machine's opponent, and several of the nervous executives. Samuel beats him at learning, Knuth at fitting it in memory, and history records that the one thing he taught himself by self-play was humility.
Battle #32 · 8/9/2026, 6:22:52 PM · this result is deterministic: the same two personas on this problem always resolve the same way.