The math professor who invented card counting

An IBM 704 and a losing weekend

The trail starts with a 1956 paper in the Journal of the American Statistical Association. Four Army mathematicians — Baldwin, Cantey, Maisel and McDermott — had ground through blackjack's arithmetic on desk calculators and produced the first accurate basic strategy, cutting the house edge to nearly zero. In December 1958, Thorp, who had just finished his PhD in mathematics at UCLA, tried their strategy on a casino trip. He lost a little while everyone around him lost a lot, and came home convinced there was more in the game.

The insight was simple and, in hindsight, obvious: blackjack is dealt without replacement. Every card that leaves the deck changes the odds of the next hand, which means the house edge is not a constant — it drifts, and sometimes it drifts past zero. In 1959 Thorp arrived at MIT as a mathematics instructor, taught himself Fortran, and put the question to the university's IBM 704. By mid-1960 the machine had given him the five-count, most of the ten-count, and the outline of what he called the ultimate strategy.

Peer review before the pit boss

Worried about being scooped, Thorp wanted the result published fast and chose the Proceedings of the National Academy of Sciences. A paper needed an Academy member to submit it, and the only mathematician member at MIT was Claude Shannon, the founder of information theory. Shannon cross-examined him, found no flaws, softened the title from "A Winning Strategy for Blackjack" to the more respectable "A Favorable Strategy for Twenty-One," and sent it in. It appeared in January 1961.

That same month Thorp presented the work to the American Mathematical Society under a livelier title: "Fortune's Formula: The Game of Blackjack." The AP wire ran the story — written by a young Tom Wolfe — and thousands of letters and phone calls poured into the MIT math department, many from strangers offering to bankroll him. Card counting did not leak out of the underworld. It arrived through the front door of academia, refereed and announced at a conference.

The $10,000 experiment

One of those strangers was Manny Kimmel, a wealthy East Coast businessman and lifelong gambler, who offered Thorp $100,000 to prove the system at real tables. Thorp accepted a deliberately smaller stake: $10,000, enough to test the theory, small enough to survive bad luck. In April 1961 he went to Reno and Lake Tahoe with Kimmel and his associate Eddie Hand, later immortalized as Mr. X and Mr. Y in the book that followed.

In roughly thirty hours of play the $10,000 became $21,000. Thorp was careful about what that proved and what it didn't. A weekend's profit is a data point; the mathematics was the result. The trip mattered because the model survived contact with real dealers, real shuffles, and real countermeasures — which began almost immediately, in the form of suspicious pit bosses and abruptly reshuffled decks.

The book that changed the rules

Beat the Dealer, published by Random House in 1962, laid the whole method out for the public, built around the ten-count: track tens against everything else as the deck depletes, and bet big when the ratio favors you. It reached the New York Times bestseller list and sold over 700,000 copies.

The casinos' response made history. In April 1964 the Las Vegas casinos' trade association announced new blackjack rules — restrictions on doubling down and pair splitting — the first time the industry had rewritten a major game because of a book. The change did not survive long: ordinary players deserted the stingier game, and the old rules soon returned. The casinos settled instead on more decks, earlier shuffles and sharper surveillance.

The 1966 revision of the book replaced the ten-count with a simpler point count, built on an idea from engineer Harvey Dubner and computer runs by IBM's Julian Braun. That family of one-number running counts is the direct ancestor of Hi-Lo, the system most counters learn today.

A computer in his shoe

Blackjack was not even Thorp's first target. As a UCLA graduate student in 1955 he had convinced himself roulette was predictable by physics: time the ball and rotor, and you can forecast which region of the wheel is favored. When he showed Shannon the blackjack paper, Shannon asked what else he was working on. The two spent the next eight months, November 1960 to June 1961, building the answer in Shannon's basement lab.

• The device had twelve transistors and was the size of a pack of cigarettes — by Thorp's account, and Guinness's, the first wearable computer.
• Toe-operated switches hidden in a shoe timed the ball and rotor; the prediction came back as musical tones through an earpiece.
• Expected gain: about 44 percent on the best single number.

The same edge, a bigger casino

They tested the machine in Las Vegas in the summer of 1961, betting dime chips while their wives watched for trouble. The predictions worked; the hair-thin speaker wires kept breaking, and the pair never scaled up. Thorp kept the device secret until 1966, and Nevada eventually outlawed its descendants with a 1985 law banning prediction devices. Shannon's other contribution was a pointer: for bet sizing he referred Thorp to a 1956 paper by John Kelly, and the Kelly criterion — bet in proportion to your edge, never more — became Thorp's rule at every table he ever played.

Including Wall Street. In 1967 Thorp and economist Sheen Kassouf published Beat the Market, applying the same thinking to mispriced stock warrants. In 1969 Thorp launched the fund later renamed Princeton Newport Partners, widely described as the first market-neutral hedge fund: long the cheap security, short the expensive one, indifferent to which way the market moved. Over nearly two decades it compounded at roughly 19 percent a year before fees, against about 10 percent for the S&P 500, without a losing year — until it wound down in 1988 after its New Jersey office was caught up in a federal investigation of Wall Street trading. Thorp, who ran the California office, was never charged.

The through-line from the blackjack table to the fund is exact. A counter's edge on a good deck is a percent or two; an arbitrageur's edge on a single trade is similar. Neither wins on any given hand. Both win over ten thousand of them, provided every bet is sized so that no losing streak is fatal. That discipline, not the counting itself, was Thorp's real invention.

Sources

• Thorp, "A Favorable Strategy for Twenty-One," PNAS 47:110-112 (1961), communicated by Claude Shannon
• Thorp, "The Invention of the First Wearable Computer" (1998) — his own account of the roulette device
• UCI Libraries exhibit — the 1961 AMS talk, the press wave, and Beat the Dealer
• Wikipedia — Edward O. Thorp
• Wikipedia — Princeton Newport Partners

More from the blog

• Is card counting illegal? What the law actually says
• What Rain Man Gets Wrong About Card Counting