Wednesday, March 6, 2013

The Story of the Way Big Data Shaped World War II

It’s odd to think that a simple street address in Mountain View (CA), a short bicycle ride from the headquarters of Google, Intuit and LinkedIn, connects the origins of Big Data, Silicon Valley with the brutal World War II Battle of the Atlantic – during which Allied forces pursued the packs of Nazi submarines that almost succeeded in severing the United Kingdom’s only supply line.
391 San Antonio Road was the original home of Shockley Semiconductor, a company founded in 1956 by the controversial, co-inventor of the transistor, Nobel Laureate William Shockley. It was Shockley’s company, or more precisely his polarizing personality, that eventually led to the birth of many of the most important silicon companies of the past half century – most notably, Intel. A decade earlier, a younger Shockley had joined a team of British and American scientists that, for the first time in the annals of warfare, used data to determine the shape of battle and so helped tip the course of history.
This episode, now fading into obscurity, is chronicled in Blackett’s War, a new book by Steven Budiansky which relays how information and analysis, pieced together from studious scrutiny of carefully collated data by a most unwarlike coterie of mathematicians and physicists, eventually persuaded leaders of the Allied navies and air-forces to set aside their dogma, deeply held prejudices and vanity to change the way they fought. It marked the birth of operations research, the science that decades later does everything from schedule the production line for the Dreamliner to displaying video ads on YouTube in front of the right pair of eyeballs.
After submarines first appeared in World War I they had been dismissed as a defensive weapon, but they quickly became a potent menace to shipping. At the outset of World War II, all combatants had come to recognize the lethal threat of these torpedo-bearing vessels. During 1940 and 1941 the Nazi submarine fleet was a nautical jackboot pressing into the shipping lanes along which convoys ferried food and munitions, and after Pearl Harbor, U.S. troops to British ports. This was the foe faced by British forces and their Secret Service.
When the blitz began, British intelligence was dominated by men who had distinguished academic results in subjects like Latin and Greek, were proficient fox-hunters or members of the right London Clubs. But counting bagged partridges and quail did not require the mastery of advanced mathematics and probability theory necessary to design adequate responses to attacks that first appeared over the skies of Britain. A group of scientists was hastily cobbled together from British Universities to help with the intelligence effort and was headed by Patrick Blackett, a Cambridge educated scientist who, three years after the War ended, was awarded the Nobel prize for physics.
Blackett’s Circus cuts its teeth – and started to establish credibility – by working for Coastal Command during the Battle of Britain. Here, they married information gleaned from primitive radar systems with the firing patterns of batteries and managed to cut the number of shells required to down one German bomber from about 20,000 to 4,000 – a task aided by the Sperry Predictor, a half ton mechanical ‘computer’ connected to the guns. They also calculated the number of submarines that the Coastal Command airplanes should expect to spot per flying hour and the most efficient scouting system. When results fell far short and it became apparent that the planes were being detected earlier than expected by U-boat crews, the paint on the bottom of the wings was changed from black to white. Subsequent air sorties met with improved success.
After the perils from the sky dissipated, Blackett’s attention turned to the marauding U-boats. Early attacks on the submarines had proved fruitless. Within a twelve-week period during the summer of 1940, 150 ships (and a million tons) were sunk. The Nazi boats were even marauding the eastern coast of the United States, sinking ships off Cape Cod, New Jersey and Cape Hatteras. Winston Churchill wrote, “The Battle of the Atlantic was the dominating factor all through the war."
Blackett’s scientists transferred the approach used to great effect during the Battle of Britain to the challenge of the Battle of the Atlantic. Before offering advice on tactics they combed operational records and accompanied missions to detect helpful patterns and accumulate a trove of data. They were also made privy to the riddles of German code signals unwrapped by the famous Enigma machine with assistance from rudimentary IBM and NCR punch-card computers. From this mélange of data, Blackett’s group of boffins estimated vital details such as the amount of time attacking aircraft had to mount an attack after a U-boat (15 seconds), the optimal setting for depth charges (25 feet) and the best bombing pattern. They also proved that it was wiser to have fewer serviceable airplanes and instead keep more airborne rather than vice versa – something that flew in the face of accepted wisdom.
The scientists, several of whom weren’t bashful about touting their leftist leanings, continued to be viewed with intense suspicion by the military brass. Proponents of heavy bombing of German cities weren’t sympathetic to the argument that the war would be shortened by diverting planes to attack U-boats, despite compelling evidence, which was often ignored by military commanders, that the mortally dangerous air-raids over German cities were not doing much to disrupt production or supplies. Thousands of airmens’ lives were lost in a similarly futile effort to destroy the enormous concrete pens the Germans had constructed to protect their U-boat fleet in ports along the coast of France. Even the heaviest of allied bombs did little more than pepper these bunkers – many of which still exist.
The navy admirals were as obdurate as their aerial counterparts and refused for a long time to abandon their belief that small convoys were safer than larger ones despite incontrovertible evidence to the contrary. Blackett’s Circus proved that each ship in a convoy of 15-24 ships had a 2.3% chance of being sunk compared to 1.1% in a convoy larger than 45 vessels. (Doubling the size of the convoy only required increasing the number of escorts by 16% because the length of the perimeter requiring protection increased very slowly as convoys grew in size.)
Eventually, the probability theories started to take a grip and during 1943 the balance was tilted when the Allies targeted U-boats travelling to and fro across the Bay of Biscay. The overall results of this early employment of operations research were startling. Despite the gruesome fact that U-boats sank about 2,800 ships, the mortality rate for the submariners was far worse. Of the 830 U-boats that took part in World War II – 784, or 94%, were lost. By today’s standards Blackett’s work might just be considered a triumph of brain over brawn but in the past seventy years no data has been quite so big.