OPINION BIG DATA NATURE|Vol 455|4 September 2008 ESSAY The Harvard computers The first mass data crunchers were people, not machines. Sue Nelson looks at the discoveries and legacy of the remarkable women of Harvard’s Observatory.

A photograph taken at the Harvard Observatory in Cam- bridge, Massachusetts, circa 1890, features eight women in what looks like a Victorian- style sitting room. They wear long skirts, have HARVARD COLL. OBSERV. COLL. HARVARD upswept hair and are surrounded by flowered wallpaper and mahogany tables. At first glance they seem to be sampler stitching or reading. In fact these ‘human computers’ are analysing photographs of the heavens, cataloguing stars. When cameras were first attached to tel- escopes, with the ability to capture the image of thousands of stars on a single photographic plate, people were needed to trawl through these new data. Observatories hired ‘com- puters’ — a term used for human processors since the early 1700s — to do the painstakingly repetitive work of measuring the brightness, position and colours of these stars. From the 1880s until the 1940s, the Harvard College Observatory amassed half a million photographic glass plates, weighing around 300 tonnes and holding images of tens of millions of stars. A team of women trawled through these photos with nothing more than magnifying glasses — often for little pay and Williamina Fleming stands in the centre of the Harvard computers as Edward Pickering looks on. with no scientific training. Despite these unpromising conditions, the attention to detail and stamina. Most of the said that the computing work at his observatory ‘Harvard computers’, who worked from the end plates are negatives; stars appear as fine grey or was so easy that even his “Scotch maid” could of the nineteenth century to the mid-1920s, black spots against a clear background. There do it. This was Williamina Fleming, a school- made tremendous contributions to astronomy. are also several thousand spectral plates, in teacher from Dundee who had emigrated to They determined how to calculate the vast dis- which starlight has been split by a prism before America with her husband in 1878. A year tances from Earth to the stars, and developed being captured. These look like nothing more later, abandoned and pregnant, she secured a star classification systems that are still used than smudged pencil marks a few millimetres job as Pickering’s maid and housekeeper. She today. From photos taken of the northern and wide; under a magnifying glass the smudge was soon working for him at the observatory southern skies, from observatories in Cam- turns into a barcode, revealing information part-time as one of his first computers. bridge, New Zealand and Peru, they produced about the chemical composition and tempera- Pickering’s apparently disparaging remark an astronomical gold mine of data. ture of the stars. about his maid belies the fact that he spotted These women were proof that ‘people power’, and nourished the untapped potential in many even from those with no formal training, is Patience personified intelligent women who worked for him. Flem- capable of great things. It is a trend that con- In 1901, William Elkin, the director of Yale ing was obviously bright and Pickering recog- tinues today: volunteers are recruited from the Observatory, expressed a view typical of the nized this; by 1881, at the age of 24, she was general population and taught to spot objects time as to who was best suited for this work. “I appointed a full-time staff member of Harvard. of interest to astronomers, from the tracks of am thoroughly in favour of employing women Seven years later, she assumed responsibility for interstellar dust left in a spacecraft’s collector, as measurers and computers,” he said. “Not the increasing number of photographic plates, to the direction of spin of a spiral galaxy. With only are women available at smaller salaries editing publications from the observatory and Harvard now working to digitize its photo- than are men, but for routine work they have hiring new computers. During her time at graphic plates, the same pictures of stars scru- important advantages. Men are more likely to Harvard, Fleming examined thousands of spec- tinized by the Harvard computers may soon be grow impatient after the novelty of the work tra and catalogued more than 10,000 stars. available to many more, equally curious, eyes. has worn off and would be harder to retain for Fleming helped Pickering to devise his Working with the repetitive and often indis- that reason.” hydrogen-based stellar classification system, tinct photographs collected at places such as Edward Pickering, the Harvard College which ranked stars according to the strength the Harvard Observatory required patience, Observatory director in 1877–1919, famously of a hydrogen spectral line — A for the

36 NATURE|Vol 455|4 September 2008 BIG DATA OPINION

strongest, then B and so on. She also played Lightman nominates the collector during a rendez- a crucial part in the discovery of the spectral Cepheid distance scale as vous with a comet. One of peculiarities of white dwarfs. Fleming was one of the most important the greatest challenges then appointed Harvard’s Curator of Astronomi- breakthroughs in twentieth- lay in finding the tracks of cal Photographs in 1899 — the first woman so century science. To celebrate tiny interstellar particles — appointed — and eight years later she became the centennial of Leavitt’s pio- rare pieces of dust that came HARVARD COLL. OBSERV. OBSERV. COLL. HARVARD the first female American citizen elected to neering work, the Harvard from distant stars — among the British Royal Astronomical Society. Dur- College Observatory is the more common tracks left ing her career it is estimated that she exam- holding a symposium in her by comet particles. These ined around 200,000 photographic plates. honour this November. tracks, at a millionth of a Pickering’s ‘harem’, as they have been called, More advanced cameras metre across, were a devil could have earned more per hour doing menial and digital photography to find on the 1,000-square- work in the local mill town. But the observatory removed the need for most centimetre collector. appealed to intelligent and educated women, photographic plates in the In 2006, the Planetary Soci- including researchers and graduates from 1980s. But the sheer number ety and the University of Cal- the new all-female colleges, who were keen of plates collected at Harvard ifornia, Berkeley, launched to find patterns in the data and draw interest- means that even now there the Stardust@Home project. ing conclusions from them. Pickering allowed are stars that have yet to be Just as the Harvard comput- the women to do their own research in their analysed. Because of this ers were trained to analyse HARVARD COLL. OBSERV. OBSERV. COLL. HARVARD spare time and their names were often cited untapped potential, and stellar spectra, members of as co-authors in scientific papers. He encour- the historical significance the public were trained, via aged them to give talks, and to be recognized of the computers’ work, the online tutorials, to scan pho- as astronomers in their own right. DASCH (Digital Access to tos of the gel on their com- Annie Jump Cannon, for example, meas- a Sky Century at Harvard) Spectra (top) pulled from star plates puter screens and identify ured and recorded the colours of 300,000 project is hoping to scan all (above), are only millimetres wide. possible tracks. No compu- stars, classifying them into spectral groups at the half-million plates over ter program exists that can a rate of up to 300 an hour. Her true achieve- the next four years; the team has so far scanned do this as well as the human eye. ment came when she was tasked with find- more than 3,000 plates and is trying to raise In the first phase of the project, 23,000 vol- ing a more meaningful way of arranging the US$4 million to complete the work. unteer ‘dusters’ searched nearly 40 million star categories in Pickering’s hydrogen-based images, flagging any photos of possible inter- system. Cannon developed a reordering and Pattern recognition est for trained scientists. Without public help, simplification that ranked stars in order from The primary motivation of this work is not it would have taken the team at least 20 years to the bluest and hottest to the coolest, red ones. historical preservation, but to dig deeper into locate tracks. With their help, it took months. The sequence — O, B, A, F, G, K, M (remem- the data. Requests have already been filed Phase two is currently under way, with dusters bered by the mnemonic Oh Be A Fine Girl, by researchers for access to the DASCH’s searching for particles in photos of a higher Kiss Me) — remains in use today. digitized collection. One Harvard graduate magnification. Henrietta Swan Leavitt, a college graduate, student, Sumin Tang, working with DASCH Inspired by the Stardust project, Galaxy Zoo joined Pickering as a research assistant. She principal investigator Josh Grindlay, recently went online in 2007. This project used volun- studied Cepheid variable stars — whose light found something unusual after studying just teers to classify spiral and elliptical galaxies from brightens and dims over periods ranging from 500 of the digitized plates: a star that bright- images taken by the Sloan Digital Sky Survey. hours to years — and found that these periods ened by nearly a factor of two over 20 years Within six months of operation, volunteers had were related to the bodies’ intrinsic brightness and then levelled off, retaining its brightness identified more than 500 overlapping galaxies in a predictable way. This relationship, once for 60 years. This new type of variable star will — when a galaxy appears behind or in front of calibrated, allowed astrono- be observed further using a another from the point of view of an observer mers to determine such a “There remain some telescope in Arizona in the on Earth — of which only 20 were previously star’s distance, based on its years to come. known to astronomers. Recently, a unique true brightness as compared areas in which the The sheer magnitude of object containing the hot gas of a normal galaxy to its apparent brightness human eye is better than data expected for the entire but without any stars was discovered through as seen from Earth. This a modern computer.” DASCH project (around 1,500 the project by a Dutch primary school teacher; ‘Cepheid distance scale’ terabytes) lends itself well a paper describing the find is in the works. became Edwin Hubble’s to analysis by today’s high- Modern astronomy could not be done with- ‘yardstick to the Universe’, eventually allowing speed computers. Yet there remain some areas out supercomputers crunching epic quantities him to discover that the Universe was expand- in which the human eye is better than a modern of data. It is nonetheless worth remembering ing. It is still crucially important in distance computer, particularly in the realm of pattern the value of the human mind in spotting com- measurements today. recognition. Classifying stars or galaxies, like plex patterns or following a hunch — as proved Leavitt died in 1921 of cancer, four years classifying species, is still something most easily by the persistent, repetitive and inspiring work before a letter was sent notifying her that she done by people with knowledge and a knack. of the pioneering women at Harvard. ■ had been nominated for a Nobel prize — an In some cases, the best results come from the Sue Nelson is a writer and broadcaster living in honour that had to be withdrawn because it use of many eyes — even if they are untrained. Hertfordshire, UK. could not be awarded posthumously. In his NASA’s 2004 Stardust mission, for example, e-mail: [email protected] 2005 book The Discoveries, physicist Alan gathered dust particle samples on an aerogel See Editorial, page 1.

37