As a woman, Dr. Ming started noticing that people treated her differently. There were small things that seemed innocuous, like men opening the door for her.
There were also troubling things, like the fact that her students asked her fewer questions about math then they had when she was a man, or that she was invited to fewer social events — a baseball game, for instance — by male colleagues and business connections.
Bias often takes forms that people may not recognize. One study that Dr. Ming cites, by researchers at Yale, found that faculty members at research universities described female applicants for a manager position as significantly less competent than male applicants with identical qualifications. Another study, published by the National Bureau of Economic Research, found that people who sent in résumés with “black-sounding” names had a considerably harder time getting called back from employers than did people who sent in résumés showing equal qualifications but with “white-sounding” names.
Everybody can pretty much agree that gender, or how people look, or the sound of a last name, shouldn’t influence hiring decisions. But Dr. Ming takes the idea of meritocracy further. She suggests that shortcuts accepted as a good proxy for talent — like where you went to school or previously worked — can also shortchange talented people and, ultimately, employers. “The traditional markers people use for hiring can be wrong, profoundly wrong,” she said.
所以 Gild 公司希望使用更客觀的公開資料來幫助別的公司聘用人才
Dr. Ming’s answer to what she calls “so much wasted talent” is to build machines that try to eliminate human bias. It’s not that traditional pedigrees should be ignored, just balanced with what she considers more sophisticated measures. In all, Gild’s algorithm crunches thousands of bits of information in calculating around 300 larger variables about an individual: the sites where a person hangs out; the types of language, positive or negative, that he or she uses to describe technology of various kinds; self-reported skills on LinkedIn; the projects a person has worked on, and for how long; and, yes, where he or she went to school, in what major, and how that school was ranked that year by U.S. News & World Report.
經過上述 300 多項的變數分析後，給每個程式設計師一個 Gild score (註 2)
The company has amassed a database of seven million programmers, ranking them based on what it calls a Gild score — a measure, the company says, of what a person can do. Ultimately, Dr. Ming wants to expand the algorithm so it can search for and assess other kinds of workers, like Web site designers, financial analysts and even sales people at, say, retail outlets.
也應用此方法找到 Mr. Dominguez 成為員工 (高中畢業的 Dominguez 年薪 115,000 美元)
Mr. Dominguez had made quite a contribution. His code for Jekyll-Bootstrap, a function used in building Web sites, was reused by an impressive 1,267 other developers. His language and habits showed a passion for product development and several programming tools, like Rails and JavaScript, which were interesting to Gild. His blogs and posts on Twitter suggested that he was opinionated, something that the company wanted on its initial team.
當然，還有其他的公司
Gild is not the only company now scouring for information. TalentBin, another San Francisco start-up firm, searches the Internet for talented programmers, trawling sites where they gather, collecting “data exhaust,” according to the company Web site, and creating lists of potential hires for employers. Another competitor is RemarkableHire, which assesses a person’s talents by looking at how his or her online contributions are rated by others.
And there’s Entelo, which tries to figure out who might be looking for a job before they even start their exploration. According to its Web site, the company uses more than 70 variables to findindications of possible career change, such as how someone presents herself on social sites. The Web site reads: “We crunch the data so you don’t have to.”
針對這些方法的疑問
The algorithm did a good job measuring what it can measure. It nailed Mr. Dominguez’s talent for working with computers. What is still unfolding is how he uses his talent over the long term, working with people. ...
Sean Gourley, co-founder and chief technology officer at Quid, a Big Data company, said that data trawling could inform recruiting and hiring, but only if used with an understanding of what the data can’t reveal. “Big Data has its own bias,” he said. “You measure what you can measure,” and “you’re denigrating what can’t be measured, like gut instinct, charisma.”
在另外一篇報導中 (註 3)，說明 IBM 以 13 億美元收購 Kenexa
Companies view work-force data mainly as a valuable asset. Last December, for example, I.B.M. completed its $1.3 billion acquisition of Kenexa, a recruiting, hiring and training company. Kenexa’s corps of more than 100 industrial organizational psychologists and researchers was one attraction, but so was its data: Kenexa surveys and assesses 40 million job applicants, workers and managers a year.根據 Kenexa 的資料分析，銷售員成功的要素為情緒勇氣 (emotional courage)
Tim Geisert, chief marketing officer for I.B.M.’s Kenexa unit, observed that an outgoing personality has traditionally been assumed to be the defining trait of successful sales people. But its research, based on millions of worker surveys and tests, as well as manager assessments, has found that the most important characteristic for sales success is a kind of emotional courage, a persistence to keep going even after initially being told no.
Google 發現最創新的員工是最快樂的
Since 2007, the company has conducted extensive surveys of its work force. Google has found that the most innovative workers — also the “happiest,” by its definition — are those who have a strong sense of mission about their work and who also feel that they have much personal autonomy. “Our people decisions are no less important than our product decisions,” Mr. Setty says. “And we’re trying to apply the same rigor to the people side as to the engineering side.”
Transcom 公司成功地應用在客服中心，降低離職率，以節省聘用成本
Transcom, a global operator of customer-service call centers, conducted a pilot project in the second half of 2012, using Evolv’s data analysis technology. To look for a trait like honesty, candidates might be asked how comfortable they are working on a personal computer and whether they know simple keyboard shortcuts for a cut-and-paste task. If they answer yes, the applicants will later be asked to perform that task.
Those who score high on honesty typically stay in their jobs 20 to 30 percent longer than those who don’t, Evolv says.
Neil Rae, an executive vice president of Transcom, was impressed with the project’s results and plans to use Evolv in the call centers he runs, which employ 12,500 workers.
In the call-center world, Mr. Rae says, 5 percent attrition a month — 60 percent a year — is stellar performance. Dropout rates are calculated at 30-day intervals, and it takes four to six weeks to train a worker. The cost of attrition — for hiring and training a replacement — is about $1,500 a worker, he says.
In the project with Evolv, Mr. Rae says, Transcom was able to hire fewer people — about 800 instead of a more typical 1,000 hires — to get 500 workers who were still on the job at least three months later. The big payoff, he says, should come in cost savings and better customer service with less worker churn in call centers.
(註 1) Matt Richtel, How Big Data Is Playing Recruiter for Specialized Worker, New York Times, April 27, 2013
(註 2) 從文章和網站看不出其如何計算。文章說明聘用 Mr. Dominguez 的方法中，其概念類似 PageRank，由別人的引用來說明其能力。如果有很多變數，通常使用線性迴歸以得到其相依的關係。
(註 3) Steve Lohr, Big Data, Trying to Build Better Workers, New York Times, April 20, 2013
沒有留言:
張貼留言