人工智能是一个“黑匣子”——神秘而且令人生畏。同时，人工智能技术新的排列组合如雨后春笋般层出不穷，在招聘领域尤其如此。然而，尽管雇主希望自己的员工组成更加多元包容，但人工智能行业本身却因为几乎都是白人男性而受到抨击。例如，纽约大学研究人员最近的一项研究指出，在像Facebook和谷歌这样的科技巨头中，女性和有色族裔员工的占比非常小，整个企业都面临着“多元化危机”。

讽刺的是，如果能够正确使用人工智能，它们“非常有希望在决策方面比人类做得更好，特别是在招聘工作中。”亚历山德拉·莫伊西洛维奇说。莫伊西洛维奇是IBM的人工智能研究员，拥有16项机器学习专利，协助开发了可以用于检查其他算法是否存在无意偏差的算法。她指出，想要利用人工智能鼓励多元化，重要的一点是要确保构建黑匣子的团队本身就是一个多元化的团队，确保这个团队拥有不同背景和不同观点。

“人工智能工具是否优秀，是否公正，取决于我们输入的数据。”莫伊西洛维奇说，“人工智能不是要取代人类的智慧，而是进行补充。”

人工智能可以帮助雇主找到并吸引来自于不同性别、年龄和种族的新员工。以下是四种主要方式：

人工智能知道如何针对最优秀的候选人宣传

招聘信息中的文字表述很重要，不仅因为文字表述经常无意中阻止一些潜在的雇员申请空缺岗位。“我们人类只能尽最大努力猜测怎么说会引起求职者的共鸣，但经常猜错。”人工智能公司Textio的联合创始人及首席执行官基兰·斯奈德说。

Textio公司用了大约5亿个真实招聘广告的数据，让人工智能分析这些广告在现实中得到的回应，建议公司应该使用和避免使用哪些词语。例如，对于客户eBay而言，“原先的经验”这一短语使男性申请人增加了50％。“但是，‘表现出来的能力’却多吸引了40％的女性，即使它和‘原先的经验’说的差不多是同一件事。”斯奈德说。

不同性别、种族、民族中性的语言“变化很快。没有‘请使用这10个词’的词汇清单。”她补充道，“但恰当的词可以在合适的时机吸引最多元化的申请者。”

人工智能扩大了符合标准的申请者范围

人工智能还能够在鞭长莫及的区域建立更广泛的网络。以校园招聘为例，雇主只能派这么多人去一部分校园进行招聘——但如果完美的求职者没有去这场人才招聘会，或干脆就去了另外一所学校呢？

人工智能公司HireVue（客户包括英特尔、甲骨文、道琼斯、唐恩都乐品牌等）的首席技术官洛伦·拉森表示，“非著名大学里的某名学生可能和‘正确的’大学里的学生一样好，甚至更好，而你可能根本就不会派人去这所大学校招。”

拉森说，在过去，这名学生得不到机会。但是通过人工智能获取潜在求职者的信息，并使用视频聊天等现代工具，你可以轻松与他们取得联系。“通过这种方式，可以让更多优秀的人进入系统里，这样你就可以‘看到’更多元化的求职群体，并进行评估。”拉森补充道。

人工智能是伯乐

简历是有效的招聘工具，但是“如果你专注于某位候选人简历上的内容，你就可能忽视很多其他人。”CareerBuilder的首席执行官伊利尼亚·诺沃谢利斯基说道，该公司的领导层现在拥有70％的女性和少数族裔，比诺沃谢利斯基在2017年加入时高了40%。

CareerBuilder网站使用人工智能帮助雇主和求职者进行最优配对，其数据库包括超过230万个招聘职位、1000万个职位、13亿技能点。算法完全瞄准某一工作所需的技能，找到有潜力、拥有这项技能的候选人——但这些候选人可能正在根据自己的背景申请其他工作。

“有些人简历中的大标题或最近一份工作不一定能够代表他们还可以做其他什么事情。”诺沃谢利斯基说。例如，客户服务代表需要耐心和解决问题的能力，“我们发现家庭医疗保健工作者拥有这些技能。没有人工智能，是不可能这么配对的。”

严格关注技能“自然会促进多样化，因为招聘标准对于每个候选人来说都是完全相同的，不分性别、种族、民族、年龄或其他任何因素。人工智能将所有无关紧要的东西剥离。”HireVue的洛伦·拉森说。大量的研究证实，在所谓的结构化面试中，面试人员询问每个候选人完全相同的问题，寻找完全相同的答案，能够最有效地消除无意识偏见。

问题是，人类面试官几乎做不到。“我们会觉得无聊，会走神，或者突然觉得牙疼。”拉森说，“人工智能从来不会。”

人工智能可以纠正自己的偏见

人们在工作时会不自觉地带入自己的经历、假设和偏好，其中一些怪癖尤其难以改变，特别是当他们潜伏在潜意识中时。相比之下，即使是最聪明的机器（至少到目前为止）也只能学习和运用程序员装进去的内容。其中可能包括，强调要欢迎所有年龄、性别和人种的最佳候选人。

“人类通常不能完全解释自己的决定，因为我们一定程度上依靠‘直觉’。”拉森说，“但是通过算法，我们可以查明无意的偏差存在于什么地方。”

HireVue的团队在一家客户公司尝试了一种算法，结果发现该算法更青睐具有深沉音色的求职者，因此，该算法在初步测试中，一直选择男性而不是那些同样称职的女性。与此同时，早期其他一些人工智能系统因为在视频采访中偏好浅肤色的求职者而招致批评。

拉森说，程序员已经学会发现并修复类似这种情况，她还说“数据驱动技术让我们有机会以前所未有的方式实现公平。”

这并不是说人工智能可以让人力资源专业人士和招聘经理退位。管理公司兼容并包的政策、与有潜力的候选人建立良好关系、确保人工智能在做本职工作，这些事情只能由人来完成。

正如IBM的亚历山德拉·莫伊西洛维奇所说：“所有的研究都表明，人类和人工智能相互配合比单打独斗更有效。”（财富中文网）

译者：Agatha

Artificial intelligence can a “black box”—mysterious and more than a little intimidating. Meanwhile, new permutations of the tech are sprouting up like mushrooms, especially for recruiting and hiring. Yet as employers have increasingly tried to make their workforces more diverse and inclusive, the A.I. industry itself has taken some flak for being almost exclusively white and male. For instance, a recent study by New York University researchers points out that at tech giants like Facebook and Google, such tiny percentages of employees are female or nonwhite that the whole business is suffering a “diversity crisis.”

The irony there is that A.I., used correctly, has “a shot at being better at decision-making than we humans are, particularly in hiring,” says Aleksandra Mojsilovic. A research fellow in A.I. at IBM, Mojsilovic holds 16 patents in machine learning, and helped develop algorithms that can check other algorithms for unintended bias. An essential part of using A.I. to encourage diversity, she notes, is making sure the teams that build what goes into the black box are themselves a diverse group, with a variety of backgrounds and points of view.

“Any A.I. tool can only be as good—and as impartial—as the data we put in,” Mojsilovic says. “It’s not about replacing human intelligence, but rather about complementing it.”

A.I. has helped companies find and attract new hires of all sexes, ages, and ethnicities. Here are four main ways it’s helped them to do that:

A.I. knows how to speak to your best candidates

The words in job postings matter, not least because they often unwittingly discourage some potential hires from applying. “We as humans take our best guess at what will resonate with job seekers, but we’re often wrong,” notes Kieran Snyder, cofounder and CEO of the A.I. firm Textio.

Using a dataset of about 500 million actual job ads, and A.I. that analyzes the real-life responses they got, Textio advises companies on which words to use—and avoid. At client eBay, for instance, the phrase “prior experience” drew a 50% increase in male applicants. “But the phrase ‘demonstrated ability’—even though it means essentially the same thing—attracted 40% more women,” Snyder says.

Language that is neutral across sexes, races, and ethnicities “changes rapidly. There is no ‘use-these-10-words’ list,” she adds. “But the right word at the right moment does attract the most diverse possible group of applicants.”

A.I. widens the pool of eligible workers

A.I. also has the power to cast a wider net across unmanageable geographies. Take, for example, campus recruiting. Employers can send only so many humans to a limited number of campuses—but what if the perfect hire skipped the job fair, or goes to a different school entirely?

“A student at an obscure college where you’d never send a recruiter could be every bit as good as, or better than, graduates of the ‘right’ schools,” observes Loren Larsen, chief technology officer at A.I. firm HireVue, which lists Intel, Oracle, Dow Jones, Dunkin’ Brands, and many others among its clients.

In the old days, says Larsen, this student wouldn’t have gotten a second sniff, let alone a first. But by sourcing the leads with A.I., and using modern tools like video chatting, you can reach them with ease. “This way, a lot more people are let into the system on their merits, so you get to ‘meet’ and assess a much more diverse group of candidates,” adds Larsen.

A.I. has an eye for talent—and skill sets

Resumes are nice, but “if you focus on what it says on someone’s resume, you risk overlooking huge numbers of people,” says Irinia Novoselsky, CEO of CareerBuilder, whose top leadership is now 70% women and minorities—up from 40% when Novoselsky joined in 2017.

The site uses A.I. to help employers and job hunters find the best match, with a database that includes more than 2.3 million job postings, 10 million job titles, and 1.3 billion skills. The algorithms zero in on exactly what skills a job requires, and find promising candidates who have them—but who may, based on their background, be applying for a different job altogether.

“Someone’s resume headline or most recent role may not necessarily translate into what else they can do,” says Novoselsky. Customer service reps need, for instance, patience and problem-solving ability, and “we’ve found that home health care workers share those skills. Without A.I., making those matches would have been impossible.”

A strict focus on skills “naturally leads to more diversity, because the hiring criteria are exactly the same for each and every candidate, regardless of sex, race, ethnicity, age, or anything else. A.I. strips out all that extraneous stuff,” says Loren Larsen at HireVue. Reams of research confirm that so-called structured interviews, where interviewers ask precisely the same questions of each candidate and look for precisely the same checklist of answers, work best at eliminating unconscious biases.

The catch is, human interviewers rarely do them. “We get bored, or we’re distracted, or we have a toothache,” Larsen notes. “A.I. never does.”

A.I. can correct its own biases

People can’t help bringing their own experiences, assumptions, and preferences with them to work in the morning, and some of those quirks—especially when they lurk in the subconscious—are notoriously slow to change. By contrast, even the smartest machines (at least so far) can learn and apply only what programmers install in them. That can include an emphasis on welcoming the best-qualified candidates of all ages, sexes, and colors.

“Humans often can’t fully explain their decisions, because they’re going partly on ‘gut feel,'” says Larsen. “But with algorithms, we can pinpoint exactly where an unintentional bias has sneaked in.”

At one client company, HireVue’s team tried out an algorithm that turned out to be biased toward job applicants with deep voices so that, in preliminary testing, it kept selecting men over women who were just as qualified. Meanwhile, other, earlier A.I. systems have drawn fire for favoring light skin tones over darker ones in video interviews.

Larsen says programmers have learned to spot—and fix—that sort of thing, adding that “data-driven technology gives us the chance to keep getting more fair in ways that weren’t possible before.”

That’s not to say that A.I. can ever push human resource professionals and hiring managers to the sidelines. The tasks of managing company policy on inclusion, building great relationships with promising candidates, and making sure that A.I. is doing its job can only be done by people.

As Aleksandra Mojsilovic at IBM puts it, “All the research shows that humans and A.I., working together, are far more effective than either alone.”