Shulou(Shulou.com)06/02 Report--

By Tomas Chamorro-Premuzic, Frida Polli, Ben Dattner

Translated by: wwl

Proofread by Wu Jindi

This article is about 2800 words, it is recommended to read for 5 minutes.

In talent management, AI models can better help reduce bias against potential employees and increase diversity and inclusiveness in organizational hiring than relying on manual screening and evaluation by hiring managers.

AI has disrupted every area of our lives-from carefully curated shopping experiences by companies like Amazon and Alibaba to channels like YouTube and Netflix using personalized recommendations to market their latest content. But when it comes to the workplace, in many ways AI is still in its infancy. This is especially true when we consider that it is beginning to change the way talent is managed. To use a familiar metaphor: AI in the workplace is in dial mode. The 5G WiFi phase isn't here yet, but we have no doubt it will.

Admittedly, there is a lot of confusion about what AI can and cannot do, and how to define it. However, AI plays a very special role in the talent war: providing organizations with more accurate and effective predictions of candidates 'professional behavior and performance potential. Unlike traditional recruitment methods, such as employee recommendations, resume screening, and face-to-face interviews, artificial intelligence can discover features that the human eye cannot capture.

Many AI systems use real people as successful models for specific roles. The set of such people, as the training dataset, typically includes managers and employees defined as "high performers." The AI system processes and compares the profiles of various job seekers with the "model" employees it creates from the training set. It then provides the company with a probability estimate of how well the candidate's attributes match those of the ideal employee.

In theory, this approach could match the right people for the right jobs faster and more efficiently. But, as you may have realized, it has become a source of hope and danger. If the training set is diverse, if unbiased demographic features are used to describe the people in it, and the algorithm is unbiased, then the technique can indeed mitigate human bias, expand diversity, and socioeconomic inclusion better than humans. However, if the training set, the actual data, or both are skewed and the algorithm is not adequately vetted, then AI will only exacerbate bias in recruitment and homogeneity problems in the organization.

In order to rapidly improve talent management and harness the full potential of AI, we need to shift our focus from developing more ethical HR systems to developing more ethical AI. Of course, eliminating bias from AI is not easy. In fact, it's hard. But our argument is based on our belief that it is far more feasible than removing it from humans themselves.

Most organizations still do their best to identify talent or potential. Recruiters only need to take a few seconds to look at resumes before deciding who gets eliminated. Hiring managers rely on cultural fit to hire, using so-called "intuition" or ignoring hard data--a problem exacerbated by a general lack of objective, rigorous performance measures. In addition, more and more companies are implementing unconscious skewed data training that is often found to be ineffective and sometimes even worse. Usually, training focuses only on individual differences and ignores structural skewness of classes with small sample sizes.

While critics argue that AI isn't better, they often forget that these systems mirror our own behavior. We're quick to blame AI for predicting that white (and possibly white male) managers will receive higher performance ratings. But this happens because we don't treat the skewed distribution of performance ratings in the training data appropriately. We are shocked by the skewed hiring decisions AI gets, but living in a world dominated by human bias can be just fine. Just look at Amazon, where critics of their biased hiring algorithms ignore overwhelming evidence that human-driven hiring in most organizations today is inevitably worse. That's an expression of concern about driverless car deaths--more than 1.2 million people die each year on the roads due to vehicle defects, distracted drivers and drunk driving.

In fact, AI systems have a better ability to guarantee accuracy and fairness than they do to influence incentives for recruiters and hiring managers. Humans are good at learning but bad at forgetting. The cognitive mechanisms that make us biased are often the tools we use to survive in our daily lives. The world is too complex for us to process logically and consciously all the time; if we did, we would be overwhelmed with information overload and unable to make simple decisions like buying a cup of coffee (after all, if we don't know the barista, why should we trust him?). That's why it's easier to ensure that our data and training set are unbiased than to change Sam's or Sally's behavior, and for a person, we can neither eliminate bias nor really capture the actual output of variables that influence their decisions. Essentially, analyzing AI algorithms is easier than understanding and changing human thinking.

To that end, organizations using AI for talent management at any stage should start with the following steps.

Educate candidates and obtain their permission. Prospective employees are questioned and their personal data is provided to the company, which is analyzed, stored, and applied to artificial intelligence systems for HR decisions. Be prepared to explain to them what, who, how, and why. Because artificial intelligence systems that rely on black box models are immoral. If a candidate has attributes that correlate with success, then not only do you need to understand why, but you also need to explain cause and effect. In short, AI systems should be designed to predict and explain causation, not just to discover correlations. In addition, you should guarantee the candidate's anonymity in order to protect personal data and comply with relevant legal provisions. Invest in systems that optimize fairness and accuracy. Historically, organizational psychologists have pointed out that accuracy declines when candidate evaluation models are optimized for fairness. For example, a large number of academic studies have shown that cognitive ability tests are consistent with job performance, especially in high-complexity jobs. This distribution adversely affects underrepresented individuals, especially those with low socio-economic status. In other words, if companies want to promote diversity and create a culture of inclusion, they need to pay less attention to cognitive testing when hiring new employees, so that diverse candidates are not disadvantaged in the hiring process. This is called fairness/accuracy trade-off.

But this trade-off relies on technology that is half a century old, long before artificial intelligence models emerged that can be used to treat data differently from traditional methods. There is growing evidence that AI can overcome this trade-off by deploying dynamic, personalized scoring algorithms that are equally sensitive to accuracy and fairness and can work together to achieve optimization. Therefore, there is no reason why AI developers should not do so. Moreover, since these new systems already exist, we should question whether widely used traditional cognitive assessments (which adversely affect minorities) should continue to exist in ways that do not mitigate bias.

Develop open source systems and third-party reviews. Hold companies and developers accountable by allowing others to evaluate this tool for analysis. One solution is open source, non-proprietary, but key parts of the AI technology are owned by companies. For proprietary components, companies can leverage third-party reviews conducted by trusted experts in the field to demonstrate to the public how they mitigate bias. Comply with the same laws regarding data collection and application as traditional recruitment. AI systems should not use any data that is not permitted to be collected or included in traditional recruitment processes for legal or ethical reasons. Personal information about physical, mental or emotional conditions, genetic information and drug use or abuse should not be entered.

If organizations address these issues, we believe that ethical AI can not only reduce bias in hiring, but also enhance talent management, making the link between talent, effort, and employee success far greater than in the past, thereby greatly improving organizations. Moreover, it would benefit the global economy. Once we reduce bias, our candidate pool will not be limited to employee referrals and Ivy League graduates. People from broader socioeconomic backgrounds will have more access to better jobs--which helps create balance and begin to bridge class divides.

However, to achieve these goals, companies need to make the right investments, not only in cutting-edge AI technologies, but also (especially) in human expertise-people who know how to take advantage of the benefits these new technologies offer while minimizing potential risks and drawbacks. In any field, the combination of AI and human intelligence has the potential to produce better results than without AI. Moral AI should be seen as one of the tools we can use to fight our own biases, not as the ultimate panacea.

Original title:

Building Ethical AI for Talent Management

Original link:

https://hbr.org/2019/11/building-ethical-ai-for-talent-management

Editor: Wang Jing

Proofread by: Yang Xuejun

About the Translator

Wang Weili, BI practitioner in the elderly medical industry. Keep learning.

- End-

https://www.toutiao.com/i6772131488432390668/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.