David Green ????的动态

I believe any algorithmic output should be vetted with an open mind as there are aspects which algorithms are blind to... algorithms are only as good as the data you feed it with

How much variability in performance does the algorithm explain? Has the algorithm been validated as predictive or is the equation predicting better job fit based solely on concurrent data? These are important differences. If an algorithm explains 50% of variability in job performance, then the other 50% of the decision should be based on human intelligence. The algorithm then has an appropriate supportive function. The BCG Utility equation can be used to forecast ROI once a sufficiently powerful sample and statistically-significant correlation have been obtained.

David Green, I think considering the algorithm's recommendation but also defining exactly what one's trying to achieve in filling the role, something the algorithm is not likely to know and finally utilizing an independent committee approach like Google is best answer.

What about involving the team that will be working the that new manager to select him/her? I missed this option in the whole article! Analytics tools should be a support and challenge for those taking the decisions, but to taken as the only tool as I think some personal characteristics and incompatibilities among humans still cannot be replaced by algorithms and do not appear with the people analytics tools.

When I make a hiring decision I try to review all data points available. This includes what we learned during the interview process and the opinions and instincts of all those involved. However the data points also include any insights collected by the various tools we may use. In the end it is a blend.

Thanks for sharing. I totally agree with the message of the article that you also highlight at the end of your post. On the other hand, I am wondering whether they also analyzed the quality of the collaboration beside the frequency of the emails. It is hard to judge whether a connection in an email network is generating value to the company or purely overloading others (e.g. internal spam and trolling). Adding qualitative data to these connections either by analyzing the texts of the emails or by including other data sources like real-time feedback between employees would probably lead to even better algorithms to judge the value of someones personal network. We published an article a few years ago in Social Networks about the comparison of on-line and off-line networks within an organization. Some interesting findings can be found here: https://www.sciencedirect.com/science/article/pii/S0378873312000123

Several articles, TED talks and tweets from Zeynep Tufekci come to my mind with AI algorithms propagating biases from its inputs. Before decision is made using the recommendation based on the algorithm, she needs to understand methodology better and what data was used.

From the limited information of the case - exactly the design of the algorithm is to predict performance, predict assimiliation (ONA), predict potentials, predict actualization, or predict what else? Without a clear understanding of the design, no evidential based conclusion can be drawn fairly and justifiably. ??????

Whatever one’s decision, as long as the tools used are evidence-based, have been properly validated, and will withstand legal scrutiny, there is no one right way...

Yes an algorithm should tell us who to promote, that's called transparency, and the algorithm should be known to all in advance.?Algorithm should be used to determine pay level and most of the things related with traditional HR. Transparency. That's how you curb cronyism and nepotism and promote merit in organizations.