| 2/2 Step 2: peer-review
The highest candidate is selected, interviewers exchange notes and are asked to: * check and discuss if their colleagues reasons’ are valid and reasonable (for example "the hello" reason might be seen as too petty by some, and will be discarded) check if the point values are reasonable
(Checking old similar data to flag outliers is a possibility) check if some of the reasons presented by other colleagues apply to their own candidates. For example, if the "hello" reason was not discarded and your candidate didn't say hello you must apply it to him too. *Interviewers are asked to bring more focus on the current best candidate's reasons. If after a peer-review round, the best candidate changes, another round is made or shorten the list and re-interview. If it does not change, the best candidate is offered the position. Claim D: this helps smooth out harsh or lenient interviewer bias. Claim E: this process if well documented should be a valid and easy defense against allegations. Data Analysis on hiring decisions becomes way more interesting. I'm sure there are tons of trends you can seek out. Claim F: this allows an earlier detection of discrimination. For example if you find that an interviewer or a committee always removes points from a certain group for a subjective reason more than others in the company. It might be an early warning sign that a problem need be addressed. Claim G: this allows for a higher quality debate on controversial issues like sexism and racism in hiring. Machine learning can be used to give suggestions to interviewers about the amount of points to give for reasons. (They should be suggestions no decisions to avoid the pitfalls the lost nuance that classic machine based decision might suffer from) Example: a fairly easy one is that after a high enough number of universities and GPAs is received one can aggregate the point values into scores that take into account the difference of grading and quality between universities. This might help decide if GPA x in Y is better or worse than GPA z in C, a very hard question to answer fairly if we didn't have the data. I think this method offers a more traceable, open and perhaps fair way of hiring without suffering from the lack of nuance that traditional automatic hiring suffers from. Problems: - Money: this process might require more man power. - Tooling: to be efficient this process requires tooling and automation I would be happy to hear your ideas, improvements and experiences with similar systems. I think the idea of this system is very similar to that of a neural network. |