03/14/2022
When AI is applied to management, especially hiring, we
must better evaluate claims made about how the AI is "fair."
"Fair" is defined different ways by different people, and it's easy to talk past each other. The differences between how a job candidate, a psychologist, a computer scientist, a statistician, an ethicist, a legal system, and a society define "fairness" and "bias" can be vast.
In such a high-stakes context as hiring, using the same word to mean different things causes a lot of problems. It creates anger and spite. It breeds mistrust. "I think this is fair but you don't; you must be wrong! You must be evil!"
We can't afford to make this mistake when people's jobs are on the line. Thinking so narrowly holds everyone back. It is bad for job candidates, and it is bad for business.
This problem is much of what led Tara Behrend and TNTLAB PI Richard Landers to explore the concept of fairness across conceptual and disciplinary lenses, to develop standards for "psychological audits," as published in this new open-access article to appear in the flagship journal of the APA, American Psychologist:https://doi.apa.org/fulltext/2022-30899-001.pdf
A psychological audit is necessary any time an AI purveyor makes a psychological claim, like:
1) AI is claimed to measure something about a person, like their "hireability" or "potential" or "leadership skill."
2) AI is claimed to make predictions about a person, like their "future job performance" or "turnover risk."
3) AI is claimed to make meaningful decisions about a person, like if they will directed towards a leadership pipeline or need unique training and development.
Do not let your own biases about "bias" mislead you. There are many, many ways to evaluate whether a particular AI system is "fair," and your personal definition is probably a by-product of the field you were trained in rather than any "objective" truth about what that word actually means. In reality, there is no "true" definition of either fairness or bias.
Thus, anyone making a claim about fairness and bias in AI must be transparent about what they mean by those words. In turn, you need to be open to learning about what they mean by it, so you can evaluate what claims they are really making - and more importantly, what they are leaving unsaid and untested.
It will only be through transparency and open exchange of data and evaluative reporting that we can have any confidence in claims about what AI can do.
https://twin-cities.umn.edu/news-events/meaningful-standards-auditing-high-stakes-artificial-intelligence
In response to claims of unfairness and bias in tools used in hiring, college admissions, predictive policing, health interventions, and more, the University of Minnesota (U of M) recently developed a new set of auditing guidelines for AI tools.
