NYC to Audit AI Hiring Systems | Roundup #4
Plus: Rideshare drivers in California are not earning enough to meet minimum wage
Welcome new readers! This is a weekly roundup where I share the 2-3 most interesting articles, quotes, charts, or other tidbits from what I’ve read in labor topics the past week. If you’re trying to quickly come up-to-date on news in labor policy, economics, and how technology is affecting workers, this is for you.
On to the update!
NYC Aims to Root Out AI Bias in Hiring
From Richard Vanderford at the Wall Street Journal:
A New York City law that comes into effect in January will require companies to conduct audits to assess biases, including along race and gender lines, in the AI systems they use in hiring. Under New York’s law, the hiring company is ultimately liable—and can face fines—for violations…
AI has steadily crept into many companies’ human-resources departments. Nearly one in four uses automation, AI, or both to support HR activities, according to research that the Society for Human Resource Management published earlier this year. The number rises to 42% among companies with more than 5,000 employees.
Using AI to augment hiring might eliminate certain types of biases, like when human interviewers make snap judgements about a candidate based on their alma mater. However, AI hiring products — such as HireVue — may replace those biases with new ones, at scale. From the same article:
One audit of HireVue’s algorithms published in 2020, for example, found that minority candidates tended to be more likely to give short answers to interview questions, saying things like “I don’t know,” which would result in their responses being flagged for human review. HireVue changed how its software deals with short answers to address the issue.
The way I understand HireVue’s product, an interviewee flagged for human review is significantly less likely to continue to the following stage. Is a respondent who says “I don’t know” any less qualified than someone who makes up a well-spoken but ultimately BS answer that wouldn’t have been caught by HireVue? Probably not.
AI systems tend to inherit many of the implicit biases of their creators and the training data they are used on. GPT-3, the large language model created by OpenAI and a core technology being used by multiple startups, has known racist and sexist biases, in part because its training data includes much of the internet (which has known racist and sexist biases). In a similar way, if systems like Hirevue are trained on a company’s historical hiring decision data (which may have been biased), it is going to lead to bad outcomes (computer scientists refer to this affectionately as ‘garbage in, garbage out’).
The NYC regulation mandates an AI audit for companies that use automated hiring tech, although the city’s Department of Consumer and Worker Protection hasn’t finalized the rules that will implement the law. Because an AI audit isn’t yet very well defined (in contrast, say, to a financial audit), it will take some time to figure out the right framework. But it is great to see the city experiment in this arena, which will hopefully prompt more cities to think about regulation and industry groups to think about self-regulation.
Do California rideshare drivers meet California minimum wage after Prop 22 and inflation?
Aarian Marshall at Wired reported on a study that was conducted by PolicyLink and Driver’s Seat Cooperative on California rideshare drivers on their net earnings, after the state implemented Prop 22 and record inflation during the pandemic:
Drivers made a median of $26.30 per hour in earnings, tips, and bonuses, the researchers found. But driving looked less well compensated when they subtracted some of the costs associated with being self-employed that might otherwise be covered by an employer, such as the cost of owning or leasing and operating a vehicle, taxes, unemployment insurance, paid sick time, and paid rest time. The result was an effective hourly wage less than half of California’s minimum wage… [t]hey make a hourly wage of $6.20…
The fixed and operating costs associated with rideshare drivers are much higher than most people — often even drivers themselves — realize. A study commissioned by the Seattle city government and conducted by researchers at UC Berkeley and the New School reported that, for Seattle drivers:
Current gross driver hourly pay is approximately $21.53. After expenses of $11.80, a driver nets $9.73 an hour, much less than the minimum wage.
In this scenario, net hourly pay was 45% of gross hourly pay. I honestly wasn’t expecting California drivers’ net hourly pay to be just 25% of their gross hourly pay, and admittedly the driver sample size they are looking at is quite small, but I could imagine this is due to significant inflation of vehicle acquisition costs, maintenance costs, and gas since the Seattle report came out, and how significant these components were in rideshare drivers’ expenses even back then:
At any rate, even applying the 45% ‘profit margin’ of Seattle to California drivers in this study still wouldn’t get them past the California minimum wage of $15/hr.
Rideshare drivers and delivery couriers lack transparency into their net pay — and Uber, Doordash, and Lyft certainly have no incentive to show their workers anything other than the gross pay, lest their workers switch over to another platform. Some may argue that the responsibility for calculating net pay falls entirely on rideshare drivers. Yet so many are left confused by misleading messaging from Uber and Lyft around making “$30/hr,” which would certainly be a gross earnings figure and also not incorporate all working time (see an older Workonomics post on this issue).
A better wage floor and universal benefits for gig workers is the obvious solution here. But gig companies could also be required to show drivers estimated net earnings information when giving drivers trip requests. They could also disclose to the public the distribution of estimated net earnings for drivers and couriers, so prospective gig workers can understand what they’re getting into. We shouldn’t forget how good old fashioned design changes can help significantly improve worker outcomes.
This is just another example that we need our institutions to better uphold standards on wages and working conditions for precarious workers so that people can make better economic decisions.
Thanks for reading Workonomics! Subscribe for free to receive new posts and support my work.
Updates on Past Workonomics Roundups
Bloomberg Law: The Art of Settling But Not Resolving Gig Worker Status Disputes [update to Roundup #3: “Uber Finally Pays NJ $100M for Unemployment Insurance”]
The Guardian: California’s fast-food industry calls for ballot proposition overturning new labor legislation [update to Roundup #1: “California’s New Bargaining Council for Fast Food Workers”]
Quarterly Journal of Economics: Does Welfare Prevent Crime? the Criminal Justice Outcomes of Youth Removed from Ssi
Quarterly Journal of Economics: Labor Market Returns and the Evolution of Cognitive Skills: Theory and Evidence
Other News & Perspectives
Noah Smith: American workers need lots and lots of robots
Full Stack Economics: I'm a professional dad who "leaned out" to support my wife's career