AI is now a top political issue. Can we beat China in the AI game? Will this “frontier” AI destroy humanity? Where will we put the data centers necessary for it to do so? These issues have occasionally perplexed Congress.
Almost three years ago, then-Majority Leader Chuck Schumer (D., N.Y.) convened his first “AI Insight Forum” so that senators could hear from the brightest lights of the tech world on how to approach AI.
I attended the first Insight Forum and was struck not by the remarks of Elon Musk or Bill Gates, but rather by Schumer’s “civil society” invitees, mostly liberals taken from the civil-rights industrial complex.
While the tech gurus were focused on big, even existential, questions about AI and the future of humanity, the civil rights activists were focused on AI applications and, in particular, making sure that their client groups got a piece of that action. That’s why, eight months later, when Schumer unveiled his “bipartisan AI roadmap” my only question for the Republicans involved was, “did you keep disparate impact out of it?”
Machine learning allows decisionmakers to sort through data without taking into account ideological pieties, especially in areas like hiring. The way the diversity game works is to allow non-quantifiable factors correlated with preferred “diverse” outcomes to trump quantifiable factors. It stands to reason that machine-learning sorting algorithms won’t do that. The only way, then, to reflect the HR nostrums of the diversity era is to instruct the algorithm to consider illegitimate inputs (like race, sex, or age). That was clearly what Schumer’s allies wanted, and why we consistently warned Republicans that issue of “application bias” was an invitation to put Kristen Clarke in charge of our iPhones.
Luckily nothing ever came of that legislatively, but that doesn’t seem to be stopping liberals and the courts from going forward with it. Take the case currently being litigated in San Francisco’s federal court, Mobley v. Workday. The case has been going on for years and is currently on its fourth amended complaint. The claim is, essentially, that Workday’s machine-learning applications for sorting job candidates has a disparate impact on disfavored groups, including women, the old, and racial minorities.
The supposed problem is that these algorithmic systems are used “in lieu of individualized human review” and they therefore “reproduce and amplify disparities embedded in prior hiring decisions and labor market structures.”
In reality the problem is that the algorithmic systems are not trained to engage in the “right sort” of explicit bias. As the plaintiffs explain, “The absence of adequate safeguards — such as bias testing, transparency mechanism, or corrective constraints — creates a substantial risk that the algorithmic systems will drift over time…produc[ing] increasingly discriminatory outcomes.” In other words, companies can’t use objective and tailored hiring mechanisms without an overlay of an algorithmic Eric Holder to make sure they get the “right” outputs.
Apart from being a ridiculous argument as a matter of public policy, the Mobley case has underlying legal issues that should terrify people concerned with actual fairness and economic competitiveness. Disparate impact is increasingly disfavored under federal law. The Office of Legal Counsel took a flamethrower to it under Title VII, while the Department of Housing and Urban Development has done the same under the Fair Housing Act. But the cause of action in Mobley is under California’s Fair Employment and Housing Act (FEHA).
According to the Mobley plaintiffs — so far with the assent of a President Joe Biden-appointed judge, Rita Lin, Workday’s software is subject to the FEHA because Workday develops and runs it in California.
Never mind that the software is used nationwide, California’s bleeding-edge leftist laws on discrimination are what control. If you’re a small business in Arkansas or Kentucky using Workday’s technology to screen applicants, you need to be subjected to California’s reverse-discrimination mandates when you do so because that’s where the software was developed and serviced.
It’s nothing short of an effort to use the historical accident of Silicon Valley being in California to force California’s progressive race and discrimination policies on the rest of the country.
In many ways this is just another aspect of the left’s new frontier of using state law in court to regulate industry nationwide. The first case the Supreme Court will hear this term, Suncor v. Boulder, will answer the question whether state law in state courts can be used to control the national oil and gas industry. It’s the same overarching progressive strategy that’s being used in Mobley.
The big questions on AI won’t be resolved any time soon. In the meantime, though, the applications of machine learning will become increasingly important to our daily life. Should those applications be under the monitorship of a robot Eric Holder to ensure that their outputs are what the left considers “fair” regardless of what the actual inputs suggest? Of course not. And that’s why the Mobley case bears a close watch from those who want actual fairness and economic strength.
Michael A. Fragoso is a partner at Torridon Law PLLC. He previously served as chief counsel to Republican Leader Mitch McConnell and on the Senate Judiciary Committee.
