Where better than universities?

Who is responsible for the decisions of AI agents? How can we protect job applicants from AI bias? And what does due process mean when a “black box” is at the center of the case?

These are thorny questions with no clear answers, but they must be reckoned with, says Ifeoma Ajunwa, a faculty associate at Harvard’s Berkman Klein Center for Internet & Society.

In this edited interview, Ajunwa — who is also the Asa Griggs Campbell Professor of Law at Emory Law School, the founding director of Emory’s AI and the Future of Work program, and an AI ethics consultant with Fortune 500 companies — explains how scholars are approaching a new area of law.

---

Gazette: When the AI era first kicked off, how did the legal profession start thinking about AI and the law? And how has that evolved in the last four to five years? 

Ifeoma Ajunwa: The first legal questions really centered around the idea of AI as a black box, which is very problematic for the law because we like to understand who’s responsible. We like to understand proximate cause, and that’s something that a black box can make rather difficult. We saw prominent academics like [Cornell University Professor of Law] Frank Pasquale writing the “The Black Box Society,” which was questioning AI’s role in the financial industry, particularly in terms of adjudicating who had access to credit and other types of shadowy financial trading where you might have the involvement of data brokers. 

Then we started to have questions about liability. How do existing laws operate when AI is serving as an intermediary, say, for housing or hiring? How can the law still operate well to maintain its objectives of equal opportunity and equal access? Whom do we hold liable? 

The third wave that we’re seeing now is how we think about governing AI wholesale. Liability remains a very valid concern, especially now that we have AI agents that can run amok. But we can move to a bigger discussion about design and implementation, what sort of use cases we want to allow and disallow, what rules and regulations we want in place to make sure AI runs as intended. How can the law play a role in bringing about a better vision of AI in society? 

Now I have to put your question back to you: How can the law play a role in bringing about a better vision of AI in society? 

Unfortunately, a lot of American law is obsessed with ex post enforcement — that is, something horrible happens, and we go to court and we litigate and we find out who’s responsible, and maybe there’s some sort of civil sanction. We ought to put more emphasis on ex ante regulation — we need to have the regulatory systems in place before we need them. And there are various entry points where we can have AI regulation or AI governance; it can’t just start and end with the legislature. 

We can think of AI governance at the education level, which is something that I think about a lot in my position: Having students understand the role of AI in society and how that might intersect with some of our privacy concerns, some of our ethical concerns. What are the bedrock legal principles we want to protect and how can we regulate AI technologies to make sure they comport with those principles? 

Beyond education, we want to think about AI designers themselves. How can we encourage industry norms where AI is created with society, and not just profit, in mind? And so on. 

You’ve written extensively about AI’s implications for employment law, including in the 2023 book “The Quantified Worker.” Can you walk us through the core argument you’re making with this work? 

Even before the Constitution, one of the bedrock principles that America was founded on was as a place where people can profit from their labor — where anyone, regardless of their birth, is able by the sweat of their brow to make their way. I argue that AI and other forms of big data are shortchanging that employment bargain. When workers are hired, they’re told they’re exchanging their labor for money, but now they’re being called upon to hand over personal data as well. Where is the compensation for that? 

A lot of my legal scaffolding goes back to legal philosophers like John Rawls: the idea that we have to think about the most vulnerable in society when we’re creating laws. In this AI revolution, are we further disadvantaging the most vulnerable in society, or can we think about how to strengthen our laws to allow those people to flourish? 

I’m reminded of your writing on this idea of the “sacrificial lamb”; can you say more about that? 

If we hearken back to some of the early days of Silicon Valley, there was this mantra: “Move fast and break things.” And unfortunately, we learned that those “things” can turn out to be people. What I saw specifically in AI and hiring was that it was minorities, it was women who were the sacrifices. Many studies have shown that African American workers are most impacted. I don’t think it’s a coincidence that the most important case right now in AI hiring, Mobley v. Workday, involves an African American man — and his legal team is trying to get that lawsuit certified as a class action now. 

Essentially, that Silicon Valley mantra turns out to mean that you need to make some sacrifices to make progress. My question is, who is being sacrificed? It’s one thing if the sacrifice is borne equally in society. It’s another thing if the sacrifice is the same people over and over again. I don’t think “move fast and break things” should dismiss the idea that we still want a society that has equal opportunity. 

I agree that it’s unfair for one group or certain groups to be harmed disproportionately, but I’m not sure what you mean by a sacrifice being borne equally. 

You’re right: There are some sacrifices that we, as a society, should not be willing to make. For example, it was heartening to see the case against Meta for, essentially, sacrificing children to the algorithm. We should never say, “All children are being sacrificed equally, so it’s OK.” It’s not OK. We don’t want our children being sacrificed in the name of technological progress. 

I wonder if one reason so many people feel unsettled in this moment is that AI and all its attendant ethical questions seem to require a kind of thoughtful debate that can be hard to come by in our digitally mediated age. 

That’s really the heart of it: How can we effectively regulate AI if we can’t agree on governance, any governance? We have this emerging technology that seems to be something like a revolution, and we’re meant to grapple with it and corral it for our own benefit, but how can we do that if we’re not united? We as a society need to break out of our echo chambers and the algorithms that are reinforcing them.

Unfortunately, in this environment, even people who have very nuanced beliefs might not be able to have thoughtful conversations with one another. I do see that spilling over into the AI governance debate, where some people are trying to make it a partisan issue when it’s clearly not. As humans, we need to figure out how to regulate AI; that’s just a common-sense statement. Really, with so many security breach incidents, we must consider regulation as an existential need. 

If not on social media, where do you see these kinds of thoughtful conversations happening most effectively? 

I want to say that I see some pockets of it at universities. I do appreciate it when I go to places like the Berkman Klein Center, for example, which has committees and subcommittees working on AI for the public interest. 

You know, in a country with a multiplicity of beliefs, religions, ideologies, and also with a separation of church and state, I think the university becomes a kind of church: It serves that function of bringing people together for moral debates, for intellectual inquiry about the best way to run a society. The university has to serve that purpose, because otherwise we’re going to have a splintered society with all of us in our own enclaves — and that would defeat the idea of the multicultural, pluralistic society that is at the basis of our founding.