Stapleford speaks on the role of virtue ethics in AI, new AI center on campus
Thomas A. Stapleford is an Associate Professor in the Program of Liberal Studies, a Concurrent Associate Professor in the Department of History, and a faculty member in the Program in History and Philosophy of Science. His research interests include history of science, history of economics, American political economy, policy history, philosophy of science, and virtue ethics. He is the recipient of a National Endowment for the Humanities (NEH) grant, leading an investigation into the use of virtue ethics in generative artificial intelligence systems.
Editor’s note: This interview was edited for the sake of brevity and clarity.
Irish Rover: Where are we now? What are the stakes of ethics in AI development and what sort of ethics currently drive its development?
Professor Stapleford: It would be helpful to go back and think about ethics. The Greek root of ethics is ethos, which is thinking about character, a habitual way of being. When Aristotle writes about ethics, he’s thinking about what kind of habitual way of acting makes for an excellent person. Since our actions aren’t random, they’re going to be governed by some set of dispositions. This influences the motivations that we have, what we find important when we encounter the world, and the way that we reason about the situations we encounter.
So, we take that as a framework. When we ask the question, “What’s the relationship between AI and ethics?”, rather than thinking of ethics as being this narrow space where we ask questions about what is right or wrong, we’re asking how AI shapes us as human beings by affecting our habits of thought, our habits of feeling, and the kinds of motivations that we have. Does it shape us to flourish as human beings, or does it diminish us or damage us in different respects?
This lends itself naturally to talk about virtues. The word virtue comes from the Greek word arete, which means excellence. A virtue-based approach to AI thinks about the excellences of human beings and the way in which AI enables or prevents us from realizing them. For the projects that I’m involved with, we also want to think about AI systems in those same terms: What makes for an excellent AI system? What habitual ways of interacting make an excellent AI system? That brings us to where we started: An excellent AI system is one that’s going to help human beings, realizing their own excellences and not hinder us.
You asked a question about what AI ethics more broadly looks like now. It’s a big space, but the way in which it most often plays out is in terms of setting up guardrails on systems. We don’t want AI systems giving unreliable answers or misinformation. We don’t want AI systems that are going to be unsafe, or that might teach people how to design nuclear weapons or that might recommend self-harm.
We also want systems that aren’t going to be unjust. We don’t want them to be racist or sexist. So [AI ethics is] like guardrails around the system preventing it from doing these bad things.
But the guardrail approach does not touch all the questions that we were thinking about earlier. There are companies right now that advertise and sell “grief bots.” Basically, someone you love passes away and dies, but you could take this AI system and train it on all the stuff that it has written, or conversations that you have recorded, and then the AI will be able to respond as if it were that person who died.
Now, as long as it’s not telling you to harm yourself or someone else, it doesn’t seem unsafe. As long as it is sort of accurate in representing this person, it’s not misinformation or hallucinating. Let’s assume for the sake of argument that it is not going to be unjust, or say anything racist or problematic. That still leaves the question: Is this a good thing? It’s a question that is not addressed, let’s say, by the “guardrails vision” of AI ethics.
We’ve touched on how AI may diminish human flourishing, but what are some real ways in which AI could potentially cultivate human flourishing?
Yeah, that’s a great question. There’s a lot we’re going to discover and it’s hard to predict. It can be super helpful for doing tasks that are just tedious, where the task doesn’t help us cultivate character virtues or intellectual virtues.
But there’s a lot of things that we think are tedious that actually teach us something important. Transcription is a good example of this. Let’s say you thought, “I don’t need to take notes on my classes. I’ll just use my phone and record the lecture.” But we know from empirical research that taking notes on a lecture actually helps you process, understand, and retain material in ways that don’t happen if you aren’t writing it down.
A virtue-based approach to ethics entails not just making judgments about what activities to outsource to AI. Our virtues-based approach is also tied with empirical research to find out what we can outsource without any loss that matters to us.
I do think it can enable us to be more productive, which will enable us to spend our time on tasks that are important, either because they really require human judgment, or because performing that task helps us develop certain skills or certain habits that are important to our own flourishing. At least in my own work, I’ve found AI to be a very helpful complement or improvement over traditional search mechanisms for scholarly literature.
Any technology, even down to basic things like a hammer, both enhances human capabilities and also deadens us to certain things. With a hammer, I can hit that nail and exert a lot of force on it, but I’ve also lost the ability of touch from my fingertips, because my fingers are on the handle. And so there’s always a trade-off involved in using any kind of technology. It’s why you want to use a hammer for hammering nails, but you don’t want to use a hammer when you’re trying to do some really fine work.
In the past year, you’ve received a National Endowment for the Humanities (NEH) grant to study the humanities and artificial intelligence. We’ve talked about AI research broadly, could you tell me more about what you’re doing with the NEH grant?
We were fortunate to receive a three-year, five hundred thousand dollar grant from the NEH to establish a Humanities Research Center on AI here at Notre Dame.
The goal is to establish a collaborative research network on AI and the humanities because there’s a lot of related work happening at Notre Dame, but in different departments, in different colleges, and different schools. The goal for this center is to be able to bring those folks together and let them share their research, forming collaborative relationships. The center itself is a collaboration between three different institutes and centers on campus: the Riley Center for Science, Technology and Values, the Lucy Institute for Data and Society, and the Institute for Ethics and the Common Good. We’re the three core partners that put together the NEH grant.
Our center is called the Notre Dame Program on AI and the Development of Ethics in Agents. The acronym there is paideia, an ancient Greek term that would have been used for a holistic formation of citizens. So we’re thinking about the kinds of formation people need in this new technological environment. This includes the users of AI systems, but particularly those who are developing and implementing AI systems, and even the systems themselves. Not that they are equivalent to human beings, but they are, in a very minimal sense, agents in the world. They are, as we know, shaped by the ways in which we engage with them and by the kinds of training data that are provided to them.
Darius Colangelo is a junior majoring in Mathematics and the Program of Liberal Studies. He loves coffee, misses mountains, and when he’s not dozing off in the Grand Reading Room, he can be reached at dcolange@nd.edu.
Photo Credit: Program of Liberal Studies
Subscribe to the Irish Rover here.
Donate to the Irish Rover here.