Episode Transcript
[00:00:01] Margaux: Welcome to UVA Data Points. I'm your host, Margaux Jacks.
A self-driving car swerves. Someone is hurt, but who's to blame?
The company, the driver.
AI can't be put on trial, it can't go to jail and it can't feel remorse.
So when AI makes a lethal mistake, who answers for it?
That question drives Culpability. An Oprah Book Club selection by Bruce Hulsinger, professor of English at the University of Virginia. He's joined today by David Danks, professor of Philosophy and Data Science, also at the University of Virginia. Listen to a medieval literature expert and an AI ethicist discuss this controversial topic because long before the age of algorithms, scholars were already asking who is responsible for our choices and what we owe to the things we create.
So if you guys want to introduce yourselves…
[00:01:00] Bruce: Yeah. My name is Bruce Holsinger. I teach in the English department here at the University of Virginia. My academic specialty is Medieval Literature and culture, but I'm also a novelist. And the reason I'm here is that my most recent novel is called Culpability.
And it's a contemporary suspenseful family drama about moral responsibility in the age of AI.
[00:01:22] David: And I'm David Danks. I'm currently jointly appointed between Philosophy and the School of Data Science. And I work on a lot of different topics, but of most immediate relevance here, I work on how we can have better AI systems in every sense of the word. Better so ethically better, technically better, socially better.
So that can lead to exactly the kinds of topics that Bruce is really engaging in.
[00:01:44] Margaux: His novel awesome.
Love to kick it off with a question just around what interests you in the deeply human fear of losing autonomy.
[00:01:55] David: So I think that autonomy and agency, our ability to pursue what we take to be the good life, we all have our own understanding of what that exactly means, but we also all value the ability to go after that, to not have somebody thwarting our ability to pursue what matters to us. And I think when it comes to AI, one of the things that a lot of people are feeling is this loss of autonomy because it's starting to take jobs that we think of as important to us, tasks that we think of as important to us. So it's not just that the AI is going to drive my car. I kind of don't worry too much. That's not part of my identity to drive a car. But if an AI were a professor, I wouldn't just be out a paycheck, I'd be out part of who I am.
I Think that that's why AI has been this sort of almost existential threat for a lot of people is because it's starting to do the things that we think of as hours rather than the stuff you have to do to get to the stuff that matters.
[00:02:57] Bruce: Yeah. Culpability begins with this same kind of crisis of autonomy. A family of five is in a minivan on their way to a lacrosse tournament. And in the driver's seat, bell but not driving, is their 17 year old son Charlie. And they're in a semi autonomous car, a car that's in autonomous mode. And they have this horrible car accident. So the whole question, the whole moral question of the novel is around who was responsible for that accident. Because everyone in the car is in some ways engaged in some kind of threatening act, whether it's texting or, or distraction on the part of the responsible adult or designing AI systems in the case of the mom in the family. And so it becomes a very immediate kind of question that the family is dealing with in this moment of crisis.
Two very different lenses on the problem, the questions that you ask.
[00:03:53] David: But I think part of the way that they come together and this is one of the things I was sort of curious if you're okay, talk about it. I don't think this is a spoiler alert.
Is that in the book, as you were hinting at, the accident is the result of a very complex set of events. It isn't the sort of thing where we can say that person was responsible or this event is why it happened.
[00:04:16] Bruce: Right. Or this autonomous system is fully responsible.
[00:04:19] David: Exactly. And I think we sometimes have this illusion, especially in the United States of ourselves as independent, autonomous actors, we stand alone. And I think one of the things your book really forces people to grapple with is that that doesn't hold. Right. The world is a complicated, messy place.
But does that make you pessimistic about the very idea of responsibility? That's something that I sort of wondered as I read the book is do you come out of it sort of. There's almost a nihilistic possibility in the book that responsibility we can't even talk about because it's so complicated.
[00:04:51] Bruce: Well, that may be true and in fact it may help explain. The book didn't begin having anything to do with, with AI. It was a very analog car.
It wasn't in self-driving mode. The family had this awful accident and yet it was still about texting, about bad parental supervision. And so in some ways I wouldn't say the AI part of it is incidental. It just came into the book in a second or third revision and made me think about the moral dilemma in a harder way. I think it made it even more complex than it already was, which I guess is a nice lens onto AI in our society more broadly. Right. I mean, your work has tackled this problem from so many different directions about how a particular system kind of ramifies in many different moral directions.
So in a novel, you're allowed to kind of just hand it off to your readers and let them deal with the fallout as the family is in the novel.
But you're wrestling with.
I know you work in work advising government.
You're a scholar.
How do you advocate for those kinds of moral questions? And what are the difficulties you face?
[00:06:08] David: Well, one of the big difficulties is exactly the one that you're already pointing to and that you tackle in the book, which is the complexity of moral responsibility.
And then, as you said, layering on top of that, the fact that AI, at least in its current incarnation, doesn't seem to be the kind of thing that we can hold responsible.
You can't blame an AI. You can't yell at it. You can't make it feel bad. You can't put it in jail. And so all of the things that we normally think about in terms of punishment, in terms of responsibility, simply aren't available to us when we're talking about AI systems.
So in my own work, when I'm dealing with these, in the practical problem, a lot of it is trying to figure out how to get people to do better the next time, or maybe even do better the first time. So my own view is that the best way to solve the problem of AI responsibility, at least in terms of blame and negative outcomes, is let's minimize the number of bad things that happen with AI systems. If nothing bad happens, then we don't need to blame anybody for it. That's obviously an impossible goal, but I think that it can help us reframe the ways in which we approach the deployment of these systems. It leads to a somewhat more small c. Conservative approach, where you don't necessarily just throw the AI out there for everybody to use, but rather think carefully about when can it be used responsibly, when can it be used to support people rather than undermining them?
[00:07:37] Bruce: Can I ask that your observation that we can't put an AI in jail is a really interesting one, and I wonder if we could push on that a little bit. If we can't put the AI in jail, who can we put in jail? I'm Thinking about that, you'll know the name of the company. The hack a few months ago, was it anthropic or.
[00:07:58] David: No, it was OpenAI.
[00:07:59] Bruce: OpenAI.
[00:08:00] David: Their agents hacked a hugging face.
[00:08:01] Bruce: Hugging face, that's right. And so basically committed a crime.
And who was responsible for that crime? Have you delved into that particular case and what.
[00:08:12] David: So that particular case is a little tricky. And obviously I want to be thoughtful that I'm not. I'm not a lawyer, so be careful in that regard. But in a lot of ways, I think what that case revealed, first of all, was some of the shortcomings in OpenAI's testing protocols. They did not have the level of safety and security that one might have thought that they should have.
Another thing that that reveals is that the systems didn't actually do anything that violated their directives, their orders, as it were. What they did is they simply found novel ways of satisfying what the humans told them to do. The humans told them to solve a particular problem, and they found a way to do wasn't the way that presumably the OpenAI engineers intended for them to solve the problem.
But that falls back on us and the challenge of instructing AI about what we actually want it to do in terms of should somebody go to jail for this?
Leave aside that particular case. I do think that there are times that designers and developers are morally and probably legally responsible for what the AI does.
The challenge is not always. They're not always. And so this really, for me, as a philosopher, naturally leads to thinking about what's the very nature of punishment. I mean, we have different reasons why we punish. Sometimes we punish as retribution. We're mad at what somebody did, and we think that they need to be hurt by it because of it.
A lot of other times, though, we do punishment for rehabilitation or for protection of the community. And those things we absolutely can't can do with AI. We can rehabilitate an AI. We can protect the community from an AI by changing when it's allowed to be deployed, by changing aspects of its coding. So I think sometimes we talk about AI punishment as though it's this impossibility, because they don't have, or we don't think they have the kinds of emotional states that we have.
But punishment is a very rich and varied thing that we use for lots of purposes, many of which I do think we can apply to AI. So I think there's a little bit of hope for something.
[00:10:19] Bruce: Yeah, yeah. Well, you know, at the end of the novel, again, without giving any spoilers, there's this moment where the family engages in a process of restorative justice, maybe that'll be a model. Who knows?
[00:10:31] David: We can certainly hope.
[00:10:32] Margaux: Yeah, yeah, yeah. So diving in a little bit more on the blame.
So when we think about the book, who do you feel like was to blame?
[00:10:42] Bruce: Well, that's up for the reader to decide. I mean, for me, I kind of. And you personally, I don't want to give too much away in case people haven't read it.
For me, everybody bears some blame or some responsibility.
And the title of the book, Culpability.
As a medievalist, when I was arguing back and forth with an agent, an editor, a publicist about what the title of the book should be, one of the reasons that I wanted to stick with culpability is it goes back to. And David, you'll know this back to old moral questions and even theological questions. In the Christian Middle Ages, there was this notion of happy guilt, or felix culpa in the Latin. And the idea is that the original sin and the fall of Adam and Eve from the Garden of Eden was not necessarily a bad thing. It was a happy guilt that they shared. And why is that? Because in the Christian worldview, if they hadn't fallen, if original sin hadn't happened, Jesus Christ wouldn't have come into the world to redeem us.
It's a very narrowly Christian, pre modern view. But what was so interesting to me about that and always has been, is that culpability is shared by everyone.
All humans share that kind of moral culpability. And I suppose I wasn't really thinking about the medieval etymology of the term when I was writing the novel, but the more I think about it, I think it's a great old word that helps us think about distributed guilt, distributed responsibility, which I think is so much at issue with AI systems like autonomous vehicles, when you have ethics teams and engineers and CEOs and COOs all jumping in to design these systems. Right? Yeah.
[00:12:42] David: Right now, I think people still have this view of AI systems as being created by the proverbial tech genius in the garage. And that's just not the world we live in anymore. You have one company that collects the training data, one company that cleans the training data, one company that prepares and sells the training, one company that builds the learning algorithm, one company that builds the model. You'll literally have a dozen companies, and at every company there will be a dozen people who are putting their hands on the system at some point along the way. And it's very psychologically tempting, I think, for people to absolve themselves of guilt and responsibility when there's that many people involved in philosophy, we sometimes call this the problem of many hands. When many hands are touching a thing, you don't know who owns it anymore.
And so one of the things that in my work in AI policy and regulation that we've really started to be wrestling with is how do you set up structures that don't allow the responsibility to get spread so thinly that nobody's ultimately responsible? I mean, we've had massive corporate failures historically. Think something like Enron.
That actually was very hard to figure out who should be held responsible for the collapse of Enron 20 plus years ago, precisely because so many people played their little tiny role and everybody could excuse away their role or they lacked the knowledge to understand what their role really was going to do.
That I think we're running into the same problem with AI. It's almost a kind of arbitrage, right? How do I structure my AI development so nobody can be held responsible?
Hopefully not giving any ideas to companies about doing this. But I think it is a real challenge that we're all starting to wrestle with is how do we preserve some way of getting people to behave differently when it's this giant collective endeavor.
[00:14:42] Bruce: And maybe not just behave differently, but design differently too, I guess, which is part of behavior. One of the most hilarious conversations I had in the years since the book came out was at an AI Institute at Colby College. And I was talking to Stacy Doerr. I don't know if you know her work. She works on autonomous vehicle systems. And she said when she started getting involved in the research, she would all these people working out in California, and she said, okay, well, why don't you bring this car to Maine in the middle of January and we'll see how it does on our roads. And it wasn't just about the ice. It was about the lines on the road, the center line, the sidelines, whatever they're called on a road are invisible for four or five months of the year. They're under snow. And she said it was just she could tell that none of these people on this one team that she was speaking to had ever lived in New England.
And she said it was just really striking, that kind of very silly problem, an obvious problem. It's not that it hadn't occurred to people, it's just that it wasn't part of the way that they were integrating it into their practice.
[00:15:47] David: Well, and those kinds of blind spots, I think, are crucial when we're thinking about the design of AI systems. Another example is my personal first experience with self-driving cars was in 2016 when they were suddenly started to be tested on the roads of the neighborhood in Pittsburgh where I lived at the time. So suddenly I'm walking to campus and seeing cars with these funny sensor grids on top. And in that particular case it was because so many of the researchers had come out of the Robotics Institute at Carnegie Mellon where I worked at the time. So I knew many of them and I asked one of them. So why are you testing here in Pittsburgh? This seems really difficult. Lots of one way streets, lots of strange social norms. There's something called the Pittsburgh left, which is this very peculiar driving thing that people do, but everybody does it, so it's a stable norm, at least at the time. And they said, no, we want to test here precisely because it's so hard. And I think that that's something that in AI, when I think about my own development of AI systems, we're always wrestling with, we want to tackle real problems that make a difference in real people's lives.
But the real world's really messy and complicated. And so how do we have challenges and systems that are useful without trivializing it in a way that you're going to get failures of the kind of taking a Waymo car and dropping it into snow covered main roads.
[00:17:14] Bruce: Right, right. Yeah. Waymo was my introduction to autonomous vehicles. Now I was actually on a tour for my last book, which was about climate displacement in the United States. And I was in Santa Monica and I had been in the middle of drafting this novel. I want to say it was like late 23, early 24, something like that. And I got into Waymo for the first time and it really shook me.
And tourists were taking pictures from the sidewalk and they were still very new.
And just to give up, when you get in an Uber or a Lyft, you're already giving up your autonomy. You're putting yourself at the mercy of a driver who's very likely looking on their phone for their next fare. And so Waymos aren't doing that well. You tell us their first priority is to get you there safely. They're also looking for their next fare, but they're doing it in a way that would not interfere with the. Yeah, so I idealize Waymos right now. So I want you to tell me why I'm wrong.
And are those studies that show they're safer per passenger hour compared to Ubers and Lyfts, like, where are we right now at those? So some listeners I assume will get in waymos when they as part of their regular lives. I'm really curious what you think.
[00:18:32] David: Right. So we'll start with that last part, which is about the safety of them.
Right now it is the case that Waymo self-driving cars are safer per passenger mile than certainly ordinary human drivers.
Now that's a little bit of a, I don't want to say misleading, but a complicated statistic because they are deployed in places that have been well mapped, that are relatively well lit. Right. They're not deployed in Waterville, Maine or these kinds of things.
So there's biases there that can make it a little difficult to measure. But to the extent that we can measure this clearly, it does seem that they are safer in some ways. That's sort of not surprising. The reality is humans are not very good drivers.
We're not good at monitoring all the things going on around us. It's kind of a miracle we don't have much more accidents than already happen.
So in that sense, it's perhaps not surprising that these systems can handle the ordinary cases better than human. And the vast majority of cases are ordinary.
Where things start to get complicated is that we don't actually know what the broader systemic effects might be. So here's a worry that I've thought about and published on a few years ago. So there's something called a Jevons paradox. He was an economist in the 19th century in England who noticed this very peculiar thing, which was at the same time that there were massive advances in the efficiency of steam engines, the demand for coal increased, which seems sort of strange. If the steam engines are getting more efficient, we should need less coal.
So what he realized is that although they were becoming more efficient, that increase in efficiency actually increased the demand for for steam engines so much more that it undid the benefits of the efficiency gains in the steam engines.
If I can make something a quarter, if I can reduce its cost by a quarter, but I double the demand for it, it's not going to get cheaper.
And we see this today.
You mentioned Uber and Lyft. When Uber and Lyft move into a town, we see this example of a Jevons paradox. In general, Uber and Lyft drivers actually are safer than people who are not Uber and Lyft drivers. They just have more practice than most of us. So it's unsurprising that they tend to be safer per mile traveled. But when Uber and Lyft move into a town, you see a very large jump in the number of vehicle miles traveled. And so although they are safer per mile, the number of miles goes up so much that the total number of accidents increases. Because the total number of accidents just multiply together the rate by the number of miles and the rate goes down a little. The miles go up a lot. So the total number of accidents go up a lot. So the question we don't know is how much of a Jevons paradox are we going to see with self-driving cars?
When I lived in Pittsburgh, I lived walking distance to Carnegie Mellon, in part because I didn't enjoy walking, spending time in a car sitting behind the wheel.
But if somebody had told me, you can spend an hour and a half in the car and you can grade papers or you can prep your lecture notes, well, maybe I would have been willing to do that because then I could live out in the country and that could be nice in lots of ways.
So we have reason to expect that as self-driving cars become more widespread that people will simply drive more.
And so this raises all kinds of complicated questions where maybe we want to allow them once they're safer than humans, but that could result in actually having more traffic accidents, more fatalities, at least in the short run, even though they're safer.
[00:22:33] Margaux: Right?
[00:22:35] Bruce: Yeah, that makes all kinds of sense.
[00:22:37] Margaux: This kind of relates to where you guys have been going. But as we think about our reliance on AI systems and as it is part of our life and becomes kind of structurally unavoidable, how does that change? Who would be responsible when things happen?
[00:22:56] Bruce: Yeah, well, it's a tough one. I think. There's certain people in certain industries that don't have a choice. Like my son, my older son who does. He works in data analysis, data engineering and statistics and risk.
And he at the moment really doesn't have a choice. His company is. They're not mandating a certain number of hours on Claude, but they're asking for more efficiencies.
They're wanting their code guys to crunch more quickly and use agents to do more tasks more quickly. My younger son, who's in finance, he too I think is in an industry that is being completely reshaped by these systems. So it's very hard for them to push back. The students in my literature courses, on the other hand, do have choices. And I think institutions that are, I don't want to be impolitic, but that are cramming large language models down the throats of humanities students in ways that aren't necessarily good for their brains, for their writing, for their reading. I think it can be up to us as faculty to start pushing back a bit and creating spaces where students really do feel like they have agency. They have some say in how they're thinking, how they're reading, how they're writing, how they're collaborating, where they do have a choice. So I think that took your question in a different direction than you were intending. But I have strong feelings about that in the classroom.
[00:24:27] David: Well, I agree. And I do think that many people have more of an ability to opt out than they might ordinarily think they do.
This is certainly true of social media. There are people who need to be on social media. It is part of how their job works. And my niece, that's a big part of her job is to manage social media for a nonprofit organization.
So for those people, sure, but not for the vast majority of people. The vast majority of people actually could get off of social media if they wanted to or if they weren't addicted to. And this raises all kinds of questions about free will and choice. And I realize that. Yeah.
But I do think that that's. So the first thing to say is I do think there's a bit more of an opt out that people can do.
When I talk to public audiences, I often emphasize that AI is an amazing tool. I use AI probably every day. But we need to make sure we know why you're. Why we're using it. You should always be asking, what am I getting from this? What am I losing from using it in this way? So that's the first thing. At the same time, though, there are people who can't necessarily really easily opt out. And so I think this raises a lot of questions, to use a different C word, complicity.
So when do we become complicit in problematic behaviors, system structures?
You know, it's a question that we all have to wrestle with in our everyday lives. The world is filled with injustices. It's filled with things that are environmentally damaging.
To what extent are we complicit? To what extent are we culpable for. You know, I need in some sense for my job to do some international travel. Well, that imposes a significant environmental cost on the world to get on those international planes.
Am I responsible for it? Am I complicit in a system? Should I be trying to change the system?
And I think that that's a question that especially people who are in the tech industry, and I'll lump myself in here as somebody who helps develop these systems that we're really wrestling with. How do we build AI systems that provide people benefit without making them complicit in injustice without making them complicit in being forced into situations where they have to act contrary to their values and their deeply held belief.
And I think as an industry, quite frankly, we're doing a pretty bad job of that right now. I think we are not acting in ways that minimize people's complicity in some of the harms that are occurring, whether environmental or societal.
And I worry that as AI becomes more ubiquitous, we're going to be pulling more and more people into these really morally compromised positions where they're having to decide whether to opt out of some system entirely or to be playing some small part in the perpetuation of harms.
[00:27:25] Bruce: Just a few hours ago, there was a piece posted on the New York Times about a Russian drone that had killed three Ukrainian civilians, operated completely autonomously, and it made that decision. I don't know if you've seen this yet. I'm sure it was probably in the news a few days ago, but it made this decision without any human intervention.
I don't quite know the specifics of it, but that it was given a task at the beginning of its mission, I suppose, and it fulfilled that task, but it was not.
There was a long. There's probably some technical word for it. No human was in the loop within the hours before it attacked. And in culpability, I have this. It's brief, but a scenario. In a congressional testimony about this fleet of autonomous drones, a swarm that gets out of control, but does the mission, but then attacks a bus full of civilians in Yemen. And the headline was about the dystopian future of war, but we're very much in the middle of it. It's not the future, it's now. And I think that's sometimes the one big worry I have is when I hear philosophers or policymakers saying, we need these guidelines. We really need to lobby Congress and industry to do X, Y and Z. I worry about the timeline, the event horizon for these kinds of things. I wonder if that's a dilemma that you struggle with.
How can you even.
In some ways, it's analogous to writing a novel about something very contemporary. Right. It already feels outdated by the time it comes out. Like, I'm always thinking about writing at the edge of where we are, but sometimes I'm not going to anticipate something.
[00:29:20] David: Yeah. I mean, so we've had autonomous weapons making decisions about who to attack in conflict for at least three to four years. Arguably, we've had it even longer, but it's becoming easier and easier to do it. I think that the article you're mentioning, the big thing that's salient there is it was done without having almost no specialized military equipment. And that makes it particularly worrisome for a lot of us.
Warfare and conflict are challenging in lots of ways. There's a sort of banal observation.
The first government entity in the entire world to have principles for the ethical and responsible use of AI was the United States Department of Defense.
They had them before anybody else in any other government entity in the world in terms of binding principles.
Military people, by and large tend to worry a lot about ethics. They worry a lot about responsibility precisely because they understand the magnitude of what they're involved in.
And I think some very smart people look and say, well, AI systems can give us more precision, right? If you have to launch a missile from a long way away, things might change in the 30 minutes it can take a missile to get to a position.
And so wouldn't this be wonderful? But of course, cases like the case in Ukraine or what you write about, only slightly fictionalized in the book, are the counter argument to that. And so I think one of the things that is happening right now is quite a bit of research on how do we improve the ways of having or ensuring that AI systems carry out what the user wants them to do.
And in the research context, I call that not plagiarizing. In the military context, they call that following commander's intent and following international humanitarian law. And so I think there is an enormous amount of dual use technology being built right now, and we need to not shy away from the fact that they're being used this way, but rather hold ourselves to that much of a higher standard because we know these are not going to just live in an academic research lab. These are systems that are going to end up being deployed on battlefields. And I say that not to endorse it. I wrestle with questions of complicity. I do some work with the US Government and with parts of the Defense Department.
I don't do anything targeting, I don't do weapons.
But I do worry about how they can use AI more responsibly, how they can use it in ways that don't undermine their ability to ethically protect various national security interests.
And I lose sleep over it because we see how quickly it can go awry, how quickly these systems can be used to give people capabilities that are far exceeding their training or their experience.
And I wish that this were just a kind of one off. But as we're seeing, it's going to become Ever more common, I think, and not just for the great powers, but also for really every military out there.
[00:32:54] Margaux: I think moving into a similar topic, but closely knit is around some of the surveillance type cameras and things like that, which have intention of keeping people safe, but the balance of losing freedom in those types of things.
[00:33:10] David: One of the things that always worries me about a lot of the surveillance technologies is that the arguments for them often rely on anecdotes. They rely on very salient cases.
Often the laws literally get the name of some person who suffered a horrific crime.
And so then people will say, well, if only we had this technology, we could have stopped this crime. Well, I mean, if we installed surveillance cameras in every room of every home in America, we'd stop a lot of crimes. But I think that that would be a completely dystopian world to live in. I don't want to live in that world.
And so I think that oftentimes the decision to increase surveillance is made sometimes for financial reasons, but often on the basis of emotional anecdotes.
And we never pull back surveillance because of emotional anecdotes. So it's sort of this ratchet. All we ever do is increase the surveillance.
And each time we increase it, people object, and then it gets normalized, and then it gets increased, and then it gets normalized.
And I fear we're moving to a place where the people who control the levers of the surveillance mechanisms are going to have an enormous amount of power that is invisible to most people in this country, in the community, in the world.
So those are the kinds of worries that I have is that you lose accountability because these things just become part of the background.
And so you can't have these questions about who's responsible when it's misused, or who's responsible for the actions people never take because they're scared.
And I think it's easy for that to get swept out of the way when there's a tragic story.
But I think policy demands better of us.
You don't make policy on the basis of individual cases.
[00:35:03] Bruce: Well, what you said about how those surveillance mechanisms can become invisible, I think one of the roles of art is to make those mechanisms precisely visible. There's a wonderful artist named Trevor Paglen. He's a MacArthur genius winner. And I don't know if you know his work.
[00:35:22] David: Fantastic.
[00:35:22] Bruce: But he has done, over the last 15 years, he's done these incredible installation pieces drawing on technologies like some of the early work in VisionNet and exploring issues of algorithmic injustice, misidentification of people with brown skin by, by security cameras and so on.
Kind of using the products of machine learning, engaging with these technologies in order to make visible some of the inequities that they give rise to.
He had this spectacular exhibit at the Barbican in London a number of years ago.
There's other artists doing that kind of work and showing how the nature of humanity is in many ways being challenged through the ways we're being surveilled, the ways we're being interlaced with these technologies. So I do think art has a role here in exposing some of the issues with these systems and what they're doing to our lives.
[00:36:26] Margaux: Love to hear from you, Bruce, about what's next. Do you have the next great book on the works?
[00:36:34] Bruce: I don't know if I have the next great book on the works, but I am working on a new novel and it's about.
This one is also a contemporary story about two families and some issues that they get in around their online lives and pornography and so on. And it's dealing with our so called masculinity crisis and it's called the beta test.
So that's in progress right now.
[00:37:01] David: Well, I think we're going to have to get back together, not for that, but actually book that I'm working on right now looks at the nature of AI responsibility and diagnoses that part of the problem is that AI systems are designed agents. And our current moral theories and legal theories don't work for that. We have designed objects and we have human or animal agents and we don't know how to think about design agents.
But you know what? There's an intellectual tradition that thought long and hard about design responsibility of designed agents, medieval theologians. So people who worried about how could I be responsible when there's an omniscient, omnipotent God who has designed me to be who I am.
And yet I am nonetheless responsible because otherwise I can't be subject to sin. I can't have committed a sin if I couldn't have done otherwise.
And yet if God is omniscient and omnipotent, then I couldn't have done otherwise. And so how do we square that circle? Was an enormous focus of conversation of people like Aquinas and Anselm and Boethius or Boethius or Duns Scotus. It occupied centuries of really deep thinkers. And so what we're doing is trying to take some of those concepts and ideas and bring them to bear on questions of AI responsibility to see if we might be able to make sense of what it would mean to actually hold designed agents responsible just as they worried about it. So I'd love to get your thoughts once the book's finally done.
[00:38:38] Bruce: I love the. I always love it when the medievalists got there first.
That's music to my ears.
[00:38:45] Margaux: Awesome.
And then last question.
Where do you guys see AI in the future in terms of in your spaces and with your backgrounds?
What do you see in the future?
[00:38:58] David: So I'm terrible at predictions, so I try to avoid making them.
I actually think there's a reasonable chance that we're moving towards a world in which lots of boring uses of AI just become normalized everywhere and we don't even notice that we're using them anymore. Sort of like the Internet or search and the high profile things largely fizzle out and don't make a big difference.
So I don't know that we're going to have millions of self-driving cars on the road, but I think every car is going to have a pretty good self-driving mode that it can be used in 90% of the time. I don't think that AI is going to be writing every academic paper out there, but I think it's going to be a part of the research and writing of every academic paper. It's not going to be a solo author, but it's going to be playing a bigger and bigger role.
I actually think that we're much more likely to see a future where AI fades a little bit into the background, but thereby becomes even more important and indispensable.
[00:40:02] Bruce: And I think the future I see, or the future I hope for, and I'll answer the question obliquely, is that there's going to be a kind of new analog revolution in different parts of our world, maybe especially in K12 and university education, particularly in the humanities, where there's a kind of revulsion against large language models, particularly in the classroom and in reading spaces and writing spaces. And I think that there's going to be some really interesting developments spearheaded by people in those fields who really think about writing and reading. And that'll be all to the good. If this world develops that David is seeing where we still have these cherished analog spaces inside a world being reshaped by AI. I think that's something to hope for.
[00:40:56] Margaux: Great. Well, thank you so much.
[00:40:58] Bruce: Thank you.
[00:40:58] David: Okay, thank you so much. Great to talk to you, David. Wonderful conversation, Bruce.
[00:41:02] Margaux: Yes, thanks for listening to UVA data points.
More information can be found@datascience virginia.edu.
if you're enjoying UVA data points. Be sure to give us a rating and review wherever you like. Listen to podcasts we'll be back soon with another conversation about the world of data science.
