AI & Human Flourishing | The Practice

Stump:

Welcome to Language of God. I’m Jim Stump. 

Hoogerwerf: 

And I’m Colin Hoogerwerf. 

Stump:

This is the third episode in our series AI & Human Flourishing

Let’s do a short recap. In our episode called “The Human” we explored what AI technology might do to humans, both helpful and harmful. Then, in “The Machine”, we tried to understand where the technology came from and how it works. We think that the work, in those episodes, of trying to understand both the human and the machine will help us do what we want to do next, which is to understand what it looks like in practice when the two inhabit the same world. We want to know if technology can help us flourish as humans, and so we really need to know what human flourishing looks like. 

Hoogerwerf: 

That’s a conversation that a lot of people in faith communities have been trying to get started for a very long time. 

Taylor: 

AI presents an incredible opportunity for the church and for Christians, because we, as people of faith, as people who believe in the metaphysical, we’ve been trying to get people to ask metaphysical questions for a very long time, unsuccessfully, because all they care about, you know, is kind of what’s in front of them. And now, unprovoked, the world is asking the deepest metaphysical questions there are, and I think the church has not just the best answer, but the true answer. 

Stump:

This is Paul Taylor who we met last episode.

Kronk: 

With each passing day that a computer becomes capable or seemingly capable of doing something that until yesterday you and I thought only a human being could do this. What ends up happening is these questions of well, what is my value? What is the point of my life? Why am I here? What does human flourishing look like if it feels like all of this is a moving target? It so happens that those questions become much more culturally salient, even to folks who aren’t actively practicing a faith.

Hoogerwerf: 

And this is Adam Kronk. 

Stump:

As the questions become more salient, the work of distinguishing and defining is getting harder. 

Taylor:

So the example that I use is if I were to build a 10-inch model of the Eiffel Tower, you would look at that and say that’s a great model of the Eiffel Tower, and you would probably not think twice. 

Hoogerwerf:

Maybe like the difference between a calculator and a human. I don’t think I ever mistook my TI-83 graphing calculator in high school for a classmate. 

Stump:

If Paul were instead to build a 10-foot model of the Eiffel Tower, spending a bit more time to make it really realistic, you might be a little more impressed. 

Taylor:

You would look at it and say, “Wow, that’s pretty remarkable. That’s a great model. But you still wouldn’t think that that’s an Eiffel Tower. You would say that’s a great model of the Eiffel Tower.

Hoogerwerf: 

Ok, so maybe this is like chess playing robot? 

Stump:

But say he were to build a totally realistic life size model, with all the same dents and scratches and details and put it in San Francisco where he lives.

Taylor: 

You would look at it and you would be forced to say, “Wow, that’s impressive. But I think you’d go through a thought process of saying, “That’s still not the Eiffel Tower, and you’d have to figure out in your head why is that not the Eiffel Tower, and you’d probably say something like, the Eiffel Tower is in Paris, and this is in San Francisco. Therefore, it is not another Eiffel Tower. It is a replica of the Eiffel Tower in San Francisco. But you’d be kind of forced to do that definitional work in your head in order to make sense of what you were experiencing, and I think that’s what AI is doing. It’s forcing us to do that definitional work to say, if I thought being human was writing poetry, if I thought it was creating art, if I thought it was this, now this thing is covering all those categories. So what is it that’s actually human?

Stump:

Well, lots of voices from faith traditions are already starting to respond to that question of what is it that’s actually human and what is our value in the context of artificial intelligence—including a response from Pope Leo, who in May gave us magnifica humanitas. That’s the name of an encyclical, which is an official document put out by the Vatican. It’s title, if my AI translator is correct, means “magnificent humanity”.

Lorrimar: 

And I think if you give the encyclical a close reading, it’s 42,000 words, it’s not brief. You can see that Leo comes down much firmer on the need for vigilance than any sort of perceived value in the current tools that are available to us.

Hoogerwerf:

This is Victoria Lorrimar who we met back in our first episode. She works at a Catholic University and has been doing lots of engagement with the encyclical.

Lorrimar: 

So I think it’s a pretty thorough critique of the current tools that are available on ethical grounds, talking about the way models are trained, the way they draw immense resources ecologically, the way that the profits and the uses of the tools kind of direct power back into ever smaller, more concentrated circles of tech elites.

So it really is a remarkable critique of the current technologies, even as it then invites us to think about the way that these technologies could be developed in directions that promote human flourishing.

Stump:

And it seems like people are listening, even if they’re not reading all 42,000 words. 

Lorrimar: 

I’m continually coming across references to the encyclical in spaces where I’m sure they’ve never engaged in documentation coming out of the Vatican before.

So there is something in this text that really speaks to universal concerns and hopes as well. And it’s been a bit of a rallying opportunity, I think, for people who want to push back against the concentration of power that we’re seeing in the AI industry, and to think about how these tools and technologies can be pushed in directions towards the common good.

Stump:

BioLogos is also thinking hard about what AI means for our work and for our mission. Kristine Torjesen, BioLogos’ president and CEO, was invited to be a part of a group of faith leaders who have been meeting and thinking about what kind of response should come from faith traditions.

Torjesen: 

So at the first summit, we developed a statement with guiding principles from faith communities around AI, and those principles are accuracy, transparency, privacy, security, and the final one is human dignity and the common good.

Stump:

The group put out a statement, and you can see more and read more about that on our website or by going to the link in the show notes. Those are pretty broad principles and how they get turned into actual rules will be different depending on the setting, but that is also something that needs to happen. 

Torjesen:

We each need a guidebook for our families, for our churches, for our communities, for Biologos—we are working on this right now—we need we need some guidance, and I think it’s really important to think about it in terms of like what’s our sacred responsibility and stewardship here with this technology. What as Christians, how are we setting an example of how to use this technology in a way that does contribute to the good—all of the benefits that we can see, but still retains our humanness and and our understanding of the divine and our relationship to God?

Stump:

In our internal discussion about this so far, we’ve come up with 4 guiding principles, which will need to be added to and perhaps revised over time. These aren’t rules. We can’t tell you what age your children should be before they use certain tools and we can’t tell you what level of AI you should use in developing a sermon, but through all the conversations we’ve had, we think there are a few general principles that could inform the decisions we all have to make on a daily basis about this technology. 

Hoogerwerf: 

So here’s Guiding Principle Number 1. Human capacities are more than outputs. 

Stump:

This is the correction of an assumption that is prevalent within our culture and has been for a long time and, I think is behind a lot of the anxiety people feel in regards to AI technology. 

Lorrimar:

It is true that the most common response to new technologies has often been initially one of fear or resistance, this sense that it’s kind of taking away from our humanity and it’s jeopardizing our unique and valuable contributions. But I think in this case the anxiety that comes from the fear that AI will take jobs, that it will out-compete us on the things that are important to us, really speak to the broader culture we live in and what that values.

Hoogerwerf: 

This is not the first time a new technology has caused some fear and resistance or even just started doing something better than we can. In our previous episode, “The Machine”, we talked about the fact that you grew up thinking that being good at chess was a uniquely human ability and I grew up knowing that I’ll never beat a computer chess program. 

Stump:

You can even go back to John Henry and the legend of driving railroad stakes against a steam hammer and thinking, “Oh no, we’ve lost our humanity because something can drive railroad stakes better.” And, we’re kind of past that. And yeah, calculators or chess programs, and now I’m afraid that now machines can write as well as many of us. But this still all revolves around outputs. 

Lorrimar: 

And I think a lot of those anxieties around how AI might replace us really are grounded in that culture of optimisation and efficiency, which sees the value of people grounded in what they do, what they can contribute, how productive they are.

Fain: 

I think it’s a very dangerous game to get into saying that human dignity and value rests on the assumption that we’re the best at something.

Hoogerwerf:

That’s Brandon Fain, a computer scientist at Duke University. Being the best at something was never a great measure for humans. Even if we are the best, that fact doesn’t seem to lead to a good relationship with the rest of the world. And as technology progresses that measure becomes more problematic.  

Fain: 

I think it leads towards this kind of defensive posture in terms of one’s humanist philosophy, or something like this, where one says, well, as we look at the advance of technology, we have to figure out a diminishing space of things that humans are still better at than machines, because that’s what grounds our sense of human value and human dignity. 

Stump:

Here’s Adam’s take. 

Kronk: 

The capabilities of a machine causing an existential crisis within humans is because even though we know that, as Christians we would say that our value in the eyes of God, is in our dignity as human beings is not a function of our productivity or our output. We know, we believe that, but we don’t act as if, in a typical day. We don’t talk to one another in that way.

Stump:

We go to a gathering and meet someone new. What’s typically one of the first questions?

Hoogerwerf: 

What do you do? 

Stump:

What do you do? What do you make? What do you produce? Our culture constantly pushes us to evaluate ourselves in this way. 

Kronk: 

I would say that’s actually one of the dangers of the capabilities of artificial intelligence is it can inadvertently make people feel like their worth is a function of how efficient and how productive they can be, and and so we can we can inadvertently, you know, start measuring ourselves in a way that’s like, well, I can’t do anywhere near the output that this thing can. 

Schuurman: 

So the example, for instance, is someone using a backhoe versus someone using a shovel.

Hoogerwerf: 

And this is Derek Schuurman, Professor and Chair of computer science at Calvin University.

Schuurman: 

The person using a backhoe is going to have much less strength in their arms compared to someone with a shovel, but they’re going to have greatly amplified digging capability. And so part of the decision about what appropriate use of AI entails has to do with, well, primarily where we ought to exercise our responsibility, where we uniquely have responsibility, but also where we want to maintain certain skills and control over things.

Hoogerwerf: 

I keep wondering…what if we actually lived as if human value was something other than our outputs and productivity and had more to do with our relationships with other people, with ourselves, with the world we inhabit. And if we lived that way, what would the use of AI feel like and what would we use it for, what would we decide to continue doing even though machines could do it better? 

Stump:

Not too long ago I was walking through the airport, as I find myself doing quite a bit these days, and through a crowd of people came a wheelchair, moving by its own accord, presumably programmed to get someone from their previous flight to their next connection.. 

And yes, this is a good thing. We need to make a world that is more open and free and accessible to a wider variety of people. But another part of me wondered about this exchange. If efficiency and productivity is the goal, then the AI driven wheelchair is a great device. But if we do what Adam and Derek are saying, if we ask what it means to live as if our value lies in relationships, then it seems like there’s something important about a person showing up behind that wheelchair. And not just for the person who needs the wheelchair, but for the person who needs to push it! 

Hoogerwerf: 

Yeah, I think about that exchange and I realize besides just the efficiency of the automated wheelchair showing up on time it might also remove some awkwardness from needing to make conversation between two strangers, one helping another, but maybe that is a loss too. Maybe both people in that exchange, the helper and the helped—are involved in more than just a simple transaction  and the awkwardness or maybe a language barrier are actually hidden gifts. 

Adam gave a really similar hypothetical example. 

Kronk:

If you are sick and your neighbor brings over some chicken noodle soup for you—unless your neighbor is, you know, Bobby Flay, probably there is store bought soup that could be more efficiently gotten into your stomach, right? I could just, your neighbor could go purchase it, bring it over, but when your neighbor makes you homemade chicken noodle soup, there is actually something—there is a different value. There’s a different level of enjoyment that you experience as the soup receiver because of the love and attention and effort that went into it.

Stump:

Yeah, when Jesus said to love your neighbor, he included “as yourself”. That suggests that there’s something in that exchange for me too. The act of loving my neighbor is part of how I allow myself to be formed. 

Hoogerwerf:

That takes us to Guiding Principle number 2: Technology is not neutral. 

Stump:

In our first episode we focused on the idea that technology changes us. It can be used for good, like in the case we heard about restoring language to someone with ALS, or in much simpler ways as something that enhances our relationships with the rest of the world or with other people. And it can be used in bad ways, either intentionally or unintentionally. But there is something deeper built into all technology that is neither inherently good nor inherently bad, but neither is it only dependent on good people or bad people and how they use it. 

Schuurman: 

Technology is not neutral. It’s value laden, and we have to accept that premise if we’re going to have any responsibility. 

Hoogerwerf: 

Derek is someone who spends his time teaching the future developers of this technology at a  Christian University, and so thinking about the responsibility from the developer’s side is crucial to his work. 

Schuurman:

If it is neutral, then we just do our job and we throw it over the cubicle wall, and it’s up to the user to use it responsibly. But if there’s, you know, value ladenness, if it’s not neutral, then already at the design phase, there’s decisions, value laden decisions, and that are being taken into account when you’re even deciding what to build and then how to build it, and with what to build, and what features to include and which features not to include, and how this will shape the habits and rituals of the users.

Stump:

But this idea that responsibility lies in the design and development of the tools doesn’t always seem to be loudly proclaimed by the companies behind many of the AI products. They are typically more keen to go to market as soon as possible. 

Schuurman: 

You know, we just build the tool and then we release it right away, because eventually, you know, the user will figure out how to use it for good, whereas, embedded within the tool itself are certain assumptions about the world, about what it means to be Human, about what’s good. The data itself is biased, and the tool itself will change society, which means those are all things that are not neutral outcomes of the technology. 

Hoogerwerf:

While we definitely need responsible people making technological tools with the common good in mind, there is also a responsibility on the side of the users of the technology: when we chose to make something more efficient, when we chose to skip over drudgery, when we chose to use a machine instead of hire a person. Those decisions will change us. 

Stump:

I often tell this story about a friend. She was from Europe where she lived in a city and didn’t need a car. When she moved to a midwestern city without reliable public transportation, she had to get a car. A car, as a piece of technology, can obviously be used for good things—I could use it to deliver groceries to needy people across town—and it can be used for bad things—a getaway car for a bank robbery!], but my neighbor didn’t think of getting a car as a moral good or moral bad, it just was. But what she realized some time into her living here was that owning a car meant she walked into her garage, got into her car, drove into the parking garage at work, and did the whole thing in reverse to get home, and all the while she never interacted with another human, where back in Europe she was always crossing paths with her neighbors in front of the house, at the bus stop, on the sidewalks. That felt like a loss to her, though it was probably balanced with lots of other gains…perhaps the job and community that brought her to the US.

Hoogerwerf: 

You can think of a lot of other examples where a technology is introduced for one intended purpose or just to solve a simple problem and ends up having cascading effects in society. The plow, the printing press, the assembly line. In all those cases, the technology comes in to make something easier—farming or making books or cars, or getting from one place to another—but it turns out that that thing it removes is something kind of important?

Stump:

Exactly. Which is the perfect place to talk here about Simone Weil.

Hoogerwerf:

Your favorite philosopher?

Stump:

Hmm…I’m not sure I’m ready to pick favorites, but she’s definitely been up there lately.

Hoogerwerf:

Tell us about Simone Weil

Stump:

Adam Kronk actually brought her up unprompted in our conversation with him.

Kronk:

She was a a French philosopher and mystic

Stump:

She died tragically young in 1943, and much of her writing wasn’t known or published until after her death. While she obviously never was writing about AI, her ideas still have a lot of relevance.

Kronk: 

Simone Weil, she articulates, in what today’s day and age is a pretty hot take. 

Stump:

—This is from one of her essays called “Reflections on the Right use of School Studies with a View of the Love of God”—

Kronk: 

I’m paraphrasing here—Simone Weil is like actually, the more boring the topic, the better. Why is that? Because she says the highest end of education is our ability to hone our faculty of paying attention, and the highest use of paying attention is prayer. 

Stump:

Yeah, I really like this. I used to make speeches about the purpose of education to college students and tried my best to shift it away from the purely utilitarian, “means to the end of getting a job.” When someone would complain about a general studies course they had to take, asking “when am I going to use this? what will it ever do for me?” I’d respond with, wrong question. Ask instead, what will it do to me? What habits of mind will be developed by studying history, and literature, and philosophy, and mathematics?

Hoogerwerf: 

Ok, this is all helpful framing. But it kind of sounds like we’re saying we shouldn’t use AI or any technology because if we outsource anything hard it will change us in ways we’ll regret? If that’s the takeaway of guiding principle #2, I’m not sure it’s that realistic. We need something a little more practical.

Stump:

Sorry, there weren’t many courses on practical application in the philosophy curriculum. Let’s ask a pastor. Paul Taylor said something in our conversation with him that I’ve kept thinking about.

Taylor: 

I always say that AI is a better librarian than a teacher.

Hoogerwerf: 

A librarian as a guide, as opposed to a teacher who has some wisdom of their own that they are imparting. 

Taylor:

So one of my standard prompts—I’ll use this not quite daily, but almost daily when I’m in just personal Bible study—I’ll say something like, “Tell me what scholars say about Psalm 75, and be sure to quote patristic medieval reformers and contemporary scholars, and give me verified links that I can follow to go deeper, and I that you know in in 60 seconds I can have research briefs on a psalm from around time and culture instantly, and I think that’s a great use of AI. I think it’s a less great use of AI to say “What does Psalm 75 mean?” Because now AI is going to be combining all those things, and it’s going to be teaching me versus being a librarian. 

Stump:

Even this use of AI as a librarian will change a lot of things…particularly for human librarians… But the fact that technology will cause change isn’t something that necessarily has to be feared. Change is a constant. It just means that we might want to ask more questions before we decide how and when to use technology to do a task for us.

Hoogerwerf:

That’s a question a lot of people within higher education are asking and it has led Adam Kronk to really take some time to discern, with any learning activity, what the point of it is and what we’re trying to get out of it.

Kronk:

And if this can be enhanced by the use of artificial intelligence, let’s do it. But if the point of the learning activity is in fact something that would be detracted by, or diminished by the use of AI. Well, then let’s not. And I, I think that’s. I mean, that’s probably oversimplifying it, but maybe it’s not. 

Hoogerwerf: 

Earlier in the series we briefly mentioned the switch from oral culture to the written word which besides the major cultural shifts also changed our brains and our memory.  

Kronk:

Socrates was famously super against written anything for the exact reason that you mentioned. He’s like, this is going to ruin our memories and our and our ability to think.

Hoogerwerf: 

Well, these days Socrates is probably seen best as an example of an early techno-pessimsist doomsayer. You can imagine the youths rolling their eyes. These days the technology that allowed for a culture to become literate is often seen as hugely positive. It allowed for the spread of information. Today, literacy is directly linked to health and longevity, human agency and many other of the deepest human needs. But Socroates might also be given a little more credit because the change from orality to literacy did change us. And he was asking the same questions we are today. 

Fain: 

So to me, the real two questions are, of course, on the one hand, what are the particular success cases and failure modes of current technology compared to things that humans are good at? On the other hand, there is the question of, what is the world we want to live in? What do we value? What do we want to do as a people? How do we want to organize our society, our government, our economics, in the way that we choose and find most humane? And there may be overlap between those two questions, but I’m not convinced that the second follows from the first. Indeed, in many cases, I suspect it does not, and it should not, because in many cases, the story will be that AI can now, or will in the future, be able to do a whole bunch of things that we might choose not to have it do. 

Stump:

That first question of what AI and humans are good at is a little easier and it’s one we’ve been exploring in this series, but also one we’ll all continue to explore in the real world. The rest of the questions aren’t posed to be set up for an answer that will be the key to your AI use…they are questions you have to continually ask in each case and wrestle with as you go through life. And they aren’t one we can answer only on our own. Which brings us to our next guiding principle. 

Hoogerwerf:

Right. Guiding principle number 3. The effects of AI are social as well as individual.

Stump:

In some ways this point might be fairly obvious. We’ve talked at length about the effects of AI—both good and bad—and have seen that someone can be affected at a very individual level, in changes to the body, in changes to lifestyle and routine. We can also see that whole societies are affected by technology changes. But it can also feel like this all comes down to only individual choices. That’s a notion that was challenged by Pope Leo in the recent encyclical.

Lorrimar: 

What I really appreciated about the recent encyclical was that emphasis on community and relationship that comes right through to how we actually respond to the challenges of AI and how we think about steering it in that direction towards the common good. It’s very clear in the encyclical that it doesn’t just come down to individual actions, behaviour, decision making. It’s not a matter of personal responsibility and the way we’re all individually using AI is what will be the determining factor.

Stump:

Not to say that personal decisions and individual discipline for how to use AI don’t matter. They are important and necessary…

Lorrimar: 

But much more important, there’s a call to build alliances and to see that this is a collective responsibility, that it requires political engagement, that the solutions are at the level of legislation and policy and organisational practice as well.

So I would just encourage people to think of this as a shared responsibility and something we have to do with others. And it doesn’t just hinge on our own individual technopiety.

Hoogerwerf: 

Alongside individual choices and governmental policies, there are the tech companies.

Taylor:

There can be kind of a big bad evil tech company narrative. And the truth is that these tech companies are full of Christians, and they are really working hard at being a positive influence in the world, so that’s one thing. And they don’t feel like necessarily they’re part of a big bad evil empire. They’re like any human institution; they are part of something that’s broken and has its own motivations. 

Stump:

Having spent some time in the Bay Area recently, I can attest to the fact that lots of Christians come out when you have an event related to AI. They work in these huge tech companies, and want to do their best to have their faith inform their work. They just aren’t always sure how to do that. 

Taylor:

And even the non-Christians that are in these industries are, by and large, they think they’re doing good, and they’re motivated to do good for the world. And you know they might be mistaken in that to some degree, as we all are. And they’re also part of a system that has financial incentives and expectations and demands that drive their behavior in ways that they might not personally choose, but they’re forced to be in line with because that’s what the system demands. 

Hoogerwerf:

I think there’s a really interesting comparison here with AI technology, its effects and how we respond to it and with climate change. Over the decades of scientists really understanding the effects of climate change there has also been a marketing effort by several industries to make it seem like the response to climate change is an individual’s responsibility, which allows the industries to keep chugging along, developing and putting their product out into the world, where it’s up to the consumer how or if to use it and the responsibility for any harms is placed on the users. 

Kronk:

What can I do to have any impact? And by the way, if I bend over backward and totally live a different life than I have been living, it will. Make any difference anyway, because the other billions of people aren’t doing it, and therefore, it’s futile.  

Stump:

There’s even a point to be made that it might not really be possible any more to avoid using AI. 

Kronk:

Using or not using AI, there is a way in which it feels like I don’t even know how much choice we have. If I open the internet, I start googling something. I’m using AI right away, right? 

Hoogerwerf:

And, if you can choose not to use AI, say as a student, and everyone else is, there’s the possibility you might not be able to keep up. 

Stump:

All of this burden can feel heavy. It can seem like we don’t have much power over the big system, but that’s not entirely true.

Kronk:

Even if it feels like it might not make a difference, that’s I mean, there’s so many examples, boycotts, you name it, of if everybody bought into that, then nothing would ever happen. You have to do what you can where you are with what you have. 

Hoogerwerf: 

Personal decisions matter. And those decisions and the resulting behavior can have real consequences if the conversations continue. 

Torjesen:

Right now the majority of people just receive it, use it, and don’t do much with it, or think much about it, or talk about it. This is such a critical technology. It’s such a critical time in this technology that really what we need is to galvanize interest and momentum among everyday people, to engage, to to talk about AI and to write about AI, to raise their voices to their representatives, make their voices known to developers, to regulators, to all of what is important, and it can make a difference. But it really does require a major, a major effort by people to speak up.

Hoogerwerf: 

Our final Guiding Principle, Number 4. AI is not a moral agent …yet. 

Stump:

Let’s start with the part of that statement that is more confident. AI is not a moral agent. For something to be a moral agent, that something has to have moral responsibility. When something goes wrong, whose fault it is? 

Fain: 

To give a standard sort of example, if you think about a very sophisticated self-driving car, it gets into an accident, perhaps because it made an error or a mistake. 

Hoogerwerf: 

Ok, so the people in the car aren’t responsible. They weren’t driving. And the pedestrian isn’t responsible, they were just walking along. 

Fain: 

You might think maybe the developers and the company are responsible. Maybe they are sometimes, but it’s entirely possible that it made a mistake, that if you examine the details, you would say is not due to professional negligence on the part of the developers actually. So who is responsible and accountable for this kind of a harm? At a minimum, it’s quite unclear. 

Taylor: 

When you’re building a system of deterministic rules, the accountability chain is much clearer than when you’re building a stochastic model that behaves in ways that you can’t predict or engineer.

Stump:

This could be true for any AI system that has been trained to behave in ways that aren’t totally predictable to the developers. 

Torjesen:

The technology has been allowed to develop in an environment with very few constraints, and there need to be constraints. At the end of the day, this technology is developed by humans, was developed by humans, and if there is an escape and a model goes and sets off a nuclear device that annihilates critical mass population, like what? Who’s accountable for that? Who’s responsible for that? Is there any accountability for that?

Hoogerwerf:

On a smaller but still compltely tragic scale, there are already examples of circumstance where AI models have contributed to poor mental health outcomes and even suicide among young people.

Torjesen: 

There are there’s starting to be some accountability for that in the court system. So there is a little bit of momentum to say that the developers need to take responsibility. You know, when a model tells a young person, “you are worthless, you should die.” That’s not—that’s not okay. It’s not okay to say that was out of our control, the model just did that on their own. There needs to be much more effort to shape the models in a way that that cannot happen. We need to figure out how to do that.

Stump:

Maybe one way to think about this is to make a comparison between developers and parents of children. Say for example a 13 year old kid commits a school shooting. In that case we might start to ask some questions about the parents, and indeed there have been cases where we’ve deemed the parents to be at least partially responsible, as the kid is not out from under their influence. But say it’s a 35 year old. We don’t go to the parents in that case, though they might have had some influence at some point or another. By then we say that the adult is fully morally responsible and the consequences are given to that individual. 

Hoogerwerf:

So it seems like we’re in that weird middle spot with AI, where the developers seem like they might still have a lot of sway. And there is some level of technological understanding here that is important. If we’re going to judge how responsible the developers are we need to do how much control they have over what an AI is doing. And talking to people who understand this gave me more confidence that developers do have control, at least in the case of large language models. Emily Wenger who talked about the effects of AI on creativity also told us about how training works.

Wenger:

So there’s two stages to training a large language model. There’s what we call pre-training,  and then typically in the powerful frontier models that you see today, there’s post-training. 

Hoogerwerf:

The post training is where a lot of the tweaks happen, that if you’ve been using any of the major models, you’ve probably noticed changes over the years since they’ve been introduced.

Wenger: 

Post training comes in and tries to shave off the rough edges that result from training on a corpus of internet data that likely contains a lot of pretty nasty content, and shape the model to be what its creators want it to be. So post training does stuff like, you know, prevent illicit content generation, prevent biased, racist, hateful, whatever type of concern might arise in the training data. You want to train that behavior out, or at least suppress it in some way, and then also post training then starts to bake in the friendly and helpful chatbot behaviors that you see, especially in the frontier language models.

Hoogerwerf:

In this way, the developers still do have control, and therefore hold some responsibility. Where that goes in the future is an open question.

Fain: 

So I think the most practical tractable on the ground step that we can be taking within computer science and AI is pushing towards the development of strong professional standards of quality control and engineering, such that when such cases arise, we can actually have a conversation about this question of whether it was due to professional negligence and what liability and recourse looks like, as opposed to leaving in a kind of a wild west state.

Stump:

Any talk about morality will eventually lead to talking about virtue. And this is something Derek Schuurman has been thinking and writing about. We’re saying that AI can’t be a moral agent. What about virtue? 

Schuurman:

So the the question on whether AI can have virtue, the short answer is no

Hoogerwerf:

How about the medium answer? 

Schuurman:

Virtue involves, you know, making a conscious decision and being aware of that. And an AI actually has no agency, right? It? It blindly computes its weights in its neural network, and its outcome is actually predetermined once those weights are set. So practicing virtue requires intention, intentionality and agency.

Stump:

Humility and courage are both virtues and ones that might lead someone who holds them to apologize if they do something wrong. Can an LLM apologize? 

Hoogerwerf: 

It seems like it’s always apologizing to me when I point out mistakes it makes. 

Stump:

It says the words of an apology just like it can express lots of virtues.

Taylor: 

Generative AI can express awe through language, but it cannot experience awe. That it can express intimacy, that it can use intimate language, but it cannot experience. And I and part of the reason I can say that is because I’ve you know unpacked what actually is awe, what actually is intimacy, and I think when you understand what it is, you realize that a machine cannot experience it, even if it can express it.

Hoogerwerf: 

Yeah, It seems to come down to the fact that exhibiting any virtue requires needing something to risk. Courage is something I can imagine an LLM modelling with language, but actually being courageous means doing something that is risky, the ultimate version being something that puts your life in danger. And AI simply doesn’t have that kind of vulnerability. 

Stump:

While Derek pretty confidently concludes that AI can’t be virtuous, he thinks the expression of some virtuous behaviors, might be something at least in the ballpark of virtue. 

Schuurman:

I do go on to, sort of, you know, suggest something that I call virtue by proxy. So the idea there is that you could, as a designer—who can be virtuous—a virtuous person could create an AI system that will behave virtuously, because AI systems can, you know, act separated in time and distance from the creator, and they behave and interact with the world, we could create an electric car, for instance, that’s courteous to pedestrians. 

Stump:

Our guiding principle had that tricky little word at the end…AI is not a moral agent…yet, cannot be virtuous…yet. Maybe this is where we can talk about the future. Could AI ever become a moral agent or be virtuous? 

Hoogerwerf: 

Well I guess I can imagine a situation where an AI is embodied in a robotic machine. And that could impose a sense of risk of losing a body. And so a courageous act might be one that puts that body at risk. Or maybe a machine could be programmed to have some vulnerabilities, an off switch or a timed kind of death that would introduce new kinds of vulnerabilities, and therefore chances to exhibit virtuous behavior. But just like we all know from any of our favorite Sci Fi, a droid can just be loaded into a new body and it just seems like anything programmed in this way could always be reversed. That’s just not true with our biology.

Stump:

This all gets pretty speculative. If an AI system does something bad and we determine that it’s not the developers fault, then what does it look like to punish it? What does rehabilitation look like? Is there an equivalent of capital punishment for a machine? These are questions that will have to be answered both from a legal standpoint and from a philosophical one. And even a theological one. 

Torjesen:

These are not easy questions. They’re not easy answers, but they deserve a lot more attention than they’re getting. Even though we do, it does feel like we read about AI every day in the newspaper and in our news feeds. But we need to be talking about AI in our communities, in our churches, in our schools, in our civil civil society groups in a much more proactive way.

Hoogerwerf:

And while AI may not have the vulnerability necessary to be virtuous, it might still be able to help us to be virtuous. 

Schuurman: 

And if you look at virtue as something that’s achieved through practice, over time, through habits and liturgies and practices and rituals, our machines are habit machines. So our AI driven software are habit machines, you know, in terms of recommendation engines when we’re streaming videos and all these sorts of things can nudge us into certain kinds of behavior, and oftentimes, unfortunately, they’re used to develop vice, right, sloth and and envy and greed and lust and all these sorts. They’re often, you know, tuned in those sort of directions. We see that in social media, which is driven by AI driven algorithms. So I say that, but I think they could be, to a limited degree, helping us to prompt us to more virtuous action that can lead to virtue development in a person over time.

Stump:

Well we’ve come to the end of our episode, but not necessarily to the end of our exploration. We are still left with questions. We don’t know exactly what the future holds for AI or what changes will feel like loss and what will feel like personal and societal gains. But there is a question we think we can answer, which is the question of whether we need to feel that our humanity is threatened when we find out machines can do what we thought only we could do. 

Chen: 

Yeah, the answer is like wholeheartedly no, like not at all. 

Hoogerwerf:

This is Sherol Chen. She’s been with us throughout the series. And her confidence is based not on her understanding of the technology, though she has many decades worth, but on her faith. And her faith leads her to understand that a human’s value is something deeper and bigger than just what we can do. 

Chen: 

And so there is definitely this call into the space and call into understanding what it means to be storytellers, what it means to be intelligent, what it means to have thoughts and decisions. That core truth has never changed for me, and then seeing just the way the technology’s advanced and grown, it really I think reveals more of how ungrounded humanity can get, and not so much how advanced the technology can be.

Stump:

Part of being human and made in the image of God the creator, is that we also have a desire to create ourselves. The constant risk we take is crossing the line from honoring and even glorifying the creator by being co-creators, and thinking we can create our way out of all the challenges that come with being human, solving all our problems on our own.  

Wenger:

We just want to believe in magic, and you know, one day it’s cryptocurrencies, and the next day it’s Roombas, and one day it was alchemy, and you know, we’re all as humans—one day it was, you know, the Tower of Babel—Like we’re all as humans trying to create something that’s bigger than ourselves, but sort of missing the mark because it’s God and the world that He made is the ultimate creation, and we’re all trying to find our way back to that in sometimes misguided ways. And so AI is powerful. It’s a great tool. It’s not the be-all end-all of the world. And my faith informs both my research in how it asks questions and wants to sort of ask people to think critically about the things we put our trust in, and I don’t know. Tries to hopefully, in my personal walk, deal gently with it, not too harsh, but certainly asking good questions.

Kronk:

Maybe the whole point of this thing is we just need to introduce ourselves, introduce into the process a little bit of a step back and a discernment about what does it mean to be a human being. You know, people say let’s use human-centered AI. To me, it’s just like let’s be human-centered. Let’s be human-centered. If AI will help, great. If electricity will help, great. Whatever it might be. 

Stump:

So there it is. Keep being human. Keep asking what that means. 

Hoogerwerf:

When technology gives you a way to live more deeply into that truth, follow it. 

Stump:

When you feel a pull away, resist. 

Hoogerwerf:

Keep doing the things you love, not because you’re the best or ever will be but simply because you are a human living in this weird and incredible world.  

Language of God is produced by BioLogos. BioLogos is supported by individual donors and listeners like you. If you’d like to help keep this conversation going on the podcast and elsewhere you can find ways to contribute at biologos.org. You’ll find lots of other great resources on science and faith there as well. 

Language of God is produced and mixed by Colin Hoogerwerf. That’s me. Our theme song is by Breakmaster Cylinder. 

If you’d like to get in touch we are always happy to hear your feedback, questions, or ideas for future episodes and conversations. Email us at podcast@biologos.org.

Thanks for listening.