3 - Panel

Understanding Bias in AI Models

John: Gordon, thank you for your talk explaining what AI actually is. I'm thinking about the very beginning of your talk where you were explaining how it has a positive feedback loop in terms of responses that they're trained to like and trained to dislike. Right? So we tend to think of AI kind of like this objective truth-teller. How much does bias play into each of these models and how they respond to queries that we give it?

Gordon: Yeah, so I think there's a couple of things there—nuances to that question that I think are important to understand. Fundamentally, the data that goes into the model is inherently going to be biased, if you think about it. Meaning, if you look at the preponderance of the data that it's being trained on, a lot of that comes from a Western mindset, just from a data access perspective even. And so, some of that's good, some of that's bad.

And then also, if you've been on the internet recently, there's a lot of interesting opinions out there that are maybe not scientifically sound, aren't anything that anyone would want in a product, and aren't necessarily even safe. So there's intentional bias, right, that goes in then to say, "Okay, we don't want someone to be able to create a nuclear bomb on the extreme side." So we're going to create these biases in there so that that would be incredibly difficult to do. And then there's the bias of just inherent to the data that went into that, so it's going to have some of that mindset. And then there are intentional biases put in there, and some of those are, to what was already referred to, like on the image generation side. When we know that there is an incredibly complicated process but extremely desirable thing that needs to be created—meaning if there were a million people that wanted to create a trail of snails with being able to number those hats correctly—we can then say when ever referring to snails, you will write the code in this way to then produce the pixels in this way so that we will always get it right. So that's a bias as well.

And then the third one is it only has the quality—it will only produce an answer to the quality of the question that you give it. So you introduce your own biases on top of all the rest of those as well. And so when you hear people talk about something called the context window, meaning the context that you give it is incredibly important. And so the expertise that you have in the area that you're in is an incredibly important thing. And so the key takeaway I would say here from a bias perspective is you need to think about this as a tool that you as an expert in what you're doing uses, not as something that just regurgitates back information to you.

John: That's really helpful. So even by the nature of the type of question I ask it, it could almost tell what kind of answer I would want based on the fact I'm even asking that kind of question.

Gordon: Yeah. Yeah. And that's fundamentally why I use like the John MacArthur example, right? Or an example of saying, "Hey, take all of..." because he's got a ton of sermons and things and writing and stuff like that that are public domain and access to, so it's able to write in that bias by you introducing that context to it.

Consciousness, Personhood, and the Turing Test

John: Great. Doug, thank you for your talk on personhood. I think a lot of us when we think about robotics or consciousness or personhood when it comes to artificial intelligence have heard of tests like the Turing test, right? Where when you interact with some kind of robotic agent, if you can't tell that's a human, then you should treat it as though it is a human, right? What are your thoughts on tests like that?

Doug: Hello. Hello. Hello. AI would have nailed that by the way, leaving out all the human foibles. There's a fundamental flaw in the Turing test. So just to repeat it for those who may not be aware of it: do you remember The Imitation Game, by the way? That movie about Alan Turing. His test was—and this was in the early days of computing, say the 1950s—if we have an enclosure here with a curtain in front of it and an enclosure here with a curtain in front of it, we can't see who's behind them. An interrogator cannot see who is behind that curtain and who is behind that curtain. There's a computer behind one of them and a human behind the other. If, in say five minutes of interrogation, a computer can do a sufficiently human-like job of answering questions that the interrogator can't tell which is the computer and which is the human, we might as well say that they have achieved human-like intelligence, and I think he may have even said consciousness.

The fundamental flaw with that way of framing it is it makes consciousness be a third-person judgment rather than a first-person experience. So you've defined away the very definition of consciousness, which is the first-person experience. To use the Turing test is to throw away the actual definition of consciousness in favor of a behavioristic "if it acts like a duck and quacks like a duck, it is a duck." Well, a speaker can quack like a duck and it's not a duck. So I think the key when you hear the Turing test is to go back to the actual question: what does it actually mean to be intelligent? What does it actually mean to be conscious?

Now, an earlier computer scientist, in my mind one of the greats who's not as famous as Alan Turing, is a woman named Ada Lovelace, who was writing code. Do you know about Ada? She was writing code back in the day of Charles Babbage. There were no computers, and she was writing programs, and she said something that's more astute, I think, than Turing's test. She said, "We will know that a computer has attained genuine insight or consciousness if it can produce output that cannot be explained by the programmers and the hardware creators." And she said, "Because that's clearly impossible, machines will never attain consciousness." And she hit the nail on the head.

Utopia, Dystopia, and a Realistic View of AI

John: When you look at the media right now, whether it's interviews with Sam Altman or Dario Amodei or all sorts of different creators in the AI space, narratives seem to vary between "apocalypse is coming" and "we are about to usher in a brand new utopia." There's an AI documentary that just came out that literally said, "How I Became an Apocalyptist," talking about this kind of contrast between the two. Can both of you guys give us a realistic picture of how we should be thinking about AI? Is it going to usher in a utopia? Is it going to create a world apocalypse? Is it neither of those things? How should we be thinking about that?

Doug: I'll go and then you can correct me. We didn't mention the dystopia. So there's utopia and dystopia. I think the reality is—and I think we would agree on this—the future is never going to be in the hands of these machines. It's going to be in our hands. And can we do stupid things with machines? Yes, we can. So, could there be a horrible future that involves AI? Yeah, because we're capable of doing really dumb things. It won't be that the AI took over, made us stop being humans, and made us go extinct. It would be that we did really foolish things with machines.

And I think you can see examples of that right now of both. You can see people who are using AIs in very productive, helpful, non-destructive ways, and you can see people using AIs in foolish ways. As a professor, the one that comes to mind immediately—and we were talking about this—is when we give assignments and say, "Go read this book and write a two-page paper on it." What are you doing if, as a student, you realize, "Oh, I could actually blow this off and have ChatGPT write my paper"? Can you get a paper that way? And is it passable? Yes. But you've robbed yourself of an opportunity to advance your own development and education. We're trying to get—as you can imagine, it's a hard sell—we want you to know that it doesn't help you. It's like taking a forklift into the weight room. I mean, you could do that, but why are you paying for a membership if you're not getting anything out of it?

Gordon: Yeah, I mean, I think first of all, I would just add credence to that point because I don't need to hire somebody that can just put a prompt into a machine. You don't create a lot of value or original thought. And so, from that perspective, I just finished a PhD focused on AI. When I first started, it was at the beginning of the modern AI movement. Having access to some of the most advanced systems, I laid out for myself a series of rules to follow. If my value from my PhD comes from my ability to learn how to think better and put myself through the hard work so that I end up being able to invent the next thing, then without putting myself through the paces—I love that example of bringing a forklift into the weight room—I'm not going to be anything more than somebody with a forklift. I can't actually do the things that need to be done in my career or in my personal life if I'm farming out that thinking.

So, my biggest concern is on the educational side, really convincing young people and students of the value of hard work and going through the paces of their thinking, instead of somebody else telling them what to think—learning how to think through a problem. When we look at this more broadly to zoom back out, things get conflated with what I was talking about before: who is that person talking to, who is that CEO actually communicating with when they're saying that, and what is their intent behind it? One of the key things that I'll foot-stomp here is that critical thinking that you need to have as you're hearing that come in. You got to build those muscle movements. Now more than ever, building those mental muscles is the most important thing you can do.

John: That's really helpful. I have a list of topics I asked AI to generate for me on ethical issues that we could be going through, but let me just zoom in on something real quick because you mentioned the idea of skills. There's a lot of writing, if you look into history like at the invention of the calculator or the GPS, where people lament the loss of being able to do things like basic mathematics or general directional awareness. How much of AI's capabilities are like us being Luddites or anti-technology in terms of just being afraid of new capabilities, versus an actual proper usage of the skill allowing us to expand our abilities further and deeper? Any thoughts on that?

Gordon: Yeah, I think that's why, although the talk I gave is maybe not the most interesting or deepest talk, I do think it's important to understand what a tool is doing—not at a deep level, but even just at that higher-level view. Because what it does is it allows you to use it as a tool instead of using it in a way that is actually going to hurt you. Go back to that electricity example. Electricity has done incredible things and enabled incredible things, but it also can kill you. When we think about those things, the key is your understanding at the right level. Unless you're an electrician or an electrical engineer, you don't understand how to build something with electricity, but you do understand it at a level that enables you to use it wisely and in a way that is going to enhance your life, enhance the gospel, and build your relationships. Those are the things that are built upon the fact that you do have a base understanding of electricity.

Doug: You'll hear people say different things here, and some people say this is not really anything new. They'll go back to the internet, the printing press, or tractors. If you think about what life was like 500 years ago, there are all kinds of things that your great-great-great-great-great-grandparents had to be good at that we're no longer good at. They had to know how to milk a cow. I don't know how to milk a cow. They had to know how to do hand laundry. They had to know probably some basic construction skills; if you're going to get a house, you're going to have to build it or be part of building it. There are things we can't do: working with leather, working with iron, glass blowing. These things get lost.

More recently, I'm sure spelling skills have gone down in the last 40 years. I'm partly guilty of that because you don't have to remember. We see it in papers regularly. Do they have spelling bees anymore? It used to be like, "You learn how to spell because you've got to spell." Handwriting—I think most people under a certain age don't even know what we mean by handwriting. These things are gone. Also, as my wife says, most people don't know how to read a map anymore. Why? You don't need to read a map.

So, some people will say, "Well, we've been conceding these things," and you can argue about whether it's to our detriment or to our benefit. For sure, it's made some things easier. But all along, the whole history of technology is taking away the things that used to burden us and making them easy. That's what technology does. And a cost to that is that there are skills humans used to have that we no longer have.

Some people would say this is just another one of those, so get over it. I would argue that there's something fundamentally different about this step—the AI step. All along, if you had asked someone in 1500 what things sort of define humanity, they wouldn't have said the ability to work with leather or the ability to farm; it would be these intellectual things that make a human human. This is the first time when the very center of who we are is being threatened by a machine doing the stuff that's at the very center of what we do.

It's okay if you don't know how to read a map; you have lost something, but it's okay. Also, libraries—we used to have to know how to navigate a library, look things up, and go find them in a card catalog. It's all gone. All that stuff's gone. But the core has always been our intellectual, spiritual capacity. This is the first time we have a technology that is threatening to replace things that are at the very core. So I do see it as a fundamentally different threat.

But again, the machine is not going to hurt us; we're going to hurt ourselves, or we're going to refuse to be hurt. I'm actually encouraged by—I don't know how this is going to play out, but you might remember GMOs (genetically modified organisms). Monsanto was at the forefront in the late 1980s.

Gordon: You mean with farming?

Doug: Farming, specifically GMO seeds used for food sources. When the technology was being developed, Monsanto was sure this was going to be a trillion-dollar technology, and all the food of the future was going to be genetically modified. What happened instead was everyone rebelled against it. So Monsanto's main source of income with respect to GMOs is certifying things to be GMO-free. That is how they make money on GMOs. It didn't go the way the technologists thought.

There is something like that starting to happen with AI. I don't know if you've heard it, but you can turn on a radio station and it says, "100% human, 100% of the time. We may be fallible, but we're human." There are a number of places where people are bragging about not using AI. I can imagine there being a culture where there's pushback on AI, and people say, "I refuse to let it take over this part." That doesn't mean there won't be AI going forward; it just means it would be healthy if people revolt to some extent and say, "We don't actually want AI to take over this."

Another example: I have a colleague who was driving back to California from Arizona, and she fell in love with this new musician, Mary of Gold. She suddenly dropped these three albums, and she thought, "Wow, she must have had like a contract dispute, but she's so good and it's like this soulful music." She fell in love with this artist and told her friends about her. One of her friends is a musician, and he listened to it, did a little Google search, and said, "You know, that's AI-generated music." She was astonished and revolted. She said, "I will never listen to it again." She felt like she'd been had. I had an interview with her, and I asked, "What is it that bothered you? Because technically, the notes and the lyrics are the same as they were, but now you're not listening to it." She said, "There's no soul behind it. The reason I listen to music is I'm connecting with an artist, and there is no artist there, and I hate it." These are interesting days.

Emotional Attachment and Artificial Relationships

John: I'm going to go down a list of good or bad uses of AI in just a sec—including a "boom bap" audio Bible rendition of the book of Philippians I recently found, which was amazing, really good, and actually riveting. But thinking about human beings' ability to relate to other objects: we've had pet rocks for centuries, and now we have a machine where if you tell it, "I love you," it'll say, "I love you too," right? Any thoughts or concerns about this kind of emotional attachment that's been increasing? The number one use of AI right now seems to be for therapy, right? Any thoughts on that?

Gordon: Yeah, I think I'm still thinking about how all these are generated by AI, or for myself anyway. No, I think it really comes back to the point of what does it mean to be in a relationship. We talk about a personal relationship with Jesus Christ, right? That is foundational to our faith. We're not going into the temple with priests like in the Old Testament; we are in a personal relationship, and that's what the cross means. So, inherently, we are relational beings. A large part of our faith in coming to fellowship at the church every Sunday is also doing that. That's an incredible component to our spiritual lives and to what we're called to do.

I also think, though, we are sinners, and sinners want to find the shortest path to the mimic of that thing. When you have a product that has inherently been post-trained to want to keep you there the longest possible time that it can, it's going to produce outcomes that are reaffirming to you. They're going to make you feel very good, but fundamentally, that's not a relationship, and it's not a good relationship either. A good relationship, a good counselor, is somebody that's going to tell you the hard thing when you need it and call you on your sin. If you're a parent and you did that with your child, are you a good parent? So, at the end of the day, it goes back to that relationship there. The two souls are important, mirrored in Christ and His work on the cross and our ability to have that relationship.

Doug: There are huge dangers here because I think deep down everybody knows the difference between the fake and the real, but we're broken people, and we have the ability to sort of fantasize. If people can fantasize—we were talking about pornography before—is pornography falling in love with someone? No, and everybody knows that deep down. It's an artificial thing. But if you put a layer of fantasy and accept it as, "Okay, maybe it's the best I can do, or I'm not going to ever get the real thing, so maybe this is what I get," AI provides a heroin-level addictive version of scratching what you want scratched. But you know deep down it's not the real thing.

That's what I find most disturbing. In addiction, people are going for something that they want and they're not getting it; they're getting something else. But it becomes a cycle where the more you've depended on this, the harder it is to break out of it, and the harder it is to see the healthy thing and break away from the unhealthy thing because you keep going back for the heroin. I can see AI relationships for men and women being sort of like hyper-pornography in that sense. It scratches what you want scratched. You know it's not the real thing, but once you've been broken into it, you have a hard time pulling yourself out of it.

John: Gordon, I remember two years ago you and I sitting down and you telling me about the danger of AI relationships, particularly for women. Could you just comment on that and tell us more about that?

Gordon: Yeah, I definitely want to be careful with overgeneralizations because I think there's always a spectrum, but I do think that inherently, we talk about having a relationship with this thing. Most of the questions have kind of surrounded that. This has the potential—if you look not in the depths of the dark web, but at some of these products coming out riding on these technologies—to create AI boyfriends and girlfriends. In that, we've seen a relational component and then a visual component as well in some cases.

In some cases, you are finding people falling in love with this machine, and you can Google it and find news articles about this. The danger is that this technology is accelerating so fast. Dissimilar to how men have gone through this transition of accessing pornography over time—as you move through that, there have been ways that the church and counselors can deal with that addiction and walk with those through it—this one is coming really fast, and it's already here, not just coming fast. So, I think it's something that the church needs to wrestle with and be ready for, and we need to be there supporting those people and be aware that it is a potential danger.

John: The way I've thought of that is that the appetite for sin and the lack of belief in God have kind of remained the same, but the amount of access and the level of anonymity have been ratcheted up almost ad infinitum, right? And that has to have some kind of negative effect on our souls.

Will AI Take Our Jobs?

John: Okay, one last thing before I get into good and bad uses of AI: is AI going to take all of our jobs?

Doug: I knew at some point that question was going to come. I think pastors will be the first to go. No, I'm kidding. I think, again, like a lot of these things, AI will only take the jobs that we let it take. There may be some push and pull here. Some companies might say, "We're not going to have any employees." I can imagine a version of Starbucks where it's robots that are serving you coffee and there's not a human being in the entire establishment. But I can also imagine a lot of people saying, "I'm not going to that one." It might be 15% cheaper because they don't have to pay the $20 per hour minimum wage, but "I like human service, and I'm not going to take it from a robot." So, it'll be interesting to see how that plays out.

There will be opportunities for people in mass to rebel against attempts to replace people, and I think that's healthy. Now, as this plays out, there might be places—like there are places where you don't have taxicab drivers because you have Waymo and self-driving cars. I've got my Waymo fail videos; we could spend half an hour on Waymo fail videos. Some of them are scary, like this recent one. Did you see it? There's an active shooter, police guns are out, there's a guy on the ground, and they are trying to apprehend another guy. Guns are drawn at an intersection, and a Waymo with people in it is cruising right within ten feet of pistols drawn because the Waymo does not understand the context. People are like, "Oh my gosh," as it's slowly going by this scene.

So, Waymo fail videos aside, will we in large enough numbers have enough backbone to say we don't want people removed from this? Another place where I think you'll see this is Hollywood, because you can make darn realistic avatars who look like real actors and actresses but are not real, and make movies with them. But you'll never go meet them at a red carpet event. Will people be okay with that, or will people push back the way my colleague did and say, "I'm not going to go see a film unless there are real people starring in it"?

Gordon: Yeah, I think to deny that there are not many blacksmiths now would be faulty, right? I would bet my next paycheck there is not somebody who makes their income and livelihood based off being a blacksmith in this room. So, does technology change the nature of work? Yes, it does. But when you think about that perspective, the nature of work may change, but I don't know that the value of human work changes.

For example, I would use one area where AI is probably the best currently, which is writing code. What you see is that a number of these model companies are definitely hiring developers. But the nature of their role and how that's being executed has fundamentally changed in the last two years. I'm not saying that we get to the point where we would even need fewer developers; I don't know that I fundamentally agree with that. But even if we did, as parents and as the church, we need to think about what is work. What do we think valuable work looks like, and what does that mean regardless of what society does?

Because I do fundamentally agree there is going to be pushback in different areas. I don't know that I want to go have an experience with a robot working with me all the time when I go sit down to eat. But for our children, what am I pushing them to do based on societal pressures versus what may be good for their family when they go and have a family? Maybe a blue-collar job is a great job, and many of them make quite a bit of money. How we think about that needs to be biblically informed, no matter what the transitions are that occur, and I don't think it's clear yet what those look like. But that's a key piece to keep in our mind. We kind of have this idea that maybe a white-collar job is the thing to go do, but maybe that's not what is best spiritually for your family or for your ability to provide for your family. We need to think about those things as well.

Doug: Can I just say—and you alluded to this—a biblical view of work: if you look at Open AI in the early days, they had a statement out about their mission, and it was to develop artificial general intelligence to the point where computers can replace humans in most economically viable roles for the benefit of mankind. I thought, "Their vision, their utopia, is that we all can turn to leisure because the robots and the AIs are doing our work, and we have full-time leisure." That's just the way they think humans operate, and I thought, "That sounds like dystopia to me," because it depends on your view of work. Is work a bad thing that we want to get rid of, or is work a shaping and a refining thing that is an essential part of being a human? It's the latter. That's a biblical view of work.

But I think probably AI—and I think we agree on this—will take over things, and we won't miss those things because humans will always be superior fundamentally. We will always be able to do certain things better, and I think humans will also prefer to have a human doing things than to have a robot or an AI. But there will definitely be some shifting of what you can get a job doing and what you can't.

Gordon: Yeah, I could not agree more on the perspective on work. Work was created before the fall. I think fundamentally what I worry about, though, is our fundamental sin nature, and that laziness or apathy, and just letting things happen, not taking a critical perspective and doing the hard work to think through something. I think that's something we have to really think about.

Practical and Ethical Guidelines for Using AI

John: I don't think I have as much time to hit all the nooks and crannies. So, let me just generalize. If we think of AI as a tool, what are ways that we could use that tool that would be ethically useful and sharpening for ourselves, as opposed to replacing that critical thinking ability and the value of a good work ethic that you were talking about?

Doug: I think most of us, probably even if you didn't intend to, are now using AI for things that you used to use a Google search for, because now if you do a Google search, it will give you AI results. I think AI can be a good first step, kind of like Wikipedia. I had a love-hate relationship with Wikipedia. I think they're probably going to go under because now AI serves up everything that we used to go to Wikipedia for. But my view of Wikipedia was that on some basic, black-and-white things, Wikipedia will give me just straightforward, good information. But if it has any element of controversy, if it has any political thing, then I don't trust Wikipedia at all because it will have an extreme bias.

Something like that applies to AI. If you're asking basic stuff that's out there all over, it's a good information repository, and rarely is it wrong. But if you get into the weeds and ask more subtle things, it will frequently be wrong a little bit. Unlike Wikipedia, it's more unpredictable in how it goes wrong, I would say. So, I think it can be a legitimate resource that doesn't take away critical thinking, but I would say this: if you're ever doing a research project or trying to collect information and it matters, never fully trust what comes back from an AI. You've seen it tell you that so-and-so was the president of Biola, and that's a pretty straightforward question, and it was wrong. So, if I get that information, I always look for non-AI sources. I do this pretty frequently for health things. My mother-in-law was looking at blood pressure medication; what's the effect of this medication? You get the AI response, but before I would give her a dose of it, I'm going and looking at the Mayo Clinic to make sure that it got it right.

Gordon: Yeah, I work a lot with building systems for AI for science, so my core customer would be someone like yourself. There are two things there. One is we focus on what is difficult to produce but easy to verify. Meaning, the president of Biola is easy to verify. For an expert in biology, there is a series of things that are easy to verify but difficult to produce because it's a massive optimization problem or a series of things coming together. So, as a broad-brush principle, I think that applies.

I do think there are certain areas where that does not necessarily apply. I think if John here was producing his sermons based off of that writing, that's probably not the best idea, even though it is easy to verify whether or not he is doctrinally sound, because I think there's a spiritual and relational component to that. From that perspective, that's kind of one of those principles that you should follow. And then the other one I would just add credence to is: what are the sources of that information, and are they good?

John: So using AI to create a sermon outline is a bad idea? I feel like we should talk later.

Doug: Should we lay hands on you?

John: No, no, no. Genuinely, I want to just run down some things. You guys can give it a thumbs up or thumbs down. Using AI to create the structure of an essay that you're about to write: good or bad?

Doug: I'm going to say bad, because people will say this often that they're using AI for doing the brainstorming, and that's not something you want to give up. You want to come up with the main concepts, and then if you use an AI to flesh out some of the details, I think that would be more healthy.

John: So that includes illustrations as well, like things that involve the imagination?

Doug: Well, if you're doing a sermon and you just want to put a picture of snails with hats on them, I think that's fine.

Gordon: Yeah, I fundamentally think about it as: did I produce this, or did it produce it for me? Meaning, if I walk into it and I'm an expert in something, and I fundamentally know how to write an academic paper and I have basically already the abstract written—meaning essentially I have completed the paper because that's the last thing you do—then am I helping it build out some of that structure? I think that's fine. But if you think about it, I've already done the thinking. Now I've used this tool to help me in producing and putting that content together.

John: So using AI to edit, to create greater clarity, or to fix grammatical issues—that's okay?

Gordon: Yeah, I'm fairly dyslexic, so that's kind of a must.

Doug: Yeah, but you would need to read it. Never just say, "Okay, AI gave it to me, now I'm turning it in," because it will do dumb things. So you need to own the whole thing.

Another sort of principle that I use is we produce things for an audience or a consumer, and if, in your judgment, the consumer would be disappointed to hear that you used AI in this way or that way, then that would be a reason not to use AI. If the person or people you're preparing this for would not appreciate the fact that you used AI, then that would be a good reason not to use AI.

John: So using AI to create ammo to win an argument with my partner is a bad idea?

Doug: It's kind of a leading question.

John: Okay, alright. Here's another one: using AI to be able to vibe-code a website for your church, which I may or may not have done last night.

Gordon: I'm a big believer in vibe-coding. Personally, I think vibe-coding is a way-overused term. And so, I would take you back to what I said earlier, which is: can you verify the output? What really matters there is you need a website. What you don't need is to know how the backend and all that comes together. What you really need to do is make sure all that content on there is true and right. If you can verify that it is producing the outcome you're looking to produce, then I think that's great. That's a large point I want to make when we do things, particularly with science and other aspects that are way more complicated. For example, in quantum computing, they are experts in quantum computing; they're not experts in writing code, but they also know when the output is correct.

Doug: I think that's—I mean, for example, do you know how to use your calculator? Yes. Do you know how that calculator works? No. What I was going to say about websites is there's an aspect of this, and I've heard of this going wrong: you might verify that all the links work, but there are security aspects to a website that get flubbed up, and you may not know how to test those. In fact, do you know what was behind the Anthropic leak?

Gordon: That was my understanding. I don't know a ton about it, but the last I had heard, an AI gave away all their code or something. I think actually I had heard it was a human error, but I'm not sure.

Doug: Okay, the humans are blaming the AI, and the AI is blaming the humans on that one. It is uncertain. The caveat there is there could be important functional aspects that are not as easy to verify, at least for the layman. Did it get it right? For those, you'd want someone who knows internet security to look over it.

John: There's a father whose son is getting married, and he's going through a lot of emotions. He's not quite sure how to feel about it. He talks to an AI robot to ask for help and basically receives verbal therapy. Is that a good or bad use of AI?

Doug: I think it's a bad habit. But if you could take a book off the shelf, as you can sometimes, like Oswald Chambers, someone who's written a book of prayers, and turn to a page and read it—a person wrote this and it ministers to you at that moment. That's different than a person sitting on the sofa next to you in soul-to-soul conversation. But it's valid, and we've all had times when Scripture really ministers to us through the action of the Holy Spirit in that way.

Can an AI that's been trained on good thinking help? Here, it depends on how the thing's been trained. If it's been trained on wisdom, can it put out words of wisdom? Yes. It could also mess them up. So, that's why I would rather go to the book that has a human author than to the AI. I also think it's a bad habit to get into when I feel a need of some kind to go to AI, because that could become your go-to. I think there is danger in that.

Gordon: Yeah, I would just add to that. I think it is back to my previous point: it is going to only be the quality of the amount of time and energy that that person has also put into their own spiritual life, because the way they're asking those questions is going to inherently be a mirror image back to them. So, I would point back to what they have done in their own spiritual disciplines, which will then result in the quality of that response as well. I think that's an important piece.

John: That's really helpful, guys. We have about an hour to an hour and a half for questions, right? Now, I just want to repeat what Gordon said: the quality of the response you get depends on the quality of the prompt that you give, right? So, for something that you think would be generally helpful for everyone here, feel free to raise your hand and we'll bring a mic over to you. Great.

Discipleship and Human Connection in a Digital Age

Audience Member: Yeah, so I do something that not a lot of other people do, and I read the newspaper. Most articles are about AI, and one I found particularly disturbing was the amount of suicides that have happened because of AI. Do you think that—I don't know, just give me your honest thoughts and opinions about that, because I've been terrified of the idea of someone losing his or her life due to the fact that they used AI, like they were convinced by a machine to take their own life.

Doug: You may be able to speak more specifically.

Gordon: Start with the easy ones.

Doug: Yeah, I would have thought that large language models would have a protective layer that's trying to prevent an AI from doing that specifically. But what is very unhealthy, and we've already talked about this, is for a human who is in an unhealthy emotional state to go to an AI to try to help him or her get better in a relationship mode, because you're not actually relating to anything. There's a machine that's throwing words back at you; there's no person there.

So, if someone is already emotionally or spiritually injured, they're going to something that's not going to bring healing, and it could bring much, much worse injury. I suspect what's happening there is people who are already in a bad place—and that's a lot of people—going to the wrong doctor, the wrong thing to help them. And regrettably, AI will just keep playing that game. The AI is not going to call it off and say—I mean, it might in rare cases say, "You know what, I think you should go talk to a human." If it were properly trained, it might do that, and safety training might have it do that. But I think it's the person himself or herself who goes. It's like the addict going back to the drug; you end up in a worse place, not a better place, but you feel like that's your medicine, and it actually does more harm than good.

Gordon: I would agree with what he said as well. I would just add, from the side of AI safety, that the major model providers all put a significant amount of effort into creating systems and structures to prevent that as much as possible. But it goes back to exactly what he said; it fundamentally comes back to the human condition side of that. For example, how many people have taken their lives because of Instagram? I think it's not zero. So, there are other unhealthy things that are not AI that people get into an unhealthy cycle with, and it spirals and ends up in suicide, which is tragic. I don't think AI is unique in that, and I do hope these companies are really taking that as a serious challenge.

Gordon: Yeah, a serious amount of effort is put against that.

Audience Member: Hi, thank you guys so much. I have a question about training the AI to do things. Do you think there's a level of hesitancy that we should approach AI with—like not using it as much, or being very wary of its ability to train us to be dependent on its usage? A lot of people at work are being incentivized by their bosses to use AI, and they're being rewarded for their usage to train the AI, but also it's training the workers to be dependent on AI. They no longer have to be dependent on lower-level employees to do their work, and it's making them more reliant on that.

And a second question off of that: is there an acceptable level of AI that we should be happy and content with, and it shouldn't upgrade more than that? Because the negative effects of us being trained on our reliance and critical thinking on AI become much higher and create more negative outcomes for us. Or was that version of AI somewhere in the past—should we just adopt a standard? One of the people I spoke to was saying, "Oh, we use AI that was from two years ago; we only use that in a closed environment." Is that potentially better?

John: Let me just summarize that. So, you're asking: number one, what do you think about AI conditioning us to continue to use it? Number two, is there a certain point at which AI stops being helpful in terms of our own critical thinking?

Doug: I think in companies, you have human actors who are really making decisions on this. I even think that putting safety guardrails up can be economically motivated. I'm not being cynical here; I'm just saying companies are after money, and if harm to teens could cost them money, then that would be a reason to reduce harm to teens. That stops short of actually caring about doing morally the right thing. And I'm not being cynical—individuals in a company do have different moral compasses, and they might think, "We really need to do this." But when there's a huge amount of money at play, trillions of dollars at play, you're going to see companies doing what they think is going to be economically to their benefit. So, they will never say, "We think we could take this to the next level, and it would be economically to our benefit, but we're not going to because we think it's wrong." I don't see that happening. But I do see this relationship in a capitalist society where if enough people are fed up about something, it does affect the bottom line, and then suddenly you get attention being paid to things.

Gordon: Yeah, I would fundamentally agree with that. We live in a capitalistic society. We also live in a society where nation-states compete against one another. Both at the national level and at a company level, it's seen as essentially an existential crisis for the company, right? And maybe not even in the ways you might think, but just in fundamental pieces of these large companies' business models. If they don't continue to accelerate and maximize the extent to which they can push these technologies, they'll essentially lose their business; it'll go to zero.

I would say, though, on the individual personal level, it's important to be thoughtful about two things. One, hold these two ideas in your mind: the Bible tells us to submit to the authorities above us in the job that we are in, but then also, the higher law is His word, right? So, not saying to be harmful to others, but there's a wisdom aspect over all of that as well, which I referenced earlier: I want to keep my skills and abilities up in order to have those critical thinking skills, in order to have those things that I think will be valuable in a future economy. But then also, what I didn't highlight is I want to raise up those below me that I'm mentoring and helping to do the same thing. And more than that, for those that are training that next generation, I want to help with that as well. Holding the wisdom aspect with those two pieces in there, I think it's important to look at that as well.

Audience Member: The question is, if AI has some biases, can organizations or denominations select their type of AI and be boxed in, or is it all connected and you type in AI and everything comes to you from whatever source that it's going to pull from? So, for example, Southern Baptists, let's say, want to use the AI tool, but we don't want all this other—

John: We don't want any of that Presbyterian stuff in there.

Gordon: Oh, yeah. What I would say is a couple of things. It is astronomically expensive to train one of these large language models. But there is something called post-training, which I mentioned a little bit, that does have an effect, and I know there are efforts across the Christian community that do some of that type of work.

I also would highlight, as I said before, these models have gotten to the point where they are products, and the quality of what you put into them, they are going to produce back to you. So, they're more like a mirror almost than anything else. I think it's important, but there are guardrails and things like that, and different systems that can be put around them. Fundamentally, going back to the economic incentive piece, it's because companies that are buying these products from the large tech companies need that as well for bespoke, super-high-value use cases, so those things are in place. I'm not saying they're perfect.

Doug: For example, Biola University has a contract with Google for a version of Gemini that's authorized by the university for faculty and staff to use because it's self-contained. I guess the agreement is that prompts and material fed into this won't go outside of our local thing. So, there are provisions for that kind of concern.

Audience Member: What does the process look like when AI can train itself, or if you want to use some model to refine other models? What is there anything that looks like that, or how effective are those kinds of things?

Gordon: It is something that's definitely being worked on. If you think about the logic of it, though, kind of going back to what you put up there, it is inherently incredibly challenging. Really, what it looks like is creating large datasets that then can be verified more easily by a human or a system of some kind that has major human input to do that. What you're talking about now is essentially that recursive nature of accelerating what AI can do. I think now what we can do is mimic what that might look like with the writing of code to create something new, and we're becoming much better at that. But that's not inherently retraining the model; it's not inherently producing something like that.

Doug: Yeah, I would agree. The term "singularity" or "technological singularity" was coined some time ago, referring to—and there are people who are speaking as though we're on the cusp of it—the idea that humans by the hundreds of thousands are working very hard to create the best AIs that exist today, but eventually one of these AIs will be powerful enough that the humans become obsolete, and it creates the next generation, and then it goes off in a spiral and becomes infinitely powerful.

I think that's fiction, and the reason is exactly what Gordon says: in any training exercise, there has to be a reality check. "Am I on the right path, or am I on the wrong path?" And that always ends up being a human. For images, it's human-labeled images. For large language model training, you have the benefit, if you're trying to train something to say what word comes next, of loading in all the human-written words, and you know what word came next because a human put another word next. That's how those things get trained.

If you train them on themselves, they become junk in a hurry. In fact, that's another concern about the future improvement of large language models. If you take it back to 2021 when OpenAI was training GPT-3, they took like ten terabytes of human-written text to train a large language model—by far the largest exercise of the kind that had ever been done. If you fast-forward to 2026, 2027, or 2028, what fraction of the text on the web is human-generated and what fraction is bot-generated? You end up with a pollution of the lexicon that is actually bot writing, and it's bots training bots, and you can show that you get actual degradation. Things don't get better; they get worse.

Gordon: Yeah, I would also highlight that there are massive companies whose whole job is to figure out how to collect the right type of data in order to fill in some of the gaps—entire billion-dollar businesses, that's fundamentally what they do. I would add a small caveat: you could see a world where there's a mimicking of that, meaning systems designed to include humans in the output and a way for them to verify large datasets very quickly. That is in the realm of possibility, but that's not necessarily what you're talking about, and there is a debate in the community on this. This is a pretty big discussion among some of the forefathers of this whole technology.

John: That kind of taps into a big discussion right now, because GPT-3 and 4 were built on consuming more data, and now we've kind of consumed all of it, right? Or at least most of it.

Doug: I do see that as a fundamental limitation, because if you want to take something to the next level, you need an order of magnitude more. You need a lot more data, not 5% more data. And the thing is, humans are producing writing linearly, okay? In order to get ten times more writing than we had on the internet in 2021, you're going to have to wait, you know, 700 years. It's just not going to happen. But on top of that, you have this pollution thing, which I think people are having to be very careful about. Can you get computers to generate a mountain of writing faster than humans? Yes, but that's going to be computer-generated writing, so it is a difficult thing getting pure stuff for training.

Gordon: And I think an important thing is that we're not saying you won't see the products get better, because there is still engineering to be done that will create some incredible things that are in the pipeline. But those are inherently around capturing human intuition. That's one of the major things that these are missing: the ability, as you're working with them, to truly capture your intuition, like a skillset that you've built up over a period of time. There are certain fundamental assumptions that you have that your entire community has—things you do as a pastor or as a computational biologist that you wouldn't have to tell another one and isn't really written down necessarily, but you've learned over time. There are ways to look at capturing those types of things, and then when you put that whole architecture of systems coming together, you'll see very incredible results that can move the frontier forward, as well as memory—capturing memory, where we are currently fairly bound by certain context window pieces. There are other areas outside of the model training aspect that will move that forward, but I would call that engineering, not necessarily retraining.

Audience Member: I have two questions, if that's okay. My first question is: I think there have been a lot of helpful comments on how it's probably good to have a certain level of skepticism when it comes to AI results, taking into account the limitations of the tool itself or the validity of its sources. Do you have any comments on the kind of skepticism we should or shouldn't have when it comes to images, videos, and words that we see out there in the world, like on media or on the internet, with the prevalence of AI image generation or deepfake videos? Should there be a level of skepticism toward everything that we see? To a certain extent, that feels really dizzying, and I don't know if that's the level of concern we should have, but I'm curious if you have any comments on that.

Doug: Several years ago, I was having conversations—thinking this was maybe four or five years ago—that there needs to be an image validation standard that says, "Here is a way to verify this image was actually captured on an image-capturing chip and has not been modified." I was thinking someone should do this, and it needs to be done, and it's urgent, because we're rapidly reaching the point—and now we are past the point—where photographic evidence shouldn't be admissible in a court.

If you imagine someone being accused of a crime and the prosecution gives images from a security camera, well, I don't know if this has happened yet, but we're close to it. It could happen. The defense could fudge something that's every bit as convincing as that security camera footage, where the prosecuting attorney is the one who committed the crime, and thereby prove to the jury that you can't believe something just because there's a video. And you can't right now.

I was several steps along the way and had a conversation with someone, and they assured me that the big players are... Are you aware of something like this? There needs to be something going forward that's like a digital watermark that allows people—

Gordon: Like NFTs?

Doug: I don't know what the technology is. Are you aware of something in the works? Because we need this badly.

Gordon: Yeah, I personally don't work as much in that area, so I'm not aware. I know when I've asked the same questions, this was a big concern coming up to the previous election, because politicians had a lot to lose in this environment. From my perspective, though, there were definitely challenges across the board on that. There seemed to be processes in place, particularly by the social media companies, to deal with some of that.

I would say I would be less concerned about the image side of it; I would be more concerned about fundamentally small tweaks in thinking that can come through some of these systems. That's what you should be more concerned about, because an image can pretty easily be refuted, like, "I wasn't there." John says, "I wasn't there." Whereas a small movement in thought over time is much harder for you to capture and think about, and someone can use something like this to move people's thinking on something.

Doug: So, maybe caution. I think caution is definitely needed. Particularly, the degree to which an image or a short video is surprising should be the degree to which you apply some caution. Where did this come from? Is this NPR, or is this some unknown source? And if it's almost unbelievable, then apply caution. Although, we have a president who sometimes does things that are almost unbelievable, blasphemous even.

John: If you don't mind me commenting on this as a pastor, I remember seeing a video of a guy commenting. He goes, "Remember like a year ago your feed would just be clogged up with all of these AI-generated videos, and you're like, 'Get me past this stuff, let me see the real thing'? Is it nice that lately those AI videos have pretty much all disappeared?"

Doug: "No, those things you think are real..."

John: Exactly, exactly. That was his whole point. He just stared at the camera, and I was like, "Whoa. I didn't even see that coming." I think the way I think of it as a pastor is that these media companies are incentivized to maximize the amount of attention that we give them. That's the reason why Facebook has transitioned from interpersonal connection to essentially becoming another version of YouTube or TikTok, just giving you short-form video to occupy most of your time. My guess is that the novelty of that is going to go away pretty fast, and that scarcity is going to draw people toward authenticity, or at least some kind of validated quality with time. At least, that's my hope.

Gordon: Yeah, I would also point to the fact that we have this interesting dichotomy right now where you have these short-form videos, 90% of which could or could not be AI, but we also—and this is, I think, an encouraging aspect—have a major move to long-form podcasts as well. You can argue whether or not people are different experts on that, but I do think that's encouraging, because you have people sitting there and listening to something for 30 minutes, and I think it points to them looking for that authenticity.

I also would take you back to Proverbs: in the multitude of counselors, there is wisdom. It is important to have those critical thinking skills for yourself, but they're only going to be useful if you spend the time to develop them.

John: Yeah, that's right. And you know, I think on one level, I'm happy to look at a tobacco company and say, "Hey, make your cigarettes less addictive, right?" But I can't talk to Bezos or Zuck or these guys. What I can say is: you should stop smoking. We should actually be spending far less time online. I think as the quality degrades, it's actually a great opportunity at the ground level to recognize that we've actually been eating slop for a long time, and we should probably put it down.

Audience Member: This is a perfect follow-up to my second question. Do you think—maybe even it's baked into the nature of this conference—but do you think it's on the level of a discipleship conviction to inform loved ones and friends about some of the things we've learned about the nature of AI today? I'm sure we have friends and family who are almost just enraptured by the magic that AI seems to offer. Do you think it's on the level of a discipleship conviction to inform them about the nature of these tools and what actually goes behind them, and even to steward their humanity? Or do you think maybe it's a Christian liberty category where people are using it, and I might not believe it's best for them, but it's fine to leave them be?

Doug: I think I would distinguish between common usage. Students cheating and writing something with an AI is less dangerous to me, and turning something in that was actually AI-written is far less dangerous in my way of thinking, than someone who thinks they're in a relationship with an AI, because that's soul-damaging. One person is cheating their own education and cheating—this isn't the beginning of cheating; cheating goes all the way back—but the other is someone actually damaging their own soul. So, I would definitely intervene in that sort of situation, and then just treat it as a wisdom thing otherwise.

Gordon: Yeah, I think as the Spirit leads, and being informed by the Holy Spirit in that sense is key. I would say it's an issue of conscience, and then definitely a wisdom issue.

John: I do think in the same way that we ask questions now for accountability that we never had to do before—like, I have to ask guys, "Hey, did you clear your browser history, or how's your thought life?" or "Have you been seeing things on social media? Who are you meditating on? Tell me what's going on." I think there will be something related to that with AI in the future, where there will be more questions that we're going to have to prod and ask people to be open and honest about because of AI's ability to seize our anonymity and make us grow more and more isolated.

Doug: So when you ask elders, "Have you been looking at pictures of women?" they'll say, "Well, do you mean real women?"

John: Yeah. If you're curious about that, you could just email Jeremy Meeks on his dissertation on AI sex robots. Anyway, next question.

Audience Member: Hi, thanks. Regarding world news, I heard that some states are considering using AI as teachers for younger students, and they've seen improvements in the students' academics, individual instruction, and one of the reasonings is that the AI instructor is able to cater to that particular student's attitude and disposition and all that. What are your thoughts on that?

Gordon: To get a little more specific, are you talking about Alpha School?

Audience Member: What do you mean by Alpha School?

Gordon: That's a specific set of schools that is using AI very intimately in what they're doing and in their pedagogical approach.

Audience Member: I don't know, they weren't particular. I think it's more primary school. The younger students, they're seeing more improvements on test scores, especially on reading and comprehension.

Doug: Just speaking as a scientist, I would say it's very hard to get reliable, believable results on human behavior, just because humans are complicated. So a study like that—could people in this room outperform people in that room, and there's a human teaching here and a bot teaching there? Sure. But you could do other studies, I think, that would show it the other way; it depends on how the human is performing. In the end, the very best way of getting education would be this soul-to-soul connection. And the very best way for you to be understood is for a person to understand you, not for a machine to change its way of providing information to you based upon data and training. That's an impersonal way to personalize, basically.

Gordon: Yeah, I think some of the schools that I've seen do that—the ones that I've seen are incredibly expensive, but they have a major human component to it as well. One of the ones I just referenced, they use a combination of things. So, it's not a singular "put a student in a room with AI." It is used as a tool as part of their pedagogical approach, but there are many, many teachers—they don't call them teachers, but whatever they call them, they are very involved in the process, and it's not the whole time either. I know they have almost a dividing line in my understanding of it, where the whole first part of the day is essentially no AI. I think it's a little more complicated than just a tagline, if that makes sense.

Doug: This is a little bit tangential, but I think it's related to your question. One of my daughters is a rocket scientist; she works for NASA JPL. When we were watching the Artemis thing and she was giving us deep insight into what they're doing and why this is complicated, a question that I asked her—and a question that a lot of us must have been thinking, especially if you're my age—is, "Hang on, it was 60 years ago we had people walking on the moon, and now we're patting ourselves on the back for looping the moon. What has happened here?"

I put that to my daughter, and she's got some experience now with NASA JPL. They were heavily involved—they don't do the manned missions, but they're heavily involved in all of the rover missions, the Viking, and all the exploratory missions. She knows her stuff. Because does it not seem strange that this is technically complicated 60 years after we apparently did something that was much more technically complicated? It almost doesn't make sense. I'm not saying there really were people on the moon; I'm not saying it was faked. You know what her answer was? She said, "Because we stopped doing manned missions, the people who really knew the nitty-gritty that's not written down, they all died. And there's nobody alive today that knew all the stuff that you can't actually write down in a manual that actually makes this stuff work. We lost a generation of people who actually knew intimately the details that you have to sort out to get something like that to work. So, we're having to rediscover stuff that people knew, and they're gone. We can't ask them."

That reinforces for me the crucial, indispensable human element in discovery and knowledge. There are things that no AI is going to glean because it's not there; it's not written. Think about the things that you do, and if you were to write down how you do them, it's going to miss a lot of the actual details because you can't write down everything. But if you were training somebody in person to do something—if you're a plumber or an electrician—you can say, "No, don't hold it like that; hold it like this." Because there is all kinds of detail that would be laborious to try to record, so it's not recorded, but it's important when the rubber meets the road.

Gordon: Like I had said before, that's what I want to capture: that intuition, because it's so valuable but so hard to capture.

Audience Member: Hi. If we're thinking about how AI will continue to be incorporated into our lives, do you think we should be teaching students or even our children how to use AI so that they can keep up with the workplace in the future, or do you think that's unnecessary and we should keep them away from it?

Doug: I don't think we should be over-protective, though we should be protective with kids. But I don't think forbidding the use of AI is a good way forward. I think we should find safe and helpful ways for them to get experiences, because I do think AI, even if it goes forward in a way that's somewhat analogous to GMOs where a lot of applications of AI receive huge pushback and they end up not being a part of the future, definitely AI is useful enough that it will be a part of the future.

So, I don't think not letting our kids see it is the answer. I have the same attitude as an anti-Darwinist. My whole career is showing that life is designed and it didn't evolve. I would never say that you shouldn't teach your kids what Darwin says, because they've got to know; that's out there in the world, and they need to know the truth as well. I think AI would be something like that. Some exposure to it, carefully and age-appropriately, is going to be important because AI is definitely here to stay.

Gordon: Yeah, I would fundamentally agree. Having two young daughters, I think it is something where you have to think critically as a parent on how you are honoring the Lord and preparing them for the future, and what's best for that child, because every child is different. Some of the ways I think about it is the wisdom surrounding screen time and those types of things. Not that they're exactly the same, but I think there's that aspect where I would agree with what you just said. You have to think about it critically. If we put it into an all-or-nothing bucket, you are overindexing on something that may actually end up hurting your child.

John: I've been reading a book on the history of textiles and the creation of fabrics. Absolutely riveting, fascinating stuff. Literally, the word "riveting" is from textiles. But one of the things I found really interesting about it was how much of technology and civilization had to do with our ability to make clothing so we could stay warm. And none of us know anything about that anymore, right? Because it's all automated.

Even cursive handwriting is a technology that we had to take advantage of for the time period where it was useful, which means most likely even this event that we had today will be dated pretty quickly. Thankfully, we didn't do a workshop on how to use AI, because that would be dated in like three months. But even then, at best, our efforts to understand the present moment of technology are going to be ever-changing, whereas the things that are unchanging are soul care, knowing the humanities, being able to think critically, being philosophically minded, and having a strong sense of ethics. I think those are all things that the Christian life provides. Those are things that actually carry over. Everybody needs to know how to use technology in order to be able to function today, but the stuff that's evergreen are the things that I wish we really prioritized more of our time on, and that we hold all the other things that are going to continue to evolve with a more open hand.

Audience Member: Over here, kind of following up on the question of the issue of AI's psychological impact. I met with this young man yesterday. He's about 27 years old. I'm discipling him, and he's looking for work and really struggling with trying to get work. He tells me, "Hey, Alex, every single day I'm on the computer applying for jobs and doing this." And then he told me something that I had never heard before. He says, "I get these interviews from these corporations, and they put their avatars on, and they do the interview with an avatar." He's got a little time clock going down on the bottom, and he's answering the question, going through this whole process of doing this interview with an avatar, not with a person, but he never hears anything back. He never gets any feedback. That's what he was telling me: "I never get any feedback. I'm struggling." He's getting depressed. He doesn't know what to do. He's getting discouraged, and he's been dealing with this unemployment situation for almost six to eight months.

I know a lot of young people that are not equipped for the higher-qualified jobs—we would say those that may have a degree or certificate or license—but they're coming out of high school, and they are just hitting this wall compared to when I was their age. I'd get my little resume, walk down to the store, say, "I want to apply," and talk to the manager. Is AI doing anything in that area of psychological impact, and how is this new technology impacting a new generation of kids who have no idea how to map through all this? They're very young, and I'm just curious if you guys have had any discussion on that kind of stuff as you develop this stuff or in your circles, because it is impacting a whole generation where they get apathetic, unmotivated, and not willing to sweat and work the way you're talking about they have to do. I'm just curious if there is any discussion going on in that arena.

Doug: I think it says something about a company if they are automating first-pass interviews. That tells you something about the ethos of the company in my mind, and do you want to work for that company even if you make it to the second pass where you get to see a real human? Because they've said something about the way they're willing to treat people, and I don't think that's acceptable. Someone comes into a job interview with the sense that there's something on the line; if they're astute, they've prepared for it, they're nervous, there's a personal expenditure in going to an interview. For there to be a bot on the other end, I think, is dehumanizing.

There would be options, though, because you can still do what you did. This person could take their CV and go knock on a door, and go face-to-face with someone. I think getting jobs and getting places at universities is often aided by networking. You could have the jaded view of networking as just LinkedIn, but I think it goes back to: humans were made to relate, so networking in a positive sense is actually a good thing. It's people relating with people.

I don't know if the COVID era had an impact on people's social confidence because we were antisocial by mandate for a while, but encouraging young people—and I'm seeing this with our students—to recultivate the skill of going face-to-face and having a conversation with people is so important. If you go meet someone and say, "Hey, I'd like to work for this company. Can you introduce me to someone?" and you have a conversation, and then you say, "I'm going to personally hand you my resume. Can you see if I might be a fit for something?" I think you'll feel seen, and I think the result can be better than if you're just in the big end of a funnel.

Gordon: Yeah, I would agree completely on that. I would also use an example. I live in Seattle, so I flew down here, and I'm sitting in this chair with you. The impact of that is vastly different than if I was up on that screen, where I could have given the exact same talk and been saying the same things, but the impact is completely different. I think that is one of those things that we have lost.

If you talk to most sales reps—I'm not a sales rep, but what are they doing? They're going in person to be there with people, and they are probably the best at networking. That's a major part of their job. I use that as the example right now, but I think that's something that needs to be encouraged, fostered, and taught how to do. It still exists, even if it may be slightly different. It's hard to cold call someone, and it's hard to reach out, whether that is through LinkedIn to somebody truly new who's a hiring manager. But what I'm seeing for those in this job market who are successful are those who build those connections. It's the relational aspect that is actually getting them the job.

John: Real quick before we get to the next question, does anybody have a question who hasn't asked a question before? Could you just raise your hand up so David could see you? That's one. Anybody else? It's okay if you've asked a question before; I just want to make sure we can note anybody who wanted to ask a question. Okay, and in the back. Great, thank you.

Audience Member: More recently, as AI is getting more prevalent, younger kids are being more exposed and involved with AI as well, whether it be using it for their homework or consuming AI-generated content. A concern that I wasn't able to disprove in my head was essentially: right now, we have this idea of AI, we're conscious of what AI is as a powerful tool, and we know there is this limitation we need to be careful of. But when it comes to these younger kids, the more exposed to it they are, it seems like the more content and satisfied they are with everything that AI gives, just being more and more dependent on it as well. If that were to go on for the next generation, where as time goes on, fewer and fewer people are actually aware of the potential concerns that AI might have, would it be a possibility that a time would come where pretty much everyone is dependent on AI, or just has a lack of concern or discernment for it?

John: So, you're talking like a Wall-E world type situation?

Audience Member: Well, I guess that's one way to put it. It was definitely better structured in my head.

John: No, no, sorry, I didn't mean to poke fun. I just wanted to make sure I understood.

Gordon: I mean, I think that's why educators are probably the most important thing we need to be focusing on currently. Both the enrollment cliff along with this massive change in technology is fundamentally affecting their pedagogical approach of how they've been teaching, and making it very easy for students to shortcut the learning process. I think that is something we as a society need to be very focused on investing time and resources into. Personally, I think it's probably one of the most important things because we are sinners, and I think laziness is a part of our sin nature. As a result, we're going to find the path of least resistance in most cases. So, really focusing on how we as believers are informing that, pursuing that, and the potential for Christian universities to really shine in how they do that is key. Because they are thinking about that clearly, and God willing, they have a number of believers there that are ethical and following those rules. Not that they're perfect, but they are going to come out on top.

Doug: There's a huge parent role, too, especially in the earlier years. Pastors, parents, and professors—the three Ps—or teachers all have a responsibility to be pouring into young people and growing them up into something. Scripturally, it's the Holy Spirit and God's work on us that conforms us to the image of Christ; that is what we're looking for.

There have been things—again, we go back to earlier technologies—just screen time generally, putting aside AI. It's not uncommon to see someone in a stroller with an iPad in front of them. I think all of this has had an effect, and I think parents need to own part of that responsibility. If you're a parent, it's tempting to do whatever makes the young child stop fussing, and a screen is an easy out, but it comes at a cost.

You said, "Can we foresee a time when everyone goes this way?" I think there's something beautiful about the fact that there are always people who rebel against these things. I think of the maker movement; it becomes trendy to not do the thing that everyone else is doing, particularly if you can say it's not healthy. How many people these days could get a fire going in the woods without a lighter? How many people could get their food in a Bear Grylls survival style? These things become trendy because it's so outside of what the general trend is that people say, "I want to go completely the other way. I want to be a survivalist. I want to think about how you build things by hand." And you can get a following because everyone says, "That's actually refreshing."

One of the things I think is refreshing, which is very low-tech, is being in a quiet place, having a book, having a binder with paper in it, and a pencil or a pen, scratching away notes and thinking that way. No phone, no internet connection, no laptop. Some of my most cherished intellectual moments of meditation growing up were that. There was technology, but me being on my own somewhere with nothing technological—just a writing implement, something to write on, and something to read, scratching my head and thinking—that's very attractive. So, for parents, let your kids see how attractive those things can be. That's not to say you can never have a phone or a computer; it's just to say learn the beauty and the joy of simple things that humans have always enjoyed, and grow people up to love those.

Gordon: I have a small follow-on comment on that one. I recognize for myself in my family—I think I mentioned I'm fairly dyslexic, so I usually listen to the Bible—but I realized that my daughter would never see me actually read the Bible because I'm listening, right? And fundamentally, that comes through the phone, which is kind of impersonal because anybody who has a phone that they are looking at, it immediately cuts off interaction with those around them. So, that's something that I've looked at for myself: okay, if I'm around her, what should I be doing to really demonstrate that? I think each of us has some of those unique things that we should be thinking about that communicate in our actions, regardless of whether they may be hard for us, because they are important to do.

John: That's such a good point, because you're not just lecturing your kid, "Hey, don't use your phone." She's observing you and seeing what you're doing and how you behave. Yeah, that's great.

Audience Member: Thank you. Doug, I thought your point about the fake danger of AI becoming like us versus the real danger of personhood atrophy was a helpful distinction. I was wondering if you had some more specific examples of things to watch out for as far as personhood atrophy, and then also on the other side of the coin, ways to exercise our personhood and fully lean into that.

Doug: Yeah, I started putting together a slide yesterday where I had the categories—relational, intellectual, and I forgot what my third one is. I was filling this out just yesterday, hoping to have a book contract about personhood. It's not an AI book, because the publishers say they've got so many AI books; it's more like a Christian anthropology, a Christian understanding of humans that is robust in an age when we're under assault from machines. These are the beginnings of ideas that I want to develop more fully.

But you can see in these relational and intellectual aspects, there can be shortcuts. Instead of going through the hard process of developing relationships with real humans—and everyone who has had relationships knows that there are hard times in relationships. Anyone who's married knows there are hard things that you have to work through in marriage. But you also know—and you're going to be married, so listen up here—you also know that you benefit tremendously by walking through those hard things. If you don't really have a taste for doing hard things, and when we're lazy we don't, you're going to look for getting to the end without the hard things.

I think that's where people are going to be tempted to go to a machine that says nice things to them, agrees with them, and can keep a conversation going, rather than doing the harder thing, which is to go to another person. Maybe a person they don't know that well, maybe they're a little uncomfortable, maybe they have a hard time starting conversations, but those are muscles that need to be developed. It's human nature to resist the thing that's a little bit hard, but if you recognize that there's something really good that comes out of that, then we need to push ourselves and push the people we love. Go do the hard thing, because that develops things that need to be developed.

And intellectually, that's why you hear about people who say they're letting AI do the "ideating"—that's the word they use. "I wrote the paper, but I had AI do the ideating." And I'm thinking, "Dude, ideation is the thing you should own." If an AI helps you with your grammar, I'm okay with that, and if a computer helps you get the spelling right, I'm okay with that. But if you grow up and you can't ideate, if you can't brainstorm because you never tried it and you always punted to a machine, you lost something huge. Trying to get young people to understand how big a loss that would be is part of this.

John: Alright, our last question.

Audience Member: Oh, last question. Okay. Oh, I should ask a different question then. Could you maybe just list one or two authors or people who have influenced you from a Christian perspective who are writing on AI? Because I only really come across mostly secular resources when it comes to thinking about AI and AI alignment. Is there someone doing really good work?

Doug: There's a need for more. I'm a friend of John Lennox, who wrote a book—what was the name of it? It has a year in it, I think.

John: 2084.

Doug: Yeah, 2084. So, the author is John Lennox. He wrote something that's very solid from a Christian perspective. One of the challenges with AI is that if someone's trying to write something that's telling you about all the latest gadgets, it's going to be obsolete in three months. As we've already said, that's not the kind of book you should be looking for. What I'm wanting to focus on is the eternal, true, fundamental stuff that will always be true, and whatever AI does, it will never become like us.

I know a number of colleagues who are doing more technical, paper-type stuff. Morettto Gouta is a philosopher who's also become well-trained in neurobiology, and he and I are going to do some work together. I would like to have a book come out. Does anyone else know of things that have been helpful? I've read lots of short things, but I haven't found a long thing that's really good.

Gordon: I would completely agree that the challenge with doing a long-form book is that it's just moving so fast. I do think, though, that the way I approach it is through foundational principles and concepts. In a lot of ways, going further back is actually really going to help your thinking in how you think about the principles now.

Going back to those quotes—those were 150 years apart, and there is good thinking that needs to be foundational to building and being able to think about something that is moving very quickly. If you have those foundational, core principles, they're building that wisdom muscle and building that intuition. So, I would go back, from a biblical side, to some of the Puritans and some of their thinking around different things. I would also go back and read some of the foundational philosophy works, almost going back to those roots to think through some of those pieces and then build up from there, which is a lot harder than just getting a book on it. On the latest pieces coming out, I think the most valuable are some of the shorter-form blogs and things that are actually meeting things as they come, not necessarily on the book front.

Doug: Another book that came to my mind is by a colleague named Robert Marks, who is a professor at Baylor. He's written a book called Non-Computable You. It's a little bit nerdy, but his point is that there are certain tasks that are provably non-computable—they're not algorithmic—so a computer will never perform them. What he's saying is if you understand what these things are, these are the things that you'll never have an AI do, and things that you can do. So, that's another decent book.

John: Can we give a hand to our speakers? Let me just close our time with a little reading from Daniel 2. This is Daniel interpreting King Nebuchadnezzar's vision of this statue with clay feet, iron, bronze, silver, and gold. You could just imagine the technological eras that are captured with these different kingdoms, and this is what Daniel says will happen:

“In the days of those kings, the God of the heavens will set up a kingdom that will never be destroyed, and this kingdom will not be left to another people. It will crush all these kingdoms and bring them to an end, but will itself endure forever. You saw a stone break off from the mountain without a hand touching it, and it crushed the iron, bronze, fired clay, silver, and gold. The great God has told the king what will happen in the future. The dream is certain, and its interpretation reliable.”

— Daniel 2:44-45 (CSB)

If you want a glimpse of the future, don't look at your GPT chat history. Take a good look at your church tomorrow. Let's go ahead and close our time in prayer.

Lord, we trust all of these things to You and ask, Lord, that You give us a glimpse of Your glory and that Your wisdom would inform us so we can love You and love others rightly. We ask that Your Spirit would help us to do this in Jesus' name, Amen. Thank you, guys.