Who’s Really Choosing? A conversation with James Brusseau

The International Schools Podcast · episode 176 · upload date 4 April 2026

Full-recording local machine transcription, subsequently reviewed and corrected by James; his prior corrections and deletions are preserved. James Brusseau explicitly confirms the speaker labels identifying his contributions; those labels are no longer provisional. John identifies himself by first name in the opening; the upload lists John Mikton. This author confirmation is distinct from independent listening or voice-identity verification, neither of which occurred in this presentation correction. It does not establish acoustic verification of every transcribed word. Earlier provisional attribution evidence and unresolved source-allocation and acoustic-boundary qualifications remain documented in the companion and notes.

The frozen original and timing/source companion retain source provenance and current-edition mappings; earlier snapshots preserve the prior editions. Review notes and wording queue describe exact limitations. No slides. No registration, indexing or evidence activation.

Interview

John Mikton: Good afternoon, good morning, good evening. This is John, your co-host at the International Schools Podcast. Wonderful to be back. Dan is quite busy and, you know, as we're recording all these various intense geopolitical things are happening around us, so our thoughts go out to everybody that is trying to deal with that. These are really some very challenging times.

One of the things that as we have done many episodes and often we've talked a bit about AI as something that it's a transaction, you ask it a question, it gives you an answer. There's been a bit of a shift now with agentic AI. There's a lot of hype around it, but there are now tools within the large language models like Claude and ChatGPT that they have an agent mode. And that changes things because now it's acting. It's actually doing things for you, often on a command and then it goes off and does it.

Either way, one of the things that I think this all brings to the table is the ethical dilemmas that we encounter as educators, as professionals, as we're working with these tools and these AI tools. Some people feel embarrassed that they use them. They feel like they're cheating. Other people are very upfront. So I use it every day. There's the spectrum of where we stand. But the ethics is very challenging because here we're dealing with intelligence and there's a wonderful paper by Luke Drago called The Intelligence Curse. And I definitely recommend it; we'll put it in the show notes. But the question is, what does this mean for humanity? What does it mean for intelligence? What does it mean for free thinking, critical thinking, metacognition? All these things come into this area when you think on the ethical level.

And today I feel very privileged because I had read an article in the Guardian about various things about ethics and deep fakes and a professor was mentioned and that's Professor James Brusseau, who's at Pace University. And I reached out to him and, guess what, he's here today. And I just feel so lucky and we feel honored to have him here today.

So, James, a warm welcome to the International Schools Podcast. And as we start, I think with the first question you can kind of tell us a bit about yourself. What first pulled you into AI ethics? Because you have published a lot, you're an author, you're a professor of philosophy. What, you know, what has pushed you into AI ethics, and what are you teaching, working on right now that feels most important?

James Brusseau: Sure. First, of course, thanks for having me. I'm happy to be here and reach an educational audience. Of course, as a professor, I've dedicated my entire life to this pursuit.

And then I think perhaps one aspect of my past, of my biography, that connects a little bit with what your audience might be interested in is, as it turns out, I was a physics major in college and I really enjoyed the sciences in high school and so I think that there's something about education and the passions that we develop in education that stay with us. So I learned physics and entered physics in college, but you remember from high school the way physics was? It was you could shoot an arrow at a certain angle at a certain velocity and you could predict where it will land. That's the way physics starts, Newtonian physics. But by the time you get toward the end of college, what happens is physics, that visual part to physics, the practical, tangible part, evaporates and you're left with just pure mathematical formulas. So physics becomes almost like math. And so, in my case—and all this is answering what got me into AI ethics—in my case, I began to lose interest in physics because it was so abstract. And for that reason I switched over to philosophy, where I could think about practical human problems again in tangible terms.

And so I went on with my life doing philosophy and things. But then, you know, about five, ten years ago or so, all of a sudden we started having artificial intelligence breakthroughs which were significant enough to change the way, as you know, the way we are living our lives. And so, happily, because I had in my background, in my physics studies, sort of the mathematical and the technical ability to fairly well understand the way artificial intelligence works and the way computer scientists think and operate. Of course, I could not contribute to those fields, but I understand what they're doing theoretically and so on. So it was a very natural connection for me to bridge from my second half of my academic career, the philosophy back to the first half. So that's how I got interested in AI ethics and brought together these two parts of my intellectual life.

And I think that's something that for education, that's why there's such great value in what teachers do in education generally, because even if it appears that a student completely loses touch with the Shakespeare he or she studied or the sociology that he or she studied at some point, it's very possible that some of that will come back for some unexpected reason later in your life. So I think that the reason I'm interested in AI ethics connects very well with the idea of education.

John Mikton: And I think your point is a really important one, because sometimes, as teachers, you have a kid that's in the front or in the back and they look disengaged, and then ten years later you hear that actually you had an impact on them. And I love the way you connected this idea of your physics and then algorithms and then the ethics.

Because you are at Pace University currently teaching and then you're also an author so these recommendation algorithms are something that you highlight. Talk to us a bit about what are some things that you're noticing on an ethical level?

James Brusseau: So I think there are two things that I think are happening today that are very important. One is recommendation algorithms, and I will talk about that in just a second, but another one as we're sort of on the subject of education, another project that I'm working on which I think is also very topical today, and perhaps even a little bit fresher is a project in something called mimetic artificial intelligence. So what is mimetic AI? Think about in the old days, we used to be able to play chess against the computer. And so you would play against a sort of generic opponent. And of course now we know that chess can beat almost any human, so the chess computer can beat us easily. So that's a generic AI tool that will play chess. But what a mimetic AI tool does is it tries to play chess not in some abstract sense, the best possible game. Instead, it tries to imitate a specific player like the Grandmaster Magnus Carlsen or Kasparov or Karpov. So the idea of mimetic AI is it doesn't try to be the best at something. It tries to imitate a human expert in something with all his or her strengths and also his or her idiosyncrasies. I'm not sure I want to say weaknesses, but the mimetic AI might very well lose in a chess game to the abstract generic chess playing computer.

But I think that this mimetic AI is very important on the subject of education for the following reason. Because every professor, at least in the humanities, develops his or her own kinds of examples, his or her own kind of vocabulary, his or her own kind of intellectual frame through which he or she teaches material. And so you were saying at the start that many of us are afraid—students, professors, et cetera—to talk about ChatGPT and how we use it. I think part of the reason for that is because the tool tends to give back answers which don't fit anywhere. So if you have a Shakespeare professor who teaches Romeo and Juliet by talking about, oh, Taylor Swift, and she has a song from her younger times sort of like a Romeo and Juliet song. And so this kind of professor teaches in a very popular culture way. And then another professor who is equally good and presents an equally compelling set of classes on Shakespeare might instead talk about the Globe Theater in England and the time that he or she spent in the Globe Theater and have a very sort of sophisticated venerable approach. So these are two different approaches. Now, if you ask ChatGPT about Shakespeare or philosophy or something else, it's gonna come back typically with one of these aggregate answers which doesn't sound like any particular person. It doesn't sound like the pop culture Shakespeare professor or the Globe Theater Shakespeare professor—imagine this person has an English accent and everything, the whole deal.

So in the field of mimetic AI, one thing, a project that I'm actually working on now, it's called The Caffeinated Professor, is we are trying to teach language models to imitate specific professors. Specifically, in my case, me, as we're taping my lectures and trying to have the chat bot or the agent be able to answer questions for students in my AI ethics courses and answer them in the way that I personally would answer them as opposed to giving this kind of generic ChatGPT answer, which doesn't fit with the kinds of examples that I typically use, the vocabulary that I employ and so on. We think that that might be much more valuable for students because it allows them to continue their learning trajectory in the language and with the ideas that they have developed from their human professors. So this is a way that I think we can use, we talked a little bit about kind of our humans turning into AI or AI turning into humans, and this is a way in some sense of kind of taming the beast. Making sure that AI serves human kinds of pedagogical interests in teaching as opposed to having the other extreme when the humans just serve ChatGPT.

John Mikton: When that happens, you know, students are just going to ChatGPT to write their essays, and professors are using ChatGPT to grade student essays. And at the end we have ChatGPT talking to ChatGPT, and the uniqueness of the professor and the uniqueness of the student, that's what disappears.

So fascinating. And can I ask you a question? So you are doing this at your university, at Pace University, using a large language model, and I assume you have a team of graduate students, your team working on this?

James Brusseau: You know the situation well, exactly that. Just today I wrote a letter for a student who's on a visa, saying that he's working for this project. So yes, that's exactly what we are doing and experimenting, yeah. So we are working on a second model. The technology is advancing so rapidly that things that happened only nine months ago are really no longer applicable in this field. In the field of mimetic AI specifically here, we are learning very quickly how to use a kind of retrieval, a kind of search, which allows these language models to go first and look through a set of data, which is the data from a particular professor, my recorded lectures, my writing, so that the tool now is able to go through and look at that data first when it responds to a student instead of mixing that data with all the math of the general language model underneath it. And that's very recent, this new ability.

So we did last year—I had an initial model about a year and a half ago. And it had some success and some failure, both technically and socially. Technically, sometimes it did replicate me; sometimes it didn't. And then let's say pedagogically, some students found it helpful. Other students did not. Typically, what we discovered is that if the questions are more specific, the tool works better.

However, that's in the past, we're redeveloping the tool from scratch now. And we believe it will get much better performance and we believe that we will be able to do things that are very helpful like the following. In my classes, in at least one of my online courses students contribute to a discussion board every week. We believe we will be able to make this tool able to respond to the student's discussion contribution immediately. That is, the student will write their short paragraph responding to a case study and then the tool will be able to respond in something like the way I would and that will allow the student to continue working on the problem. Now for a human professor or me, that's impossible. I cannot be online 24 hours a day to always be responding to the students. So this is an example of how we think we might be able to get a better response from the students with this second tool because it will work better and that will allow us to use it for examples like I just stated: a live interaction with students on a discussion board.

John Mikton: So I think that's the overarching theme I get, is that, you know, because there's a lot of talk—I just have some camera issues, so ignore me—but one of the things on the ethical side is, like, how much are we subcontracting out? And there's this issue of cognition loss and, you know, critical thinking disappearing. I mean, I'm blowing it out of proportion, but there is definitely a sense that as we subcontract, this cognitive offloading takes off. But what I'm hearing you're saying: wait, hold on, there's this way of doing AI that is really very much richer because you have the uniqueness of this individual, their expertise, their knowledge base, their idiosyncrasies, their humor. And this really gives a much more personal, authentic interaction with a large language model that has these guardrails of, for example, James, the AI ethics professor at Pace University.

This is obviously something that not many people are doing. Why do you think sometimes, and I really, you know, when AI helps people to choose what to read, watch, buy or learn, how is it also influencing choices without us really noticing it?

James Brusseau: Yeah, right. So this is your original question back at the start with what are some areas that I think are very important in AI. And so I think that they're, in my experience audience percents right now, there are two areas which are very important right now, and the one is that the mimetic AI we just talked about briefly, and the other is this area of recommendation algorithms. And so recommenders are almost invisible but extraordinarily powerful.

So, for example, I saw a statistic several months ago. And I don't know exactly how trustworthy it is, but I'm imagining it's fairly close to right. It was a graph, actually, of how couples, romantic couples, had met over the last century or so. And of course, 50 years ago they met through—100 years ago they met mainly through family and friends. Through the 70s and 80s we started to see people meeting more and more at work. But then, just in the last 10 years, this explosion to online meeting has been remarkable. And now, according to these statistics, there are more people that meet their current partner online through presumably these kinds of dating apps like Hinge and so on. They meet them online more than in every other way put together. So recommenders are controlling, to a significant extent, our personal lives, because it's a recommendation algorithm that initially decides who the people are. I'm looking for my phone. Here's my—I won't ever show my phone because I might be advertising something. But, you know, I have my phone, swiping left and right. It is a recommendation algorithm that decides from whom I will choose that initial group. Let's say, in my case, how many women are there in the world? Well, in the United States, 150 million women, I suppose, are out there somewhere. Obviously I can't go through all of them. I only have time, you know, when I'm looking on these apps, you know, I would only have time to go through a hundred every day if I was very aggressive in it. So who makes that choice? Who narrows that list of possibilities from 150 million or something like that to 100? Well, that's a recommendation algorithm.

Something similar is happening, for example, in LinkedIn, the kinds of jobs we are exposed to, the kinds of people that we meet professionally. Of course, these tools do a tremendous job of finding ways to connect us with people who have similar interests, can help or participate in our career. So this is not a criticism in any way. However, it is to point out that all of these connections that we get on LinkedIn are a tiny subset of the possible connections that are out there and the recommendation algorithm is deciding that for us invisibly.

And then the easiest example, I think, is something like Netflix. There's a good experiment I read about. It was one of the magazines, Atlantic magazine or some such, where someone talked about the experience where they checked into an Airbnb, and somebody else's Netflix account was signed in. And they were astonished to find that all of these videos had no connection whatsoever with the videos they watched on their Netflix. They had no idea these videos even existed. So that's just a simple human example of how there are all these movies out in the world. But we have the experience, those of us who have Netflix, of feeling like, well, there's only very few. There's only this little group of 20 or 30 that seem to keep popping up all the time. Well, so there too, from entertainment to our professional life to our personal lives, recommendation algorithms are choosing what we can choose from. And so that's how they work invisibly.

Or maybe just one more quick example I might give, of what recommendations do. If you think about travel, we travel frequently to discover new things, new food, new ways of dressing and living and so on. But what recommendation algorithms do, more or less, is they decide for us where we are going to travel to. And then once we get there, we can kind of look around ourselves and find this restaurant or that restaurant. But the big decision, you know, where we're going, that's what the recommender does. And the recommender does that in LinkedIn, the dating apps, Netflix, et cetera. So that's the power that the recommendation algorithm has. It has tremendous tremendous benefits, gives us terrific opportunities. But it does that by narrowing our possibilities dramatically and invisibly. It's not a bad thing.

John Mikton: Do you think then that actually as these recommenders and this idea where, you know, what we watch, what we buy, even what we learn as you're talking about the Netflix and the LinkedIn—is it coming to the point that actually we are not going to be making our own choices or we're going to be making choices only by, let's imagine Spotify just as an example, where you get your discover weekly. Well, I don't find that usually what they're discovering for me is what I want to listen to or they make assumptions. So my question is, is there a tipping point here at some point in your thought ethically where the recommendations now are entering a new phase where actually we don't have independent choice?

James Brusseau: Right. So I would say two things. The one is, intellectually, as a thought experiment, it is curious to ask whether or not we want choice and freedom. Now, I myself do want those things. However, it's reasonable to imagine, if you think, why do you and I have freedom? Why do we have autonomy? Well, usually, if you just ask someone on the street, the reason I have freedom to choose is so that I can get the things I want. That's why I want freedom. But if computers that analyze patterns of behavior can analyze these patterns well enough to know beforehand the kinds of things that I'm going to want, then you can make a rational case that freedom is redundant, that we don't want freedom because we can get the rewards of freedom even better by being slaves to these machines, assuming, of course, that the predictive algorithms are sufficiently expert to know what we want. I've talked about this a little bit theoretically under the idea of euphoria convenience. Euphoria convenience is that kind of convenience where you get what you want even before you know that you want it or despite what you think you want.

Just as a footnote here, I think you could make an interesting case that maybe sacrificing our freedom is not necessarily a bad thing. However, I myself do not make that case. I go the other way. I believe that, just intrinsically, freedom is good. And further, I would make the case that freedom is better than happiness. If you had to choose, I would say it's better to be free than to be happy. And my thought experiment there might be something like a drug. Like, let's say you could take a drug, give somebody this pleasure, like heroin or something, and you had an endless supply of this heroin. You could just keep taking it, but of course, once you're taking it, you're addicted. You can't do anything else, but you're perfectly happy, you're extremely happy. I myself would reject that way of living. So it seems like it follows that what I'm saying, in essence, is that I am choosing freedom over happiness if I have to choose, at least in some sense. We could have a long debate about that. Yeah, that would be the position I would take.

John Mikton: But people always talk about that, James, is this idea of the seamlessness, the frictionless, oh it's great, I just wanted those shoes, I didn't even think I needed those shoes and that's the pair I was actually looking for or we're going to make a walnut cake, oh I didn't know they were, yeah. So I think your point's a really interesting one, this idea that you know freedom or happiness, it has a bit of a brave new world tone to it.

James Brusseau: Yeah. So then you have the question of how do you operate that? Or if you are a freedom advocate as I am, how can we try and manage AI to preserve our freedom as opposed to just giving us happiness? And so there—and this leads a little bit back to the Caffeinated Professor, actually—there, one tool that I proposed, and others working in this area also propose, is a kind of transparency with respect to the confidence the tool has about what it is predicting. So let's say you have Netflix predict—just a simple example—you have Netflix predicting movies. Right now what Netflix does is it just says, we think you will like this one, or we think you would like that one. Or sometimes they add, and I like this, they say, because you watched Mission Impossible, we will recommend Die Hard, right? Movies from our generation. Okay, so explained a little bit. But a stronger version would be something like Netflix saying, we have a 90 percent certainty that you will like this movie, but only an 80 percent certainty that you will like this movie. This kind of labeling of the confidence that the tool has might be one way that humans can keep engaged with the idea that they themselves are making decisions along with the tool, as opposed to letting the tool decide for them.

John Mikton: I think that always the logical stance to take is that to preserve freedom, we need to find ways for humans to make decisions with AI instead of letting AI make decisions for humans. And so that's the research direction, and the question is, how do we do that without making people annoyed, right? Yeah, you could have a rating, a star rating system. I mean, if you think of the way social media and reviews are, I'm sure there has to be a way of doing it. I think it's a very interesting concept. Because so often we just take for granted we're gonna use the Netflix thing. Netflix is gonna maybe pay us money because we keep mentioning them. But the Netflix is that we just assume—like your great example where the person went to an Airbnb and suddenly all the videos are like, I never knew these videos existed—is that we take quite confidently what Spotify and Netflix and Apple give us. We just assume because there's enough, like if it's 80 or 70 percent, there's enough places where you're like, yeah, that's a good choice, that maybe reinforces enough that you ignore the ones that weren't good.

James Brusseau: Right, right, exactly. And then the problem comes, though, and I'm the first person who's guilty. When I come home at eight o'clock at night and I sit in front of my TV, I just want the TV to show me something I want to watch. I don't want to bother with trying to struggle to be free and that kind of thing. So there's a real human problem here. You and I can talk—for me at least, it's in the middle of the day, I've had my coffee, so I'm in sort of good analytical shape. And so I enjoy talking about the idea that I will preserve my freedom and all these theoretical reasons and practical ways that will do that. But just like everyone else, it gets eight o'clock at night and I just want to be entertained, so it's going to be a real human problem.

And so I think that this will be maybe in a certain sense a wrap-up part of this question. One half of the answer will be developing technologies and ways of approaching technology which allow us or even to an extent force us to have intentions in making our decisions, and as I said one way to do that is to label how much confidence the tool has in one or another prediction. So that would be one kind of battle, the battle in this technology. But I think to a certain extent AI is going to be a human battle with humans. It's with us, with ourselves, and the tool is almost just an excuse. And in some sense I think that heroin example is not a bad one. I think in some sense it's like the drug example, and there is that temptation. And at the end the only thing that is going to stop you is you.

John Mikton: Yeah, your drug is almost gonna be an excuse. And I think so often we blame it on AI and we don't blame it on our own personal discipline and capacity for critical thinking, and also the resilience of keeping that, you know, the human in the loop.

One of the things that you had talked about, you know, when you're talking about these recommendations and this transparency, non-transparency, there is this term, the black box. What does this actually mean in simple words, when we cannot see the reasons behind the decision? Yeah, so of course, as the word itself indicates, right, we don't know what goes on in the box.

James Brusseau: I have sort of a good example of that, I think. One of the classic tasks of AI is to identify, classify, is to identify dogs versus cats or kinds of dogs. So in these early days of classification AI, a tool was developed that did an extraordinarily good job of identifying huskies. That is the Alaskan snow dog, right? The one that pulls the sleds and that kind of thing. Why does this tool do so well to identify these dogs? I mean, we have so much trouble just otherwise pulling apart dogs and cats and that kind of thing. These guys must be experts or just geniuses at figuring out how to know what a dog is. Well, people kind of kept looking and trying different pictures and so on, and it became apparent at some point that what the computer was doing was identifying a dog as a husky because there was snow in the background. That was it. The picture had snow in the background, then the dog was identified. That worked a lot, actually.

So when we say the black box, you need to understand exactly what that means. The black box doesn't mean that we cannot trace exactly how the computer makes its decisions. We can do that, though it's extraordinarily painstaking. In practical life, we could not do that for every decision. But for any particular decision with enough time and effort, we could trace how the computer makes its decision. The problem is that the computer has no human common sense. That is, the computer doesn't realize when getting the right answer is getting the right answer for the right reason. The computer doesn't do cause and effect. So the black box really means that. It's not exactly that we don't know how the computer comes to conclusions. It is instead that we don't know what the computer is basing its conclusions on. And so it could be the snow in the background of the picture as opposed to the face of the dog. And so that leads to all kinds of problems since we don't know what is at stake when the computer is classifying. We don't know what evidence the computers rely on. In some sense, that's even worse than the other version of the problem which is that we can't even understand how the computer works. It's worse to understand how the computer works but not know what it's concentrating on so we can't draw any conclusions. So I hope that's a human example of the black box. We think that the computer is identifying dogs as huskies because they look like huskies. Nope, that's not it. You're identifying dogs as huskies because there's snow in the background.

John Mikton: That's just so interesting. And I think your point is really interesting because so often that black box is hyped as this like black box, we have no idea how these things are working and this is doomsday. So I think it's really important because there's so much hype and often with the hype misconceptions of definitions of words. So this, I think, is really interesting. But also you bring about the issue: well, we don't know what the evidence is, and that in itself is a dilemma which is almost more important than how it actually physically works.

James Brusseau: And so that kind of leads us back to the question. One of the oldest questions in computer science, and just in these discussions generally, is: what counts as intelligence or even, in a certain sense, what counts as consciousness? And we all remember that one of the first tests is this idea of the Turing test, right? That if you put a human being behind one curtain and a computer behind the other curtain and you ask them questions, just text questions and have them hand back their text, the idea would be that the computer is intelligent if you cannot discern which texts come from the computer as opposed to which texts come from the human. So, if you take that kind of pure behaviorist approach, you can see how we could end up twisted into pretzels by AI logic because AI does get a lot of things right, as I say, for zany reasons that make no sense, at least that make no sense to the human mind. Like, as I say, identifying a dog as a husky because there's snow in the background. So if we take that Turing test approach, just the pure behavioral approach, that if it looks like a duck, it sounds like a duck, it's a duck. If it looks like it's intelligent, it sounds like intelligence, it is intelligent. Then AI is going to be very, very competitive. It already is very competitive with humans in many ways. So if we go that route, then AI is kind of one of us already, I think.

The other route to go is to demand, or rather to distinguish or separate, ways of thinking. Not so much strength of thought, but ways of thought. So there would be something like, in math, like a calculator, of course, is infinitely more powerful than a human at doing math. But we don't consider calculators to be threatening to us mathematically because the way we do math is distinct and we've grown accustomed to that. I mean, I imagine when the calculator was first invented, you know, people were dumbstruck and thought, oh, this is amazing, this calculator can do eight times seven divided by four, square root, all cubed. And I'm 17 instantly in my mind.

John Mikton: I'm aging myself, but in my math class calculators were not allowed. We had the slide, you know, and most people very likely don't even know what a slide rule is.

But anyway, one of the things that many of our audience are dealing with is kids and students interacting with AI as you are with your students and they're young adults and adults. And I'm wondering how much can these recommendation systems that you talked about actually shape an identity?

James Brusseau: Yeah, I think tremendously. And my thoughts about this question are really fairly close to just the analog or the pre-AI thoughts. And if you think about, well, I'll just use my own life. I was shaped tremendously by my professors. The novels that I enjoy reading, for instance, I had a terrific Shakespeare teacher. And so I actually enjoy reading Shakespeare, but there, and I know myself, there is no way that I would enjoy reading Shakespeare were it not for that one particular professor. Right? If I just stumbled upon Shakespeare and read it, I would last about 15 minutes and say, I don't understand what's going on, and that would be the end of it. So I think that our professors shape tremendously the way we think and exist in the world because of what they choose for us to read and because of the way they present it. I suspect that AI will have a similar effect on us, though the tools that AI will use to suit us or to fit us to one or another kind of item will be different. But I suspect that AI will be like that best professor.

So let me give you another example from education. I talked about the project that I am working on where we're trying to make the professor a mimetic AI professor, just like me. But there's a different project by competing companies and things. And they are working on a generic tutor but they want the tutor to be suited to, personalized not to a professor but to an individual student. So what this tutor will be trained on is the student's own educational background, the student's own interests, the student's own past and so on. And that's also a very provocative line of research. And once we have a tool that can do that efficiently, and I believe we're fairly close to that, then we are going to see bursting and exploding onto the educational scene these personalized AI teaching tools, which will take your child and put everything into the kinds of examples in the context that makes sense to him or her. So if your child loves NFL football, then this tool will be able to put Shakespeare – I don't mean to trivialize Shakespeare. I'm just using the simple examples here.

John Mikton: Yeah, no, no, but it makes sense. Absolutely. Right.

James Brusseau: But put Shakespeare into the kind of terms that someone who watches football and is aware of the rivalries and so on could understand. Well, it's like the Detroit Lions versus the Chicago Bears, the House of Capulets versus this other house and so on. And I believe that that play to personalization will be very, very powerful in shaping students.

I'm from St. Paul, Minnesota. And I remember that when I found out that F. Scott Fitzgerald who wrote The Great Gatsby was from Minnesota, went to my high school actually, right away, I thought, oh, I've got to read that book. And I did read it and I loved it. But I think part of the reason was because it was so personalized to my past. In only five years, I believe that all students are going to have tutors that are able to personalize almost everything through their own interests and experiences. And when that happens, judging from my analog experience, the fact that when I had a great teacher or when I had a book that connected with my life, it really took me in those directions. So if the AI can do that, then the AI will have the power through this personalization to take students, young people, part of our audience here, of course, in many different directions.

Of course, who will decide which directions they get taken in? Under what conditions? Why? What counts? In this case, old questions are going to become very important again. You know, what counts as education? What are we trying to get out of this? What counts as a child who has learned? And now the answers are basically the SAT and which college you get into. And those work fairly well, given the fact that it's an imperfect world, I mean, there are worse solutions than this.

John Mikton: And it's still working. We, many of our listeners do the IB curriculum and the kids get their scores or they do the IGCSEs out of Cambridge University or even the ASATs. So absolutely, absolutely. I mean, these systems work.

When you're talking about this, James, is that I get a sense then there are some significant risks. And what, for you, is the biggest risk? You know, because you're being very balanced, you're saying yes, this is good. This is good. And I really appreciate that because I think a balanced view is important because with the hype sometimes you think it's all doomsday and, you know, the Terminator is going to take over here next week. I mean, I'm exaggerating, but it's that sense that it's hard to be grounded. But you, who really are understanding this and working with creating this professor that's very much like you and studying AI ethics, what do you see as the biggest risk that, you know, when you go to bed at night, you're thinking, shoot?

James Brusseau: Right, I worry about my kids, right? What do I worry about for my children, is it? Yeah. I would say there are two things I worry about for my children. The one is the monoculture. That leads back to recommenders. Recommenders are very effective in bringing people or filtering to the surface the kinds of music or movies or careers that are most popular and making those things attractive to people. Now, we could enter into psychology here and have a long discussion, but to the extent that we are all relying upon the same recommendation algorithms or the same family of recommenders to decide what jobs we want, what movies we watch, who we will meet and marry, it seems to me that one tremendous risk is that we lose fundamental diversity. The idea that different people might want different jobs, the idea that for me, the woman I think is beautiful is different than the woman that you might think is beautiful. This kind of basic human diversity we have every day, I think is under deep threat by the power of recommenders which will funnel us toward a common single understanding of what counts as a good movie, what counts as a good job and so on. So I think that's one side, that's the recommender problem.

The other side, the other big problem I think is sometimes called the dignity problem. And this is a problem that is difficult to talk about because it's quite ineffable. It is that in some sense, our interaction with AI to the extent we are going to be learning with artificial intelligence and learning from it, our interactions with these tools cheapen or they counterfeit or they collapse the value of human discussion. So just think about the way that we talk with AI. At least I am, I'm very careless. I will just say, ChatGPT, give me an academic citation of X. I'm not polite, I'm impatient. In some sense, I lose a kind of human skill in discussing with ChatGPT. And, of course, we know that ChatGPT is quite sycophantic, so I don't get any pushback. I don't get any sense of resistance, whereas from humans, I could tell you, I get pushback all the time. And so I think that engagement with pushback and the idea that language has a kind of traction, that it does things for you but also works against you. And the idea that human language, this is the hardest part to prove, the idea that human language has intrinsic value as opposed to machine language. I think that that human dignity side is the second big risk that in a sense, instead of the machines helping us communicate, we will start to communicate like the machines.

Just a simple, silly example of this is more and more people are using, and I'm using like emojis and mindless ways of communicating with each other. It's almost like the corruption of human language, the reduction of human language to banal symbols. And I think that there is potential there for AI to accelerate that loss. Then the problem is to conclude this, I say that loss, then you have to ask, well, Brusseau, what is this loss exactly that we're losing? The problem is it's very difficult to define exactly what is this human quality of language and interaction, which is different from our interactions with machines that can be programmed. But I think that there is a quality there, and that quality would be called human dignity, taken from Kant's work on human dignity. And so yes, to me those are the two real risks, the monoculture risk from recommendation algorithms and the computerization of human interaction, the loss of dignity risk from just computers that are very efficient, natural language models which are very efficient in talking with us in sycophantic ways.

John Mikton: James, when you talk about this idea of the monoculture I think it's really important because we have a very multicultural, we have listeners all over the world from all kinds of backgrounds, beliefs and life experiences. But if you look at the large AI companies the ones that are kind of running the show, you know, there are five of them, I think they always say. Amy Webb, she says the nine big companies are kind of five American companies and four Chinese companies that dominate that space in the sense they have the resources, the money, the financial backing, et cetera. Sandra, do you think that also it being kind of centric to a Chinese corporate vision and an American corporate vision only exacerbates that idea of the dilution of multiculturalism in opinions?

James Brusseau: I think that's clearly the case. I always want to rush on this kind of subject to insist that it's not that these companies—Chinese—are evil. And this is not their intention, necessarily, to create a monoculture. It seems rather to be simply an effect of the world as they found it, that the training data is overwhelmingly in the English language and produced through that kind of American slash European Western democratic... Now, I know, of course, I said that the Chinese model, but I think even more that Western democratic liberal style, which I advocate for. But certainly it's quite different from the cultural traditions, beliefs, aspirations, fears and orientations, which you would find in many African or Asian and other localities. So on the one hand, I always want to say that I don't think these companies are sort of evil like comic book strip evil characters trying to do this. However, the reality, regardless of what their intentions are, because of the training data, because of the world as they found it, absolutely it's a threat.

And it's not clear how we can respond to that because the only way, at least it seems to me the only way to respond productively is by training tools to preserve or to advocate for, to work within different cultures. But that takes us back to the monoculture problem. If everything that's cool, if everything that's great, if everything that's happening is Hollywood and ChatGPT and English language, and I see this myself because I teach also in Italy at the University of Trento—it's extremely difficult to get an 18 year old to say, look, let's really treasure our local culture when you have all this excitement happening somewhere else. So here again, we talked a little bit earlier about it's kind of humans against themselves. Here again, it's kind of that basic human drive to go to whatever everyone else is going to, and follow the exciting lines. Young people, I like that especially. I was like that.

John Mikton: Yeah, and that's going to cause a problem. So when you think of your students, your undergraduates, and many of our audience are teaching kids as young as three up to 18. So how are we, you know, what are some things that you highlight about making sure we preserve our thinking, that human dignity, that human something, you know, that magic sauce that we bring? What are you, you know, you're obviously working with students and you're trying to scaffold some strategies. What would be some things you would say to educators or even schools regarding that? You know, to keep that special sauce that is human dignity and human intelligence.

James Brusseau: Right. So the term I use with the tool we are making is Cascading Fidelity. Cascading fidelity means that the tool that represents me as a professor carries with it, and we talked about this briefly earlier, carries with it some signal as to whether or not the words that the tool is producing, it is producing with high confidence that those words are the words I would say. So if the tool is textual, it might print out in black words that the tool was certain that I myself would say. It might print out in yellow words the tool thought I might or might not say. It might print out in some other color words which it wasn't so sure about. And the same kind of change could be made with a verbal tool. That is, we could train a tool to have a synthetic voice, which is just like mine. And then in cases of high confidence, there was high fidelity with my words, the machine's words, it could use my voice and then in lower confidence thresholds that voice could change. So I think something like that, always allowing human beings to understand the degree to which the computer is being true to the person that it's imitating, for example to me, that is, from what I'm looking at now, the best way to ensure continued human interaction with the tool. That is, the human is constantly asking, well, is this tool really imitating with high fidelity what it claims to be imitating? And then I think that will be important across a range of things.

So just to put that into a slightly different context, I was at a museum last summer, and it had a good little display. It had a gladiator who was a chatbot, and it was answering student questions there in the museum, just like the Roman gladiator would have. I thought that was very enlightening for the students. But I thought it would have been more enlightening had the gladiator occasionally broken from character and said, well, remember, I'm just an AI. And about this subject matter, historians disagree. Some historians think that the gladiator did X. Other historians think the gladiator did Y. Whereas by contrast, what this tool did in the museum is it just answered everything at 100%. It always was fully in character. And even if the historians are uncertain about what the truth was, the machine just stated it as fact. I thought that—it was my opinion that that was a mistaken way of using AI and that this idea of cascading fidelity is better. Then when we're really confident, we let the AI impersonate the gladiator, but when it's not certain then it should break character and show clearly that it is AI.

John Mikton: I think that's one way of responding. That is really challenging sometimes when you get, you know, a group of kids, especially in primary school, where many of them—you know, I've noticed, I teach three-year-olds—many of them have had a year of Alexa, you know, or the iPad. And so when they refer to it, they refer to it as very sure, reliable. It's always there, it actually never gets angry at them, it answers any question they ask, and they're just thinking this is amazing. Because sometimes when I ask mom and dad, who've just come back from a long day at work, stuck in traffic, and now have to cook dinner and read stories and give baths, that kind of effusiveness might be tempered. So I just think your point's a very important one, and I think that's so important for educators, in that lens, thinking of that cascading fidelity.

We're going we're coming to the end but I wanted to ask you this question it would you know and thank you because we actually talked about the questions before because I thought this was really important is if AI's influence decisions harm someone whose fault is it no yes right so this we could do a doctoral thesis on that I know but I just thought I can't leave this podcast without you at least addressing that

James Brusseau: And so the a very good way to put this into context I think is that it is the driverless car so a driverless car like let's say a Tesla is driving down the road it is a kind of a black box in the sense that we don't know why it's making the decision it makes. And if the driverless car runs into someone, then it's very difficult to know who to blame for that. Do we blame the person who designed the software? Do we blame the person who chose the training data to make the software? Do we blame the car maker? And there are other possibilities as well for who could be blamed for this Tesla driving into someone Now the Tesla has driven in has killed people and the response that Tesla has is interesting what they say is they don't even enter into the debate about who what or what might be at fault what they say is that the Tesla car on autopilot is safer than human drivers. And I suspect they're right about that. I suspect that, you know, human drivers sometimes drink or sometimes fall asleep or all kinds of stuff that probably the road would be safer if it was all just, in many places, if it was all just Tesla cars.

So one serious question that is arising there is, would we be willing as a culture to say, look, I am going to accept that a certain amount of people die in traffic accidents, and there is going to be no one to blame. There's no accountability at all for that. But in exchange, in exchange for accepting that lack of accountability, I get less people overall who will die. So on the one hand you have sort of across the culture, you have greater happiness because fewer people die. But amongst those who sadly are lethally injured by an automobile, and their family and those who are close to them, the suffering is much greater because not only do they die, but there's no way to hold anyone accountable for this. There's no way to gain any kind of sense of justice in someone being punished because the car did something it should not have done. So I think that this dilemma is a new dilemma and it's going to be tremendously important as time goes forward. Because it is true there's a performance trade-off. The AI that we cannot understand exactly what it is relying on when it makes its decisions does perform better consistently. So for the driverless car, that might mean coming to a decision amongst ourselves and say, look, if someone dies in a car accident, it's no one's fault. It just happened. We can't blame anybody. But in exchange for that, we have fewer people dying in a car accident. Now, some people will disagree with that.

John Mikton: I myself you say I was even handed before I myself don't have it I'm not sure what I really don't know I mean that's a great provocation I mean it's a bit like you know if you play russian roulette and you know you shoot yourself is it the gun maker is it the bullet yeah no I think that and I think these kind of questions are ones that sometimes get we're not spending as much time on and maybe I'm making an assumption here because we're so so busy getting to use the tool and understand the tool, and now there's agentic. It's kind of, it's a bit like you go into the shop. You know, every time you turn around there is the new version. It's a bit like, you know, buying a computer. You know, oh, you need to get the latest version because it's better than the previous version and it just feels like we're always in catch up. But your question is super challenging. I'm gonna leave it at there and I'm gonna let people kind of work out that question because I think it's a great way to stop.

James I want to thank you very much for coming on I want to remind our audience in the show notes you can actually see some of James's writing there are various articles he's an author you can also go to the University there is information about him and then he's got Phil people org he also has some writings there but definitely do spend some time on the show notes because there's some fantastic places to explore this further

And James as we wrap up is there just usually when you know finish a course or it's the end in graduation do you leave or is there some little nugget of advice you get ask people on an ethical level just to reflect upon as they walk away?

James Brusseau: I normally stop just where we stopped before with a question like the one when we gave. I think I will leave it there.

John Mikton: Yeah, that's a good stopping point. It's excellent and I want to thank our audience again more people joining our LinkedIn for the comments and the sharing and also people reaching out saying they have some interesting things to share and always LinkedIn, the International Schools Podcast. LinkedIn is the best place to communicate with us. And James, thank you again so much for your time, your insights and the provocations that I think are going to leave us all thinking here for a few more days. And we really appreciate you taking the time from your work at Pace University and your own work as an officer. So James, thank you again for being on the International Schools Podcast.

James Brusseau: Terrific, happy to be here.