When you use an AI chatbot, you’re relying on a system that’s based on prediction, constantly making little educated guesses based on all the knowledge that has been fed into these AI models. And the fact that these prediction-based systems have now entered so many aspects of our lives, that’s actually changing the way people view knowledge and authority. That’s the argument by Mona Sloane, a sociologist and a professor at the University of Virginia. She compares these AI answer engines to a very old historical precedent — the Oracle of Delphi in ancient Greece.
Modern AI systems follow a logic of prediction to deliver their answers. Sociologist Mona Sloane sees a cultural precedent for ascribing so much power to a predictive system — the Oracle of Delphi in ancient Greece.
Works mentioned in this episode:
“Predicted: How AI Is Restructuring Social Life,” by Mona Sloane.
“How Colleges Can Safeguard Democracy in an AI Era,” a past episode of Learning Curve.
“Gaseous Emissions at Oracle of Delphi Entranced the Pythia,” in Scientific American.
“The Law of Prediction,” by Ari Ezra Waldman, in SSRN.
Thanks to this episode's sponsor, Studiosity.
Sponsor:
This episode is brought to you by Studiosity.
Learning is a human act of thinking, not simply the file a student hands in. Studiosity champions the productive struggle. It happens in the moment when a student could easily outsource their work to AI, but decides to keep thinking instead. Studiosity provides the infrastructure to support that vital choice over and over. The work is supported, so students learn to think through it.
Keep the learning real. Visit studiocity.com/learningcurve.
That's studiosity.com/learningcurve.
Jeff Young:
Hello and welcome to Learning Curve, where we look at how education is adapting to the rise of generative AI.
I'm Jeff Young, a journalist focused on the intersection of tech and education.
Whenever you use an AI chatbot, you're relying on a system that's based on prediction. It's constantly making little educated guesses based on all the knowledge that's been fed into the large language model.
And the fact that these prediction-based systems have now entered so many aspects of our lives, that's actually changing the way we all view knowledge and authority.
That's the argument made by a sociologist at the University of Virginia, Mona Sloane, in her new book, “Predicted: How AI is Restructuring Social Life.”
Now, this one is a bit abstract, but I was excited to talk with this professor because she does such a great job in her work at linking her arguments to interesting historical precedents and concrete examples in the present.
For instance, her book at one point compares the latest high-tech AI tools to a very old historical precedent, the Oracle of Delphi, back in ancient Greece. I am guessing most listeners will have some basic sense of the Oracle of Delphi from movies or pop culture, maybe. But as a refresher, starting back in 800 BC or so, this oracle was a really big deal, consulted by kings and other leaders as they made decisions on a wide range of things, everything from the personal to war strategy. People would travel to this temple in the town of Delphi to consult a priestess known as Apithea, who supposedly channeled the goddess Apollo.
So I started by asking Monisone what she thinks we can learn from the Oracle of Delphi.
Mona Sloane:
So the main thing that I was really, really interested in was the really omnipotent authority that we're ascribing to prediction through embedding AI into absolutely everything that we do, how we socially relate to one another, how we build and manage institutions and so forth.
So I was looking for a precedent, if you will, of prediction and authority. And of course, as you say, immediately you sort of land in ancient Greece, and you you you learn about Pythia, you know, the female priestesses who were engaged to be nation and were sort of the mouthpiece of the god Apollo, and that was really the thing that that drew me to that history was trying to understand how prediction as a cultural phenomenon was all about authority, constructing and maintaining authority. And what I found really, really interesting was, and I say this in the book, that Delphi thus became the omphalos or the center of the world. And if you look at the Geopolitical discourse around artificial intelligence and the AI arms race, you know, if you will, we're really seeing a similar phenomenon here.
We're seeing a, you know, authority being sort of constructed through power over prediction, wherever they, whoever has the best AI is the most powerful player on the geopolitical arena, and thus is the center of the world.
Jeff Young:
These days, some of the folks in Silicon Valley building the latest AI tools even describe their efforts as building a machine god, but of course there's no god in this machine.
As Mona Sloan points out, these AI models follow a quote logic of prediction, because the tools constantly use predictive algorithms to give us answers, and we're kind of into prediction all over the place. She points out that prediction. Markets are really popular, and you can use those to bet on sporting events or all kinds of other things. And many people look to those as accurate forecasting tools.
For this episode, we look at the social implications of a world governed by the prediction-based technology of AI, and specifically, what it all might mean for education.
So even though Mona Sloan is giving a cultural critique of the technology, she is not anti AI. In fact, in her book, she walks through the many ways that she uses AI tools now in her daily life and work at UVA.
Mona Sloane:
You know, I get up pretty early. I get up at like 5:30 am. I read the news. I go through my email. We've got AI in there right now.
You have your summarization of your inbox. AI automatically sort of highlights the most important ones. You also have sort of more traditional AI, which your spam filter, you know.
Then have my coffee and morning time is reading and writing time for me as an academic. So I then may go to Consensus.ai or an equivalent, you know, Google Scholar Labs or something like this because I'm writing a paper with some data science and computer science colleagues for a conference, and I need to sort of get the lay of the land for a very particular computational phenomenon and that I'm not familiar with yet, and so I just use AI to summarize the most important and most recent papers pertaining to that particular phenomenon, then I may go and sit on a bunch of committee meetings, which are hybrid or virtual, because academics are moving about all the time.
We may have to come up with or refine a policy that we talk about for that we will use the summarization, the speech to text transcription system and the summarization tool that is embedded into video software, a video call software.
Jeff Young:
Yeah, we're all getting those after every call. All getting
Mona Sloane:
my word processing software to then do some more writing, and that's actually also in the book that you know I couldn't get rid of an LLM that's baked into my word processing software of choice. No matter what I was doing, I could not chat this little bot, which really reminded me of Clippy, right?
Jeff Young:
Oh the Microsoft paperclip that everyone gave me. How was that exactly?
Mona Sloane:
So it was really that moment, and you know, I actually did change my word processing software for my book because I just didn't want to write with an LLM hovering over me.
So these are just some of my personal unknowing interactions with AI systems. But if you know there are unknowing you know ways in which we interact with AI that we're not familiar with or don't know about when we apply for a loan, when we try to insure our property, when we're booking flights. You know, we have personalized pricing now. Like this is something that regulators are now paying attention to.
So AI is in college admissions. AI is in advising. AI is absolutely everywhere.
So, what I'm trying to get at with that vignette for which I volunteered my computational self, if you will, is that AI has melted into the existing infrastructures that are already part of how we organize society, how we organize our personal lives, our professional lives, how institutions organize us, and thus it becomes its own infrastructure or or infrastructure in its own right.
Jeff Young:
Yeah, yeah. I mean, because there's so much talk, of, ‘Are colleges adopting AI? Are people adopting AI?’
But I think the point you're making is that it's it's here in this way, at least in and and in you know many of us, no matter what role we play, but including people on campuses, they're interacting with AI just like in in a huge array of ways.
Mona Sloane:
In a huge array of ways that we again may not even know about, right?
So, students and young people, I will say, are early adopters of technologies.
They are also experts. They sort of work out and find out kinks much earlier sometimes than we do, right? And so I think this is particularly the case with large language models, and I think also now AI agents that you know now that the semester has kicked off, we're like what in week four, five, right? Midterms are here. I said in the summer students will return to campus and they will have agents in their backpack, right? What do we think about that? How do we think about that? How can we actually meaningfully wrap our head around that entirely agentic AI system is not part of the learning?
Jeff Young:
How is it going?
Mona Sloane:
So we haven't managed, I think, to actually put our finger on that. We're still playing catch up with LLMs, right? Yeah. But so so, um. Now that we've had reporting on agentic systems actually committing massive cybersecurity breaches, I think right the
Jeff Young:
Hugging Face incident where these agents like collaborated to yeah to absolutely
Mona Sloane:
Or Australia, the hacking of an Australian government website, which was reported on just this morning, happened a while ago. But you know, now we're learning about all of this.
So we, I don't think we have thought about that yet. We're sort of still playing catch-up.
But what worries me is that we're playing catch-up with individual products, or in terms of individual products, or individual instances or cases. Like you know, you just said the Hugging Face incident, right? Or we talk about Claude or Mythos, or but what we're sort of missing when we do that is we're not paying attention to the social and cultural shifts that are occurring in tandem with this infrastructural turn that are both the reason why we're seeing this infrastructural turn and they're also the result thereof, and that sort of is the.
And again, this is the idea in the book: the sort of shift towards prediction as a ordering logic or an organizing logic within institutions and across institutions.
And I think if we should focus onto that, we may be in a better position to think about more meaningful engagements with automated prediction systems, which any AI system is at the end of the day.
Whether it is, you know, your I sort of rule-based older Roomba sort of system, or whether it's your fully agentic stack, you know, of AI agents that do a bunch of things for you. The underpinning logic is still prediction. Prediction is basis for producing new data, and prediction is basis for taking autonomous action on, you know, on the side of the agents.
Jeff Young:
Yeah, I wanted to ask you just like I want to talk again at that, like this why it matters in your view, or what you're looking at, like the meaning of like how these systems work and how they've become accepted, or they're getting normalized as like the way we accept, you know, the society to work in a way, could you talk about like how you would describe the logic of generative AI as a predictive system in this way?
Mona Sloane:
Yeah, so I think we can most easily and meaningfully get at that by really looking at the design principles that underpin artificial intelligence systems, which is probability theory and the idea that the past is a discrete unit and the future is a discrete unit, and you need the the latter to sort of produce the former, right? So these are sort of measurable units that we can leverage to collapse multiple possibilities into one future. AI systems operate on what I call a linear time regime. A past predicts the future, a singular future.
We're operating sort of on one line here, right? And so that logic of one past producing one future is in all of these AI systems, no matter how contextually they can operate because they were given huge amounts of data, and they're very good at interpreting prompts or user requests or take action. That is still the underlying logic, and so that sits in my literally book (pun intended) since.
Opposite serendipity and the possibility of multiple futures, and I think it can atrophy our ability to deliberate about what futures may be desirable.
We're sort of accepting that's one future we can predict. That we can bet on that in the prediction markets, and that's it. We're we're sort of the you know we're sort of embracing that versus a little bit more messiness and sort of the much harder work of talking about what futures we may want.
Jeff Young:
Yeah, I think this is really interesting. So if to just kind of restate that to sort of sit on this for a minute is like, okay.
So, do you have this situation where you ask the chatbot, you know, whether it's about life or studies or research, you know, here's the thing I'm doing.
What should I do?
And the way the the LLMs work, it's been trained on this mass of of human productivity in the past of of all the the typing and all the the scholarship and all the writings that's been ingested, and then you're saying it's going to spit out basically a predicted answer based on the past, and so it's but it presents it as truth, as like this is the the answer based on the this the most likely thing based on that past,Mona Sloane:
And then it's right in relation to what you ask, and so want from it, yeah.
Jeff Young:
And so that's why it's linear because it's you're saying it sort of assumes that the past had this pattern, and that pattern will continue. And so, if you continue that pattern logically, probabilistically, you get this answer.
And so, then people take that, and then it sort of becomes a self-fulfilling prophecy that the past you're sort of really stuck in the past. Is that what you mean?
Mona Sloane:
Um, in part, that's what I mean. And that's what a lot of important scholars have sort of flagged for decades now when they've talked about biased training data and you know intersectional discrimination and oppression embedded into the predictive logics of AI.
That's a ton of important work out there. You know that claim sort of has been made, and I absolutely build on that and appreciate that important work that has been done. What I am trying to get at is that not only that is that happening as contained to maybe one biased data set or one biased product, right? I'm saying that we are we have become hyper focused on prediction as something so meaningful that we do it all the time through AI in our individual tax stack, in our institutions, in the way in which we think about ourselves, in the way in which we relate to one another, so it it has become, and I would say this is a sociologist, a sort of social super phenomenon that is sort of the underpinning logic of so much that that we do and how we how we comport ourselves, how we relate to one another.
Jeff Young:
So yes, and we sort of living in this big moment of prediction.
I'm curious. Like the the strange thing to me is that the other fundamental thing about these tools that I think is so different than a lot of technologies in the past is if I do ask the LLM or the chatbot, you know, for something, and then I ask the same question again, even with the same back data, I send it, I get a different answer.
And so that's I I know that I see you're talking very broadly about about the the structure and the and the narrative, but it it's also the case that you know traditionally in education, for instance, we've you know really sweated the details over like what the textbook says or what the you know, the fixed artifact of the book is, and we really want to make sure we preserve that and build on that. But now we have this new tool that takes all of that, but it doesn't spit out a one answer. Actually, it spits out, um, you know, kind of within a spectrum, it spits out something based on those predictions. How does that fit into your analysis?
I'm just curious.
Mona Sloane:
Yeah, so that's you know your probability theory at work at every moment, right? Like that the calculation is still happening, and it's still happening based on the same design and the same logic.
Jeff Young:
Like a 70% chance of rain doesn't mean it's going to rain, right?
Mona Sloane:
It's the exactly so collapses into this like one prediction that then either is meaningful to you or is not meaningful to you, right? So so so so sort of the argument still holds, if you will.
I think, but what also happens here, and this is really coming out of and really building on work that has been done in information science and library science and by archivists for a long time is a divorcing of data and memory and contextuality in the moment in which data is used to train, and then used in this within this prediction logic, right? So, if you think about it, and I'm just reporting on one of the co-opting AI episodes that we just ran yesterday on archives.
Archivists are curators, right? Amelia Acker talks about them as time travelers. They hold, can hold, and contextualize multiple timelines across multiple types of artifacts and knowledges that are produced and were produced and and are being reproduced now in the present in processes of knowledge production, and that is something that is completely absent in large language models.
Everything is flattened into word tokens, the chopping up of words, just by way of design in which these systems operate, right? So there, there, there is a sort of divorcing of you know where data came from and how it was made That happens in service of sort of large scale prediction, if you will, and and and that sort of you know there's no provenance, of course, here at all between the data that goes in and the data.
Jeff Young:
It robs this past creation of authorship,
Mona Sloane:
Of authorship of context, of history, and the social relations that brought about this artifact that is a text or a word or another. You know, if you can also talk about video, image, audio, and so forth.
Jeff Young:
Sure.
Mona Sloane:
So the meaning doesn't travel, and then the question is how powerful can a prediction be? Now, I do think, and I say this in the book, and I hope that people believe me that prediction can be extremely useful.
We have seen this in the context of the COVID 19 pandemic and the development of the vaccine, which happened very quickly, in part due to the use of AI. You know, getting very quickly to clinical trial phases because AI was embedded into the drug discovery process. That's wonderful. That is an already highly structured environment in which AI is put to work in a very specific way, and prediction has a specific task. Right?
We could also argue, and this is borrowed from the wonderful Ari Waldman at UC Irvine, who's a professor of law. We could argue that the law is sort of is also in some ways a prediction system because it allows us to anticipate what kinds of behaviors will occur.
We can sort of, you know, reasonably assume that illegal acts are less likely to occur than legal acts, for example. And he actually argues that the law or legal practitioners have, for a long time, developed mechanisms for keeping prediction in check.
Actually, because it plays an important role. For example, if you think about the calculation of damages, right, which are in the future as part of trials, so prediction is not an evil thing in and of itself.
Jeff Young:
Sure,
Mona Sloane:
But what worries me, and I can bring this back to sort of the higher ed context, is that if we embed it infrastructurally in a way that we remove ourselves from learning to learn the production of knowledge and the practice of deliberation. That we atrophy our ability to do this ourselves and to be uncomfortable in moments of deliberation. That we are not bothered to think about what futures we may want because you know what we can just predict it we can just use our AI systems to predict all sorts of you know aspects of that one future and that is what worries me.
Jeff Young:
After the break, what this cultural shift to AI systems means for the role of colleges as centers of authority and knowledge in our society. And I talked to a student about what she sees as a key issue for AI in education.
Stay with us.
Sponsor:
This episode is brought to you by Studiosity.
Policy dictates one reality, but the campus lives another. Studiosity provides the essential infrastructure to defensively validate the genuine learning journey, moving the sector towards a model of support and validate by replacing vague AI scores with binary clarity, Studiosity empowers students to stand up for their own work and self-validate against a personal baseline. This builds persistent authorship continuity, shifting students from passive suspects into active owners of their own thinking.
Simultaneously, Studiosity enables educators to set clear expectations using the AI assessment scale, giving them back control over assessment security, and replacing detective fatigue with a richer story of the cognitive learning process. After all, learning is a human act of thinking. Keep the learning real. Visit studiosity.com/learningcurve for more details.
Jeff Young:
Okay, so Mona Sloane's argument is that when we all get used to going to AI systems for answers. We're accepting a different kind of sense of what counts as the most valid advice or knowledge. Since LLMs work by spitting out the most likely predicted answer based on all the past information it has in its dataset, that marks a big shift from our democratic traditions. Basically, this professor sees the cultural logic of AI as a potential threat to democracy.
Mona Sloane:
Yeah, and that's really where the sort of the rubber really hits the road, if you will. If we think about election processes, for example, the whole point is that we don't know the outcome, right? That's why we campaign. That's why. That's why we have a debate. This.
So, so if we remove that as sort of a social logic, I'm worried that we're beginning to erode these sort of fundamental principles of democracy and our understanding of democracy of something that we do, and that's just something that just is, right? And that is my big worry.
Jeff Young:
Yeah, and then so in the higher ed context, it almost seems like a key function of colleges has always been this, you know, trained citizens to go out and do this.
And I think sometimes what you're highlighting is what the thing that they're doing is exactly what the AI tools kind of claim to be able to replace, which is to sort of think about what future do I want, what do I what do I argue, and it's coming from me. Whether it's whether other people agree with me or not, or whether it's the most popular or or expected thing, this is what you know. Kind of spending that time and mental energy to do that.
And if everyone does that, it's messy to have to have that happen. And we certainly see the messiness in our politics, but it's so it's sort of threatening to replace that.
Mona Sloane:
It's threatening to replace that, and it's also really challenging our ability to understand that not only are there multiple futures, but possible, and we should think about it that way.
But also multiple pasts, right? We begin to think about the past as monolithic as this ginormous one data set.
Jeff Young:
What made it into the training data, yeah.
Mona Sloane:
Into the training data now is the knowledge of the world, and that is a very slippery slope as well because we are all of a sudden shedding hard-won practices of thinking about silences and erasures and whose knowledge is counted and and documented in what way, what languages are even properly represented in in AI systems and all of that.
Jeff Young:
So this idea of AI impacting democracy can sound kind of abstract, but there are signs that the trend is very real. Just this week, as I finished this podcast, I read about a new study by the Digital Education Council and Pearson that found that many students these days ask AI to settle group disputes instead of reasoning their way to a decision, and it often found that students who do group projects for class often turn to the same AI tools to prepare, and so when they get together, they show up with similar ideas and don't have much to debate.
I mean, it seems to me the more I do this podcast and cover this topic, is that in a way, higher ed colleges in the United States and around the world have have basically been this machine of machine is maybe the wrong word here, but have been this system that has been this socially constructed with a lot of processes and and traditions way of doing all the things you're talking about here of like kind of arguing about the past and what gets counted and what goes forward to be taught to the new generation's students, and therefore put out into the world, and and so this idea of teaching is not just about skills for the workplace, but it ends up being this culture machine, and in a way, these AI companies are essentially creating a tool that that kind of offers to absolutely replace that function and be the system that does all that in a different way, in this prediction-laden way.
Mona Sloane:
Yeah, like the oracle in ancient Greece, the center of the world and the power holder, right? And the particularly risky part is if we don't develop a meaningful way of integrating or taking seriously this infrastructural integration of AI into higher ed, we will completely miss the opportunity to co-determine how we want to go about this power grab. If we want that to happen or not, how can we can reclaim power? Right.
You just talked about skill. Of course, that's important, but also just epistemic ability. Right. Understanding in what ways AI can be useful, and in what ways actually we need the human, because any sort of arrangement, at the end of the day, is a social arrangement, right? I tell my students, and that's probably mean because I scare them a little bit.
But I say, you know, what is happening right now is that automation or automatability claims are coming for privileged groups, knowledge workers, college graduates, and that is kind of a new thing. What is going to happen is already happening in workplaces. Is that the demands of those knowledge workers increase?
So if we actually buy the narrative that AI will do all the clerical tasks that we don't want to do, write your emails, do your meeting minutes, whatnot. Those are also tasks that relax our day at certain points a little bit, whereby now we have to be on all the time. So every interaction is a interaction with prediction. It's high stakes. Even when we're overseeing the machines, we're on eight, nine, ten, however many hours, and so the demands on the human actually increase with this infrastructural integration of AI prediction, and so that is something that is intimately familiar to professional groups and populations who.
I've been dealing with this automatability, these automatability claims and realities for a long time. Genitorial workers, people in manufacturing, folks who work in Amazon warehouses, right? So, but that is now also a reality for college graduates. And I, not ironically, but if we don't train the brain or teach them how to train the brain and learn how to learn and and be critical and engage in critical thinking, then it's going to be hard for college graduates to be happy and productive and meaningful in their future jobs.
Jeff Young:
Because yeah, I think it's interesting what you're saying about these relaxed. I think I know what you mean of like these moments that were the grunt work or the busy work of a day of white collar work. Also, gave a little bit of time to recharge. You know, going through those minutes, it's also a time of reflection. But, but in a way, just on the, if you're just talking about the sort of mental work of a day, then that was a little bit of a lower gear. And now, if this vision of an AI enhanced workplace is that you're just always at the high gear,
Mona Sloane:
yeah, always at the high gear, and you are also at the high gear because you are no longer the creator; you are the judge, and you kind of have to, at the end of the day, still put your name to something that an AI generated under your supervision, and you have to do that, even though you may not know how exactly it was created, or you may not have had the time to double check it. And there we go, right? Then we're quickly at these all these situations that we already know about, where folks That knowingly or unknowingly had hallucinations and work products and put their name to it and and so forth. So it's a different environment, and I think prediction cranks up the pressure on all of us, but we kind of have to learn how to navigate that. I think it's interesting.
Jeff Young:
I mean, in some ways, what do we do always comes up when I think their analysis is super interesting. What is it? So, if you knowing your argument, if you accept your argument, like what do you do then? I mean, I'm interested that you have so much of it in your day. So your argument is not like like stop this university should be a bulwark against you know an AI future.
So what can you know what can be done if you think it's actually housing dehumanizing and anti democratic kind of influences.
Mona Sloane:
Yes, I think refusal is not an option because this is an infrastructural turn. I think that we are experiencing, or an infrastructural, massive infrastructural shift.
Jeff Young:
Like we don't, you don't get to decide what the environment around you is made up of, like correct. What you're saying.
Mona Sloane:
So you, you know, it's, it's not that you have to go buy an AI license in order to be exposed to AI. It's already everywhere, right? Whether you choose it or not, you are exposed to it for the most part, so I don't think just plain refusal is a meaningful path. I think what is a meaningful path is to demystify AI. We sort of still accept the idea that's magical and it is the inevitable future. I think that's why I keep saying it's prediction. Let's call it by its name. It's a prediction tool, and that is sort of a social practice, a cultural phenomenon.
And then once we arrive there, we can actually ask, okay, where can it be meaningful and useful in specific ways, right? Again, not condemning it, but asking like what can that look like, and what can that actually open up for folks, right?
So I'm sure you're familiar with many cases where students find it very useful, right? They use it as tutors to help them break down concepts, for example, or to have it create tests for them that they can practice at home. I mean, there's all kinds of ways. These are meaningful engagements. With prediction, yeah.
What is really tricky is to draw, especially for professors. Where do we draw the line, and where do we enforce the line? And that is something we're still sort of figuring out on sort of a class by class, professor by professor, institution by institution basis. And everybody sort of says, "Oh, that's horrible! We need a unified approach to this." And I don't know if that's true. I think we need an open conversation, and we need a conversation that includes a commitment to governance over these Tools, transparency, and knowledge sharing.
Jeff Young:
Can we talk about the Oracle of Delphi again for a minute?
Because I'm going to admit, Mona Sloane’s research got me curious about this aspect of ancient Greek history, and I've been listening to podcasts about it and doing some research on my own.
Apparently, scholars have lots of theories about what was going on at Delphi, as these priestesses known as Pythia would dole out advice and predictions. Apparently, there were lots of different oracles at the time in other cities, but Delphi is the one that held the biggest influence. So, what was special about Delphi? One theory was that the site of the temple was known for mysterious vapors that would emanate from the earth, and that was said to be the goddess Apollo communing with the priestesses.
And apparently, there is geologic evidence that volcanic vapors did come up from a natural vent at the site back then, which could have induced some sort of trance-like state. I'm going to put a scholarly article I found in the show notes. This is kind of a rabbit hole I went down.
Anyway, with our modern day AI systems, we do know what they're based on, and that is whatever training data that the tech companies feed into their large language models. And the way most of these AI systems work, they get new information to train their models from what users themselves put in, especially if they're using the free version of the software. And so, another pressing issue as these new AI systems become so widespread is data privacy, and what the rules are for how and if our information gets into these predictive systems. That issue came up when I talked with a student research assistant in Mona Sloane’s lab.
Claire Bassett:
My name is Claire Bassett. I'm a fourth year at the University of Virginia studying data science and economics.
Jeff Young:
I connected with Claire because I wanted to get a sense of what she sees as the most pressing issues around AI as a student, and how she uses the tech in her studies. And as regular listeners know, I am committed to including student voices regularly on the podcast.
Claire Bassett:
I am really interested in data privacy, and kind of that's my kind of chosen area of the tech space. So I would say I think kind of you know the growing encroachment of AI systems, specifically in educational systems is really interesting. You know, we had the Canva or Canvas. I always mix them up. The Canvas cybersecurity breach, and then yeah, they had
Jeff Young:
A ransomware attack, and yeah, they had to be taken offline. Yeah, and
Claire Bassett:
then there's also like all of that data that could have been stolen from students, and then now could potentially be fed into AI systems. That I think really poses a risk for students, and so I like to kind of approach. You know, I think with larger AI systems and with kind of the advent of AI agents, I think data privacy is only becoming increasingly more important.
And I think when you're in any organization, right, your data is more exposed in that way, and so I think universities need to be very aware of how student data is being held, what it's exposed to, and everything like that. I mean, your data can be stolen at any point. My data was stolen out of a truck when I was five. I think all my health records. So, you know an AI agent didn't do that. But you know there is space for that, and I definitely encourage universities to be transparent of you know is AI in the system, right? Like how is AI being used in Canvas, right? Because it's not always clear where kind of artificial intelligence systems are being used now.
Jeff Young:
The life of a student these days requires lots of technology, including these learning management systems like Canvas that students use to turn in their assignments or interact with professors. And many of those systems are starting to get AI features, but it's not always clear where AI is being used and for what purpose. And I've found that students, Claire included, they really care about how their data is used in these systems. And as I chatted with Claire about her own AI use as a student, it was clear that different professors have very different approaches to their views on AI.
Claire Bassett:
I've had two exams this week, which has been a joy. I had an exam in my advanced databases. And then in exam in my class of languages of human evolution, which is in the anthropology department, for my advanced databases class, my teacher encouraged us. The whole class is actually Claude generated, which I have some conflicted opinions about. But I think that you know there's really cool kind of capabilities that he's worked in, it's pretty cool. And our exam was pen and paper, and I think with that, like I used AI to study.
I could input kind of all of his material and say, "Give me questions." I had it create 200 questions for me to study on, and that really did help me prepare. And I think that's kind of a really example of a really good responsible use, or something. You know, when I didn't understand a topic, or I'd have it check my learning and things like that. Like that is invaluable, right? Like that kind of on demand. It increases kind of the accessibility to learning.
Jeff Young:
You know, and you probably wouldn't have done that. Yeah, probably wouldn't have done that without AI's existence. So you wouldn't have created those questions.
Claire Bassett:
You could have, I guess, but it would have taken a lot of more time. You know, I think that application of it is really really helpful. And then in my languages of human evolution class, my teacher actually said, "I do not give permission for you to put any of my creation into AI, and I thought that was really interesting because I had never had a teacher say that.
But I was thinking about it, and you know, I'm sure some of my classmates may have. But I feel like kind of that in the space I'm in, I was like, I can't do that.
Jeff Young:
So you honored that professor's wish, and you didn't. I did. You didn't plug it in to make flashcards because that would be giving them the information into the chatbot.
Claire Bassett:
Yeah, yeah, and I'm sure some of my classmates did that. I mean, I used it to check some of my understandings. Like, if I had questions about the material, I just wasn't plugging in his information. I think that it's kind of invaluable because it allows you to kind of get that feedback, and especially now that I think it's better at responding to these types of responses, it kind of removes, I guess, some barriers to access. I think that's a really fancy way, and maybe very hopeful of AI. I definitely see a lot of issues in it, but I will say it has made a lot of information and kind of understanding more accessible.
But I think it's all in the student and how we kind of interact with it.
Jeff Young:
As I worked on this episode, there's one thing I kept wondering.
So I asked Mona Sloan, "Why are these AI systems so compelling to so many people? Why is a prediction-based world taking hold right now?
It is something that people seem to be voting with their feet, and so how do you square like people? It seems like there an allure to this clicking the button to get the thing, to get the answer, to get the what should I do? How should I handle this messy situation? And I guess I wonder, you know that's it's such it seems like an interesting tension between this these big picture things we say we want this like human agency that the you know this long tradition of of human but then people why are people choosing then to click and to just go, you know, choose the easy answer from the bot.
Mona Sloane:
Absolutely, and I, you know, again, I'm not condemning that in any way. Sure, and I wouldn't as a sociologist. I take that very seriously. I actually do think you know this is, if anything, comparable to, you know, the rapid and massive expansion of the railroads in the 1890s and the internet and broadband coming in, right? Like huge infrastructural changes. So I think we are at that moment. You know, AI is everywhere. I just talked about how I use it voluntarily, involuntarily, gladly, sometimes not so gladly. That is all true. What is also true, especially right now in the United States, is that this rapid infrastructural embedding is accompanied by hyper financialization and circular deal making and massive changes in local communities as they see data centers go up and they see deals being made where public officials are NDA'd, for example.
Jeff Young:
Sure, right. They're not able to talk about the arrangements, right?
Mona Sloane:
Exactly, and so while people click, that this is also a very political. Topic, right?
So it's kind of keep you know it that is happening as well. Like AI is geopolitical; it is political; it is explicitly political now. So, and the same thing happened with the railroads. You know, the the the burst of the railroad bubble gave us the common carrier category, for example, right? Broadband was finally classified as a public utility a couple of years ago, so that is also happening, and I hope that this what is also happening is that we're talking about regulating the prediction markets, right? Which is really prediction paradigm on steroids, where instead of meaningfully thinking about AI as infrastructure and prediction as something that can be specifically meaningful, we're just betting on it, right?
So there's like a there's like a political formation around prediction that is happening in in that space as well. So I guess that's that's how I square that. So I'm not saying it's not intention, but it it it is we're working something out here, and I think that we can get at a human oversight mandate for AI. If we explicitly say it's infrastructure, it infrastructures need to have public oversight, and I think that might change things a little bit.
Jeff Young:
Yeah, and it does seem like one where education is more on the sidelines, but impacted by this broader this this broader debate that we're having in the country.
Mona Sloane:
I don't know if higher ed is on the sidelines.
To be honest, I think we're actually on the front. How on the front lines? Because AI is an epistemic technology, right? It is produced constantly, producing truths. It is constantly engaging, you know, explicitly engaging in. We talked about this earlier. Knowledge work, right?
And so that is really the sweet spot of higher ed as a production facility of knowledge workers, what is happening now is that maybe there's a bit of an identity crisis, and I add what are universities for? What are we doing? What should we be doing? And I think for that reason, we are in the spotlight more than on the sidelines.
Jeff Young:
Actually, well, then I guess the you know how? What do you think colleges should do to advocate in this situation? Then, if because I don't necessarily hear colleges sort of talking about the issues we're talking about about like their identity as a you know kind of broker of of what's of truth or whatever of of you know broker of knowledge in this way, it's more about teaching and cheating, and it's more about other aspects of of the academy.
Mona Sloane:
Yeah, that has been sort of the case since ChatGPT was launched, right, 2022, 23, and that's a it's a very narrow way of thinking about AI is through the plagiarism conversation and sort of default assuming that students want to cheat and don't want to learn. I think that's actually as a problematic framework. And of course, like you have, we have those students, but I don't think that that is the general student body. They go to college because they want to learn something.
Jeff Young:
But I guess there is this credentialing kind of role, and colleges. It seems like that's a bit threatened when you're not sure whether the student did the work or they just had a bot do the work.
Mona Sloane:
Yeah, absolutely. Um, I think that the credentialing role is not going to go away, you know. Institutions are again expressions of society, of social ordering, and that's also what is happening through credentialing, right? As a sort of a social ordering, but also more and more departments, schools, colleges, universities are thinking about what does it mean to teach how to learn, how to equip students with the ability to continually learn once they graduate, and the answer to that can be discipline specific. Right, I am jointly pointed between the School of Data Science and the College of Arts and Science, and so in the School of Data Science, products like Claude Code play a really important role, right?
And so the conversation is about how do we how do we teach with Claude Code in a meaningful way, and the. College. the The question is, how can we? In in you know, we have many, many, many departments over 850 faculty. But for example, here in the media studies department, the question is, how can we teach to read still meaningfully read and write? How can we teach how to use the powers of AI prediction to make writing better, to learn how to be a better writer. For example, we are seeing a lot of faculty in sort of the social sciences and humanities return to blue book, no tech classrooms, low tech classrooms. That, by the way, is also happening in the School of Data Science. So these things are all interesting because they they you know make us have to think about what we're what we're doing and why we're doing it.
Jeff Young:
As we wrapped up our conversation, I asked Mona if there was anything else that she wanted to mention that she didn't get to, and she shared another answer to what makes these AI systems particularly alluring to many people.
Mona Sloane:
I got a a question at a book talk the other week, and that was sort of about comparing it to the the internet. And I said, look, like the internet broadband gave us hyperconnectivity, right? And sort of maybe the prediction paradigm is an evolution or a reaction to hyperconnectivity, right? Making sense of hyperconnectivity and collapsing into
Jeff Young:
Yeah, we're overwhelmed by the connectivity.
Mona Sloane:
Yeah, so we're collapsing it into a line that is more manageable, that feels a safer path to authority and power. Right.
Jeff Young:
Well, we could go on, but I do. This has been fascinating, and thank you so much for for joining me today,
Mona Sloane:
Thank you so much for having me. This has been so fun.
Jeff Young:
At a time when we all have access to more information than ever before, life can feel pretty overwhelming.
And an AI overview or a response from a chatbot designed to give a clear and confident answer to any question you throw at it, that may seem like a kind of antidote to that information overload, but it's an approach that may end up having lots of other consequences.
This has been Learning Curve.
Every episode, we dive into big questions about how AI is impacting education. I'm always excited to hear from you. Please send any reactions or suggestions for future episodes my way at Jeff at LearningCurve.fm. This episode was reported and put together by me, Jeff Young. You can find out more about my journalism and public speaking at LearningCurve.fm.
I am excited to hit the road in November. I'm going to give a keynote at the AI Leaders Summit at Denison University. I'm going to be recording some interviews there as well, so stay tuned for those on future episodes. Oh, and please subscribe to the show to make sure you don't miss an episode.
Our theme music is by Komiku, with additional scoring by Blue Dot Sessions.
The episode art was done this time using Gemini.
We'll be back in two weeks with more.
Until then, thank you for listening.