Boyle42 · Strategic Use Required: AI, Literacy, and the New Learning Frontier

🎧 Skip to Transcript | 📚 Show Notes | ✉️ Contact TLC

Episode Description

In this episode, Matt from the Rosemary B. Fuss TLC speaks with librarians Sophie Forrester and Stacey Wilson about artificial intelligence, information literacy, and what it means to teach students discernment in the age of autofill. Topics include:

Annotated Bibliographies and AI hallucinations (or "What happens when the bot makes stuff up?").
Strategic use cases for summaries, search scaffolds, and accessibility.
The ethics of automation, fake research, and ecological impact.
Michelle Branch, seven-fingered deepfakes, and Denny’s dislike of war metaphors.

Show Notes

Additional links and resources mentioned in (or related to) the episode:

Research Tools

Notebook LM (Google)
Looking for one of those reading companions Matt mentioned? For summarizing and engaging with sources. Try this tool out.
Canvas “Summarize” Feature Guide
Namechecked in the podcast. A brief how-to of how to make an AI summarize a Canvas Discussion.

Strategic Use Examples

Writing for Career Readiness – Sara Large
A local example of strategic AI use from our WAC Coordinator: engaging ChatGPT to develop stronger resume bullet points.
Annotated Bibliography Test with ChatGPT’s Deep Research Mode (4.23.2025)
A more sophisticated annotated bibliography than the one discussed in the episode. Matt generated it using ChatGPT's Deep Research Mode. Stronger results, but with costs: higher server load, no access beyond paywalls, and students without premium access out of luck. Also? Not carbon-neutral.

Misinformation and Limitations

"A mediocre, but not completely incompetent grad student."
Interview with mathematician Terence Tao, on the limits and possibilities of large language models.
AI Tools Fuel Rise of Fake Research Papers – Forbes
Referenced by Stacey. AI-generated junk science is making its way into Google Scholar and beyond.
Problems with Generative AI
Includes several of the articles Sophie referenced. An excellent dive into the ethics of its electricity consumption, plagiaristic impulses, privacy concerns, and overall usefulness.

Text Transcript

This transcript was generated from automatic speech recognition and lightly edited for clarity. Matt apologizes for any transcription errors.

Introduction

Matt: Hello and welcome back to this wonderfully unnamed podcast from The Rosemary B. Fuss Teaching and Learning Center. In this episode, Matt talks with Sophie Forester and Stacey Wilson about artificial intelligence, information literacy, and the tension between speed and substance.

Full disclosure, we recorded this before ChatGPT Plus's Deep Research mode and before tools like perplexity hit student workflows. So that's AI for you. It evolves faster than Matt can post a podcast episode. However, the core problem hasn't changed. When students are lost in the cognitive woods, or when they're burned out, they don't reach for the best tools. They reach for whatever's closest and whatever's fastest. And even if they have better tools, if they aren't practicing discernment, if they don't know what questions to ask, they won't get better answers. If anything, AI just makes the shortcuts feel smarter.

Let's get into it.

"What is the Use Case"?" AI and the Annotated Bibliography Hallucination

Matt: So welcome back, everybody. Today I want to talk a little bit about boundaries of AI for research. And joining me today are Stacey Wilson who is the head of research and learning here at Lasell. And we also have Sophie Forrester, who is the head of digital services at Lasell. Welcome to you both.

Sophie Forrester: Thank you.

Stacey Wilson: Thanks so much for having us.

Matt: I guess the genesis of this conversation is I was wondering what the appropriate boundaries are for using artificial intelligence to interact with academic research. And what I meant by interaction was there seem to be so many ways that people are advertising AI for this purpose these days that feel like Cliff notes to me, like, you know, here, get an easy way of getting the information that you don't have time to get otherwise. So I see these things like upload a PDF and talk to the text, anthropomorphizing it as your little buddy who has read the thing is going to explain it to you or annotation, which I find to be more useful, which is sort of like copying pasting words where you have no idea what the word means or phrases and saying, "What does this mean? Can you explain it to me? And, you know, give me an analogy or an example."

But there's also summarization, which just seems to be a default now. Canvas has this discussion summary feature that people can click and will summarize entire discussions. And Google wants to stick this. Well -- wants to? Has! -- stuck a little summary at the top, along with all the ads and all the other stuff, and it's shoved down the top results and I guess when it comes to research, I'm wondering what the appropriate boundaries are for stuff like copyright for stuff like cognition, just learning in general. And I was hoping to get your all's take on it.

Sophie: I guess the way that I would begin approaching it is not what are the boundaries, but what is the use case? You know, in what circumstance is AI superior to whatever other methods we would use? Speaking as someone who is very, very critical of AI, I think that there is one thing that I love AI for, which is customer reviews summaries. Amazon is doing this -- I think target is doing this -- where it will say "Customers appreciate the durability of this product. They say that it looks really nice. However, they are concerned about blah blah blah." And it's like fantastic because if it's something that has thousands of reviews, I no longer have to wade through ten pages of reviews saying, "Great shipping times! Thanks!" in order to get to something.

But with research, what I see people working towards or trying to do is using AI for annotated bibliographies. So I decided to do a little test run, and I put in ChatGPT, "Find me peer reviewed articles on the effect of social media on teenagers relationships," which is kind of a classic librarian search query because something that's very easy to track. ChatGPT gave me:

Social media use and relationship quality among adolescents
The impact of social media on teen relationships: A review of the literature,
Digital communication on adolescent relationships: A systematic review
Navigating friendship in the age of social media: Effects on emotional well-being
Online social networking and adolescent well-being. A review of the literature.

And then it gave, you know, a sentence-long summary of each of these. So there were two things that I immediately picked up on with this list. First thing is that there were only titles. It didn't give me journals. They didn't give me -- I say "they" -- it didn't give me journal titles, it didn't give me authors. Just article titles. The second thing is each of these titles has a very particular pattern to it. It just rephrased my query. "Social media use and relationship quality among adolescents." "The impact in social media on teen relationships." It's just occasionally using a synonym and then putting them in a different order, and then tacking on "A review of the literature," "a systematic review," like some common, meaningless academic subtitle.

So my suspicions were raised and I said, "Provide me a full citation for each of the above articles." "I don't have access to specific articles or citations directly." And then it like tried to tell me how to do a general APA citation. I'm thinking, "Okay, you haven't answered my question." So then I gave what I hoped was a simple question for it to answer. "What journals are each of those articles from?" "Since the titles I provided illustrative examples rather than actual articles, I can specify the journals they come from." This is immediately after it said. "Here are some peer reviewed articles you can look for."

So, when we talk about AI in the context of research, specifically, what we're often talking about is LLMs, large language models, which are essentially text predictors. They are essentially a souped-up version autofill on your phone so that I can tap "I had a...dream...last... night...about,..", but I thought I would give it another go and that I would try sending right up front everything I need. So I said "Provide me full citations for peer-reviewed journal articles on the sustainability of fast fashion." And it generated six citations for me.

Now, the first one is mostly correct. It's an actual article that exists, and the volume and page numbers of authors, that it says it does. Um, but the DOI that it gave me does not exist.
Now, the second one, it said "Bridges T and Heinz T (2019)," a critical article title, sustainability 11.3.8931, and the DOI. Well, that DOI actually pulled up a completely unrelated article in that issue of sustainability, and the article that it named does not exist.
The third one again gave me a non-existent DOI for a non-existent article that it said was written by Henninger C.E., Elevoza, PJ. and W.W. There is no name, just WW. And interestingly, Henninger is a real person who really does write in this field.
The fourth one again, a DOI for a unrelated article in the journal that it gave me in the year that it gave me but a different issue, and the authors it gave me were Aniato, F., Careedy, M., Ripa, L., and Moretto A., some of whom are real people who write in that field. Um, and the article in that title it gave me was "Supply Chain Management in the Fast Fashion Industry: a review of the literature," which was kind of a mash up of an article title that one of those authors really did write. And an article that cited another one on them.
The next one is another non-existent article with a DOI for an unrelated article in the same issue.
And then the final one is a book. And the book really does exist, ChatGPT said. "While this is a book, it is often cited in sustainability literature regarding fashion." It gave me the wrong year, but certainly I could've -- Well, I did find that book -- So that one's more successful.

But going back to what I said about thought, what's the use case? How far did that actually get? Because that ultimately found me one article and a book. And if I had gone into Laser Search, if I had just Googled or DuckDuckGo'ed "fast fashion and sustainability," I would have found those things just as quickly and with a significantly lighter environmental impact because AI uses I believe orders in magnitude more electricity and water than general search functions do. And obviously we live in the age of climate change. That's something that ethically we need to be conscious of. And we also have to be conscious. I believe as people who live in the United States, even though we live in a coastal state built on a harbor, that we aren't the people who are going to be bearing the brunt of that impact. And we have people talking about decolonizing AI. And I don't believe that that's ultimately possible because the AI boom is contributing heavily to global warming. Who is that going to affect? It's not going to be the first world. It's not going to be a global north. It's going to be the colonized people of South America and South Asia particularly.

Matt: You bring up excellent points because AI for an annotated bibliography, is not as useful as just sort of googling or searching for it traditionally because of hallucination, just the basis of it. It's ah, it's it cares about probability, not accuracy. And it just gives you a mishmash of real stuff and fake stuff, and stuff that is both real and fake. And it's very confusing and, uh, sometimes surprising. I've always thought of it sort of a little bit as the modern era's version of, um, text to speech errors, AI results, hallucinations. They're funny sometimes, but not so funny when you consider the things underneath them. Ecological effects and the like as well of rampant unstrategic use.

Copies of Copies? When AI Summaries Work and When They Flatten Thought

Matt: Stacey. Did you have thoughts about that particular point?

Stacey: Yeah, I mean, there's a lot to unpack there. I'm looking at the three things you proposed, um, initially in terms of boundaries for AI, and I have a lot of, um, moral and other issues with AI tools. But I think in terms of like a productive way forward. Actually, I didn't think this until I was in your class. Matt, but the way you used it for summary I found really interesting and useful, although my biggest concern there is that AI inherently it's distilling and distilling and distilling and distilling and discussion boards are inherently, you know, kind of distilled thoughts and brief impressions. And there's -- I compared it in class, like a lossy file or a photocopy of a copy. It's like every copy you make, you're losing quality, you're losing texture, you're losing nuance. You may be losing the author's initial point. Um, so assuming we're dealing with real documents, um, not hallucinated, uh, documents, presuming students do the research in traditional ways, I think there's a lot to be critical of.

And this is kind of an unformed thoughts, but I also think a benefit could be -- You know, a lot of this information is behind paywalls. You know, our students have access to a great wealth of information because they're here at Lasell. But the moment they graduate, that kind of all disappears. So AI can be -- has the potential to be a tool -- to kind of break down some of those barriers to entry, um, to research for the general population, which is something I think a lot about. I'm concerned with summary and annotation, um, just because of hallucinations, and my whole job is instructing students in information literacy and getting them to the point where they can evaluate information and be independent thinkers and critical thinkers about what they consume in the world. And I think until they hone those skills to a degree that is expert level, The danger of encouraging AI in the classroom in this way is that the nuance isn't there. They may not be able to recognize errors or, um, and I'm sure we'll talk later about problematic language and things like that. Um, stereotypes and other things that AI is prone to.

So I would, I mean, I agree with a lot of what Sophie said in terms of like the problems there because there are numerous problems. But in terms of thinking about a way to use it, those are kind of my general thoughts.

Sophie: It seems like what we're both getting at is that the use case is where, um, ironically, I can't remember...

Stacey: Well, longer thinking like a really good use case. And I'll give credit to Dr. Sara Large for this one is developing a list of keywords for each topic synonyms and things, um, for students to search in databases. It's a great tool if you're not familiar at all with a topic, to kind of get oriented, to see what the conversations are in an area that you're completely unfamiliar with. Although, again, like with caveats with, you know, being extra vigilant for red flags and hallucinations,

Matt: That's very interesting because it reminds me of what Sophie said at the very beginning, using it to sort of summarize 50,000 Amazon reviews that you don't want to sift through, because some of them will be fake, some of them will be, uh, mendacious attempts to get gift cards. And some of will just be, you know, like "The shipping error!" or like, "oh, this chair broke when I stood on it!" Like, "yeah, of course, it did you idiot. What are you talking about?" So there's a whole lot of irrelevancies that it can smooth out. And the sort of fear of smoothing out the good stuff with the irrelevancy is, is what I hear coming out. And if you had a point you wanted to bring up, Sophie. I had a question I wanted to ask you to.

Sophie: And that leads beautifully into what I was thinking, which is, I think what we're all kind of working towards is that the use case is where brevity and conciseness is more valuable than richness and detail, and where it is so much more valuable that it becomes worth the cost. Because I don't love, you know, from an ethical perspective, I don't like the notion that of black and white thinking. I think it's more useful to think in terms, uh, "okay. Let's be clear eyed about -- this is using a lot of water. This is using a lot of electricity. There are significant concerns of plagiarism. There are significant concerns of economic disruption to people's livelihoods. At what point kind of do the scales tip in favor of this one individual use?" Because we can't control AI. That's a much bigger thing than me sitting at my computer. It's even a much bigger thing than all of us at Lasell sitting in front of our computer, it's a much bigger thing than Everybody in Massachusetts sitting in front of their computers. What I have control over is, you know, in this specific instance, me on this day, at this time, sitting in front of my computer, what tool do I use?

For the admittedly minuscule impacts that it will have financially, economically, environmentally. You know, is that borne out by benefits relative to another method? And in something like summarizing customer reviews? Yeah. Because then you think, okay, well, how much electricity would it use for me to scroll through five pages of reviews? You know, and how much time will that take?

Matt: It's interesting because this whole cost benefit analysis of how much am I harming the environment and how much of my harming my fellow person is behind a lot of the strategic use you're advocating for. Use it for something that is small and that does lead it into, um, another question I have, because I agree with you that if you use it in prompt engineering, basically, if you use it in the way that you described earlier to gather up Peer reviewed articles for an annotated bibliography, You'll get that weird mishmash of "Here is stuff that I know is part of a soup of what you're asking for," but it is clearly not what you're asking for, although some of it may be useful, who the Hell knows? But obviously you can use a different thing and get better results.

Cannons, Cognitive Walls, and Accessibility: When Big Tools Are Worth the Cost

Matt: I have had success though, with some sort of iterative just talking with a difficult article. And that's just I'm reading a paragraph at a time and doing critical reading. I'm copy pasting this one sentence and saying, I don't really understand what this is.\ Here's the context and here's what I think it's saying. And here is the word in particular that is blowing my mind. And here is the phrase, could you explain it as if I was a 10th grader, or an eighth grader, or a 12th grader or something like that. Please give me specific examples and then give me an analogy to help me see it, and so on. It helps build momentum in my reading if I use it that way. And I feel like that, if we're looking at it from a strategic use case, would be useful. But ecologically speaking, then it changes the cost benefit analysis because no longer are we using a thing which is like using a cannon to take care of a fly, but rather we're using the cannon to take care of a wall of cognitive misunderstanding. And it is a really useful cannon. And how do we think about that Cost benefit, I think is sort of what I'm wondering about? Yeah.

Sophie: And what I would say is it's not just you using a cannon to take down a wall instead of a fly. You're using a cannon where you have no other weaponry at hand. Like, I think...the two pieces of it for me, Number one, "How good is I at doing this thing?" Number two, "Is there something else that can do that thing?"

You know. AI is better at summarizing a large corpus of writing than other tools are. It just is. AI is not as good at making an annotated bibliography. Um, even if it were. Even if AI were good at making an annotated bibliography, Do you have other tools at your disposal? Of course. But we've been making annotated bibliographies for decades. Um, of course there are other ways to do that. Are there other ways to summarize a thousand Amazon reviews? Are there other ways to get analysis and feedback on reading you're struggling with? Maybe not.

Stacey: I was really impressed when you demoed it in your WRT102 class because it's something that I had never really spent much time engaging with in that way, that summary. And I do think it's an accessibility tool to some extent. You know, it's making these, uh, these articles, these challenging pieces of text, like, really accessible to students. And I think that's something that can't be discounted.

Matt: I think that's a really important point, because that gets to what my question about this kind of was, was, is this going to flatten people's cognitive understanding of nuance in research and discussion of research, or does it create an entry point for people who never had a chance to get into research? And maybe the answer is both. And you have to be strategic about whether or not you're facing a wall or you're facing a fly, because -- Denny would be so unhappy with the amount of war metaphors I'm using here -- He's always trying to get away from war metaphors. And here I am throwing cannons and walls. Or flies?
This just keeps on being the first go-to metaphor.

Sophie: It's not a war metaphor. It's a medieval metaphor.

Matt: There you go. That's how I'm -- If you're listening. Denny. This is about medieval, not about war.

Sophie: It's history!

Matt: This is historical! There you go everybody listening. You get some of Denny even when Denny's not here. But it does bring up a question here that you bring up Stacey. Where does a person not have strategic skills? Right. We said "use it where it's most appropriate." But what about if people don't know?

Garbage In, Colonized Out: AI, Equity, and Progress Myths

Matt: I keep coming back to this idea of academic ghost writing. Um. The idea of Let's say you got a doctor who wants to write an article about a thing, but this doctor is not a good writer or communicator but is an excellent specialist and doesn't have the time to write this article.

So he's like, "hey, grad assistant. Hey, um, whomever, hey, person on Fiver or some gig economy online thing that will write my paper for five bucks. Here are a page of bullet points. Please write an article for WebMD." And that becomes the corpus of knowledge for that on the web that you know is accessible for AI. And then that gets fed into Wikipedia and it gets fed into whatever else. And then it just becomes this ouroboros of stuff that was originally -- I don't know. It's this kind of comments that people say that "it will get better. AI will get better. You know, give it time, it will get better." And I keep on saying "Will it get better if the foundations are inherently built on a game of telephone?" Like you said earlier, photocopies of photocopies when the actual original was a photocopy of someone else's mind itself?

And that kind of danger of misrepresentation, particularly how it's built on stuff like old stereotypes too, is just deeply problematic in ways that are both reifying dangerous perspectives and inaccurate ones. Do you have thoughts on that particular danger?

Sophie: So I think. That brings up a couple of columns that I came across, um, as I was looking for some information about I. One of these is just written by a random person on the internet, such as myself, who I think phrased it very well. "ChatGPT lies not because it's regurgitating falsehoods that it found out on the internet. It lies because it invents new falsehoods on its own." Which I would slightly disagree with because it actually does both. Which I think is what you're trying to say, Matt. And I have an example of this that I would be delighted to share later on. The other thing is, um, I found a blog post written by Ed Zitron, who is famous in some small field that's not big enough for me to know about, but he was writing about OpenAI's claim that they had created a five-level system to chart the development of artificial intelligence, and they say, "well, we're on level one, but, um, cusp of level two." And Ed Zitron was analyzing their plans nd found, and I quote, "to reach reasoning level, large language models would have to do things they are currently incapable of doing. To reach level three, you need a level of sentience dependent on on entirely new forms of mathematics."

Um. So I think that speaks to your point about "Is it getting better?" And, you know, we saw a real explosion last year. Of huge dramatic improvements in various forms of AI. That took place really over the course of about 3 or 4 months. I remember seeing with AI generated art just how fast, not how fast, but like how impressive the difference was from something that somebody made in April compared to January. However, since then there's been no significant improvement. And I think it's an example of the mindset of corporate America, of infinite growth. Um. You know, "oh, we just have to get the ball rolling and it will go forever and get faster and faster and better and better." And oftentimes that's not what happens. I think that what you're seeing when, on a technical level with the development of AI, is very much like what we are seeing economically with it, which is a sudden Extreme growth and then plateauing. I'm not convinced that we're going to rise from this plateau.

I think, as humans, we have always been fascinated by the idea of artificial intelligence. You know, Isaac Asimov lived -- this is your trivia for the day -- lived just down the road from Lasell, you know. Um, so we've been talking about, you know, the concept of artificial intelligence can think for itself for itself since the 40s. I think that we love the idea. We want it to be true because it's such a fascinating thing to think about, you know? I love my stories about tragic, sad robots, uh. I will shed tears for my Laprus and 2B and my favorite robot video game characters long before I will cry for their human counterparts. But I think that has maybe unduly influenced our impressions of factually what is happening and what we can expect.

Stacey: Yeah. I mean, well, they're certainly trying. There's a huge labor population that these companies are employing. This one article describes a workforce in Kenya and in the U.S. as well, you know,
monitoring both reports that users have submitted both says this is either, um, problematic or incorrect or whatever, but also just kind of monitoring. AI and while these people are evidently vastly underpaid, they're still employed and they're still working towards improvement. But I wonder the monetization piece, It's going to come in a big way. I mean, I was thinking when we were talking about summarizing paragraphs or getting, you know, analogies and things like, would you do it if it cost $5? You know, would you do it if it if you had to pay for it? And I know there are AI tools that, um, currently cost money, but, you know, it's vastly accessible right now. So I'm always thinking about, you know. They're putting massive investments into these tools and what that's going to look like. Um, going forward, I couldn't say.

Matt: I can't either, but I do suspect there is a very strong possibility that it's -- as you described Sophie -- more of a fallacy of exponential growth. I have wondered how much of this is just vaporware, recommendations of "It's gonna get there." It is more about the shareholders, you know, rather than it is about destiny.

Sophie: Yes.

Matt: And that is an excellent point about labor exploitation, because that's part of the ecological damage equation, but on a different angle of extraction. We've been talking about this in our book club recently, how AI is built upon systems that use however much water with each individual search, built upon systems that need to -- for those image generators to get those fantastic results -- However many people need to just sort of sit and click and say that the cat that's a dog, that's a this or that, and maybe it's not just monotonous and boring, but rather here are a series of deeply disturbing bits of imagery that you have to content filter out so that the AI doesn't just give you inaccurate stuff, but literally insane materials. And it's built upon that foundation, and it's built upon foundations of, um, gender and racial stereotypes. I keep coming back to this example from the book we were reading, which was talking about the corpus of data and the categorization methods of it, and how haphazard it was.
And an example that keeps coming back to me is just the way that they labeled. Categories of people, and one category was just "Loser," and it was just a person. And it's not just that it was calling him a loser, but also how do you identify a loser by sight? It's not really a quality that a person has by visual means. It's more of a relationship or a thing.

I mean, how do you use that as an organizational thing and what is that going to produce that method in the output in the end?

Sophie: Right, garbage in, garbage out.

Matt: Yeah, that's the short version of this particular point. So if I can take anything away from our discussion, it's that because of all those deep garbage in, garbage out caveats, strategic use is the most important incumbency upon everybody. Because I keep on, I find myself always down on AI. But it does feel like it does have value for, you mentioned new mathematics, for instance. I was just reading an article from a mathematician and I can't remember his name, but he was identified as probably the strongest mathematician alive in some way, and he was talking about it being able to be used.
I think he described ChatGPT's best model, as a not-incompetent grad assistant. And he basically was saying, yeah, we probably could invent new math with this.

If that capability is there, could this thing cure cancer, for instance? If you're trying to say, what is my use case, I would like to use it to summarize a annotated bibliography. Well, it's not going to do a good job and your result isn't going to be useful. So that's not good. I would like to use it to cure this disease. That does seem like something that's worth using.

Sophie: And I think that is a case where it's certainly to go back to our war metaphors. It's certainly something where it could be part of an arsenal. Um, there's a program. I can't remember what it's called, but you can download the program. I think it comes from the centers for Disease Control, where somebody, uh. But you can link up your computer to this network, and it will use your computer's GPU to run complicated graphical work and manifold work to, like, map out the genome of cancer. And, and I think that, I don't know that it will happen, but I can imagine that could happen, that we could have programs where that's part, um, the strategic plan. And using AI is part of the strategic planning if we have a bunch of different methodologies. then they can come together and be more than the sum of their parts. I think that that would be a good use of AI. I haven't seen that happen.

Stacey: That's a ways off. Optimistically.

Matt: That's the vaporware thing. It would be nice if we lived in a utopia.

Sophie: That's right.

Matt: And just saying the word utopia. Doesn't get you the money, or it does if you're Sam Altman apparently. But, you know, it depends upon your strategic use, I guess.

Misinformation and the Michelle Branch Singularity

Matt: I think that we've covered everything that we have in here except misinformation. And in a lot of ways, we have covered misinformation because we've covered the foundations of AI effectively being flawed and people needing to be very strategic and thoughtful about to use, because that is the next step of what will happen, that you'll have a flood of misinformation and have had a flood of information, deepfakes, so on and so forth. And the only thing that we have on our side is to figure out how many fingers are on the the images that we have. Yes, that's the only way to know if that picture was not real. It's like, "Oh, that person has seven fingers. That's a fake one!"

Sophie: Look at the pupils. That's another good one. Yeah. AI is very bad at drawing pupils.

Stacey: One thing I'll add about misinformation is an article I read recently, um, in in the Harvard Kennedy School about the prevalence of kind of the next advancement or the opposite of advancement to what Sophie described earlier. Um, is that on Google Scholar, there's a ton of like, articles that are generated by AI, completely fake, but linked to real journals. They're not published in the journals. But they look like they are. It's like the next level of fake news. It's like fake research. And they're finding more and more of these and of course a lot about vaccines. But alot about immigration, etc. So it's getting harder and harder -- which again I'll go back to the Information Literacy Foundation that I think we're all responsible for embedding in our students -- And it just getting more and more challenging to kind of do with these things out.

Sophie: Yeah. And I think what that reminds me of Stacey is. Follow the money. Like why why are we even sitting here having this conversation? It's because some very, not only very rich, but very powerful people have invested a lot of money. You know why does Google add those AI summaries? Because they've invested a lot of money in AI. They want -- It's a political ploy, not political in the sense of governmental, but in the sense of political theory. They want people to think positively about it. They want AI to provoke a positive bias in people's minds because it's their product.

And since that's all I have to say about that, I want to jump back to misinformation. And even -- you mentioned earlier, Matt, the Ouroboros effect -- which is something that has happened even before AI. I actually encountered -- This is my personal area of knowledge, pop music -- Um, it's something that I encounted the day before yesterday, I think. I was adding an album to, uh, Music brains, which is a metadata service that collects data on albums and songs and performance. And, um, there was one of the tracks was a cover of a Michelle Branch song that this performer had co-written. And that song was in Musicbrains already. But it was only credited to Michelle Branch and John Shanks. Not Tiffany Arbuckle or Matt Brown.

So I was like, "oh, that's a little weird." I looked it up on Wikipedia and sure enough, it says, oh yes, this was written by Michelle Branch and John Shanks. And I thought, "I don't think she was lying about co-writing the song," so I looked it up in BMI. BMI shows Michelle Branch, John Shanks, Tiffany Arbuckle, Matt Brown. But somewhere at some point, someone heard an interview with Michelle Branch where she talked about co-writing the song with John Shanks. And they wrote a sentence about it on Wikipedia, and then that made it into Wikipedia's info box, and then that made it to Wikidata, and then that made it some MusicBrains. So even before AI, we see this chain of events very much in the same mode.

I wrote to ChatGPT, and I said. "Give me a list of all songs co-written by Kylie Minogue and Karen Poole," because I'm a huge fan. I know she's friends with this other songwriter. I know they've written a bunch of songs together. And fully expecting ChatGPT to actually give me the correct answer and its says.

Breathe
The One.
Can't get you out of my head. Additional writing credit.

And I'm thinking to myself. Karen did not co-write any of those songs. So, I said, "Give me full songwriting credits for the songs above." And it gave me just completely wrong. information. Completely wrong songwriting credits, which should be the absolute easiest thing for it to take in. Because it's right in Wikipedia, it's in Wikidata, it's everywhere. And, even like made people up that it claimed were songwriters. So then I finally said, "Who wrote the Kylie Minogue song 'Wow'?"

"Song "Wow" by Kylie Minogue was written by Kylie Minogue. Karen Poole and Greg Kirsten." I'm like, "Yes it was. Yes it was. And Funnily enough, it wasn't in that list of three other songs that were not that you gave me," so that, I, went in knowing what the correct answer was and I really had to work for it to give me that correct answer. And in the realm of information literacy. I think that's something that's very dangerous.

Final Thoughts: Use Strategically, Teach Relentlessly

Matt: Yeah. It reminds me of, um, a key term that I think we come back to which is a person's responsibility itself. Because it [the LLM] has no responsibility. If you had said you were wrong, actually. Because earlier you directly contradicted yourself, it would have been like, "oh, I'm sorry, as a large language language model. I don't know what I'm doing, you know. Buyer beware. I'm an idiot over here. Sorry about that."

But the degree of anthropomorphization behind it, where we describe it as a thing, it has agency. It continues to make us think that "Well, you know, it's the systems that are making these mistakes, the things that we're relying upon..." An no, it's our own decisions to rely upon the systems in different situations that in small details like songwriting credits or in big details like different narratives of what happened on specific days can become enormously Orwellian. Just change history. That didn't happen because the information corpus that we have ironed it out of history, That's for you too, Denny, by the way.

Sophie: This podcast is like Stefon. This podcast has everything.

Matt: [laughs] References to Denny.

Sophie: We had cannons and Michelle Branch and [inaudible], and Beees.

Matt: And Bees, yes, yes. It's just bees will attack you here. It's like "Stefon, that's not what I asked for. Oh I'm sorry..." That that's an interesting comparison. I've never heard anyone compare AI to Stefon, but that would be at least a funny way for the end of the world.

Stacey: I think that's a wrapping point.

Sophie: I think that's a great high to go out on.

Matt: I think so to. So thank you very much for joining. Thank you very much I appreciate it. Goodbye!

Thanks for listening to us, everybody. If you want to learn more about teaching, learning and educational technology here at lasell, please reach out to tlc@lasell.edu or give us a call (617) 243-2001. Thanks. Creative Commons, music by Jason Shaw at audionautix.com