SIO529: AI in Schools Makes Students Feel Smarter. The Tests Say Otherwise.

Transcript

This transcript was generated automatically and may contain errors — the audio is the record.

0:00:13 Thomas: Hello and welcome to Serious Inquiries Only, this is episode 529, I'm Thomas, and joining me again in the intro as well, celebrity on this podcast, Dr. Jenessa Seymour, how you doing? Weird computer stuff that's that's quaint because aren't we talking about AI like people don't even talk about like I feel like you sound old if you're like computers because now it's just like AI it's

0:00:26 Jenessa: Good. I'm to talk about some weird computer stuff today. That'll be fun.

0:00:34 Jenessa: Yeah. You're probably right. Yeah I oh god I probably do sound on so crotchety.

0:00:47 Thomas: Like when I tried it and I predicted this a long time ago because I think about stupid things like this but even before I had kids I predicted this meaningless but it came true trying to tell my kids what the internet is they don't understand that it's the internet that's them it's just like yeah things phone and the ipad has all this stuff the tv has all the stuff they don't really use computers yet but I'm like oh there's no internet on this they're like what are you talking about you know because they don't understand the concept of like it being connected

0:01:15 Jenessa: They have to have their like, Helen Keller with the water in the mug moment of like, the computer's on, but there's no internet. There is only a Word document.

0:01:25 Thomas: Yeah yeah it's really funny um yeah I in schools you want

0:01:29 Jenessa: Specifically how it affects the learning process. So I, you know, if you're trying to to learn new information, obtain a skill, how is that going to impact you?

0:01:42 Thomas: I feel like it's pretty early to have too much information on that sure

0:01:49 Jenessa: Early for sure. So we're going to cover some of the, you'd be surprised the speed of publishing these days. I wonder why. But there is some good research coming out on it.

0:01:55 Thomas: Probably yeah yeah this has a bunch of m dashes and says ai is great weird it says

0:02:05 Jenessa: It. The diagram is just a weird picture of a rat.

0:02:09 Thomas: Ai is not great it's actually this like it has a a bunch of that language construction for sure

0:02:15 Jenessa: There unfortunately quite a bit of that, but there's some good stuff coming out. And luckily, a lot of these issues are not new. So we have plenty of data on how people learn and how different forms of technology have benefited And harmed that and the ways that interacts. So it's, think for those of us who have experience with education or have been following the science on education, a lot of it's not very surprising. Lining up. And so it doesn't feel too early to talk about it because it really fits what's expected, given how other technologies have interfaced with it. It

0:02:55 Thomas: Well I'm really curious to talk about that something I am wondering about I don't think I don't think there's any ai in my kids stuff yet I'm pretty I feel like we're pretty lucky I love the our little school our little public school you know it's like it's it's nothing it's special but it just feels like a lot of pretty sensible people you know where it'slike when I hear these other horror stories I'm like I don't know if that's made up or if we just we're have lucky with our little nice little school you know I don't know there's a lot of weird stuff happening around the country but uh no ai stuff yet there's definitely a lotof technology and I wonder how that goes that's part of it all right well we'll take our quick break and then we can dig into all that patreon.com/seriouspod uh you'd like to support learning cool science stuff from Jenessa especially especially lately. So she says.

0:03:40 Jenessa: We've a computer real human over here yeah

0:03:57 Thomas: Right. So I'm very curious how much data we have, what it is. What are you going to tell us here?

0:04:02 Jenessa: So uh definitely an issue that's been uh in the news over here in new york because we relatively recently banned cell phones uh and then uh you might have seen mom donnie just put out a policy of for the next year no AI K -8 and studying it for high like what is the impact, positive or negative, reduction in use. Definitely a lot of interest in what exactly is going on here. Of this is going to interact with the more general issue of screens, screens so to just mention that briefly um from screens versus paper people learn more and they're better able to assess their learning um of that might just be screens are associated with so many distracting things like your computer screen is not just just for reading. It is for internet and social media and playing games.

0:05:00 Thomas: Mine's not at all for reading. I banned reading

0:05:02 Jenessa: And so, yeah, I mean, it's, I can certainly attest to like, I know it's an anecdote, but like personally I'm, I'm murdering all the trees. I'm printing everything because I just, I can't read on a screen. It's, it's so distracting. Uh,

0:05:02 Thomas: On mine. I can't read anywhere though. This isn't racist against screens. I just can't.

0:05:21 Jenessa: Uh, I generally am a pretty, like, I'm not a fast reader, but I'm, enjoy reading. I'm pretty good at it. I can read and absorb and, you know, I for what it's worth, I score good on those tests, whatever that means. On a screen, I know people are gonna be like, you can highlight on the screen. And like, it's not the same thing. The And data backs that up, for me, at least. And same thing is happening with writing too, like taking handwritten notes versus typing your notes has a huge impact.

0:05:50 Jenessa: I think there's anything magic about moving your hand. It always comes off kind of wooey to me. Like there might be some impact of like motor. Okay, here's what I'll say

0:06:02 Thomas: I could see it. I mean, there's a lot of stuff that's like, Like, you memory or the movement helps you remember it. I mean, I do theater and there's all kinds of things where it's like I could never memorize all of that. And then once you've associated the blocking and the movement, the whatever, like helps. I could see. I don't think it's crazy to think that like the physical writing. I tend to associate a little bit now, not that I do this anymore, but as a kid when I did read and write, I tend do associate to sort of where remember things were on a page a little little bit and like have some of those spatial memories with it

0:06:32 Jenessa: So I definitely think that makes sense of like I used to do this a lot where I would rewrite over and over again if there was something I just had to memorize there was no other way around it um would re -write it in the exact same location notation on a piece of paper over and over again. And I might even color code it. And then very conspicuously, so nobody thinks I'm cheating, pull out a pieceof paper and do it again. When I sat down for the test, you could just rewrite it .And now there it was. So there is a motor and visual memory aspect to that kind of thing. If you want to just straight up memorize something. Yeah, Yeah, I visual probably has a little more to do with that. But certainly the aspect of like moving your hand. Yeah, you could use that for memory. Taking your notes in class. I don't know that that is going to have the biggest impact. I think most of what's going on is my students who are typing their notes. Well, first of all, they could just.

0:07:31 Thomas: Not processing and listening they're just not I'm sorry I wasn't a notes in class guy I hated I did not in high school I refused to take any notes ever unless the teacher made me and I just not I it's don't not like I learned everything but like that's not the baseline neither do the other kids I was one of the best students I learned a lot when it was math I was perfect at it because it's just is what I liked. And it was just being able to pay attention. Now, I don't know if ADHD would make that better or worse, but the writing, I always felt like if I'm trying to type, especially the, the kids who would try to type like every word the teacher's saying or something, you know, it's just like, there's no way you're

0:07:34 Jenessa: don'T think so. So yes, right.

0:08:05 Thomas: Actually processing this.

0:08:08 Jenessa: I think that's exactly the problem is because you can type so much faster than you can write, you could just kind of transcribe what I'm saying and like, what's that doing? Nothing, really. You don't need to write down everything I said. If you're handwriting, you can't write everything I say. So you have to stop and think, what's important here? That was an aside. That was nothing. Sorry, I'm teaching live. I'm not perfect.

0:08:37 Jenessa: Sometimes I say, sorry, I try my best, whatever. Most of what I'm saying is useful for educational purposes, but some of it is an example that's trying to help you illustrate and understand the point. You don't have to write down the example. You just have to to write down the larger point. So you're actually doing active listening and synthesis. And you might realize, oh, I have no idea what point you're making. I can't write anything down because I don't know what to write. And then you raise your hand and ask me, what the hell are you talking about? What is the point of that? Are some of my favorite questions where just like, I don't get it. Help me. And I could actually have a back and forth with a student of like, tell me what you you don't understand. Like, let's talk it through. So I think that's a huge part of it is in addition to the fact that the screen itself is distracting, the student who's typing notes is kind of just transcribing what I'm saying. And that's so useless. That's not what you need to be doing.

0:08:37 Thomas: You're a little hard on on yourself. Yeah.

0:09:38 Jenessa: That's distracting itself. So, Some of this is just very simply, I would expect to see detriments just from having a screen in the classroom. The student who is using AI in the the classroom is using a screen in the classroom. We have to pull that confound out somehow. Of the larger things that's going on here, I don't know how much we've talked about this before. You've probably heard the term. It's a really great predictor of learning. It's basically your understanding of your own learning. A couple of ways we can break it down. One is ease of learning, so am I good at learning? What will be more challenging for me? I know the biology class is going to go well for me. I know that the physics class just doesn't get my attention. I'm just not into it, whatever.

0:10:30 Thomas: God, you weirdo. Kind of a pervert are you? I know that math and physics will be great and I'll love it and everything else will be annoying and I don't want to do it.

0:10:39 Jenessa: Well, so this is important. If you know that, you know how many resources you need to devote to which class and you can prepare properly. Or, I mean, depending on your trajectory, you can choose not to take the things that you're just not interested in. That's ease of learning. It's like before class assessment. A judgment of learning would be what have I learned well? What did I fail to retain? This is an after class assessment? What do I need to go study better? Sat I through the class, whether that's a single class or the semester, how am I doing? What Did I understand? There's There's feeling of knowing. I know that I know this. Who can't answer a trivia question, you could say to them, okay, but how sure are you if I gave it to you as a multiple choice, do you know the answer? That is a really common feeling, right? Like it's on the tip of my tongue. I swear to God, I know I know This. As soon as I see it, I'm going to be so mad at myself that that I didn't know it.

0:11:42 Thomas: That's me with everything. Yeah,

0:11:43 Jenessa: Yeah.

0:11:43 Thomas: Sorry.

0:11:44 Jenessa: When people say that, their prediction is above chance. So people are correctly reporting their internal mental state.

0:11:52 Thomas: I've lost track of we how started on this thread. Is

0:11:54 Jenessa: This is all types of metacognition. We had ease of learning, judgments of learning feeling of knowing and another type of metaccognition, the fourth one, confidence judgment. So feeling of knowing before I give you the question, do you feel like you know the answer? Confidence, judgment, I'm sure I have given you the right answer. So this would be after a test, for example. You feel Like you have answered those test questions correctly? And your confidence does tend to be a good predictor of your test scores. Does not tend to Be a good predictor Of things like, for Example, eyewitness testimony. Kinds of things you you could frequently encounter in everyday life. People are very confident and they're wrong. They're low confidence and they write, they're all over the place. But in a learning environment, people are usually pretty good at like, I did well on that test and they did, or, oh, I failed that and they didn't. Is really important for learning. Poor metacognition in both directions is bad for you. So if you say, oh man, I think I failed that test and you actually did really well on it, that is a predictor of overall doing poorly in school. You are not good at assessing your own learning and your own abilities. And in the long run, maybe you did good on that test, but you're not going to do well in general. You need to be able to say, I did do well on that. That low confidence is really going to hurt you. It's just overall low confidence, you tend to underrate yourself. We can work on that if you are all over the place. I think I did poorly and I did well. I did poorly. Now have poor metacognition. We are are

0:13:42 Jenessa: Not correctly assessing ourselves.

0:13:41 Thomas: It's interesting to think about that as separate skill from just knowing the material because it's like if you know the material you know what you knew but if you don't know it then you don'T that's it so it's like so it'S yeah okay so it is still yeah it makes sense so it's like the metacognition in that you can kind of track what you learned and didn't and you're like oh I'm pretty sure I didn't remember this right or I didn't learn this part versus the overconfident like you took a test you're great and then you just failed that's that's not good

0:14:19 Jenessa: Right. So like the cognition is knowing the stuff, whether that's information or having acquired the skill or whatever. The metacognition is being able to assess yourself. Do I actually know what I know? There's like that famous, what's that quote from?

0:14:39 Thomas: The known unknowns unknown unknowns yeah it's

0:14:41 Jenessa: I put that on my slides in class.

0:14:44 Thomas: Not a bad quote just in general if you take it apart from everything else but yeah

0:14:49 Jenessa: Divorce it from the Iraq war, it's actually a great description.

0:14:50 Thomas: Take out the war crimes that's a solid quote right there

0:14:56 Jenessa: Really a great description of metacognition. There are things that I know I don't know, and then there are unknown unknowns. I don't even know that I don't know it, and that's a problem.

0:15:05 Thomas: Why couldn't someone else have said that in a better context

0:15:10 Jenessa: Yeah, it's not great. I dunno. My students probably don't have a lot of personal emotional involvement in the Iraq War.

0:15:14 Thomas: I know it's just funny oh

0:15:19 Jenessa: Okay, why am I talking about this? This is gonna be of our big obstacles is we don't need to just learn things. We need to know when we did and didn't learn things?

0:15:29 Thomas: Well okay so just in general like using ai doesn't mean you You just know everything because you're just kind of passively taking on information or even just being like a through line of information, like a relay of information.

0:15:43 Jenessa: So you have to be aware of the fact that you were given information, but that doesn't mean you learned it. That doesn't means it's in your brain. It's in the computer. Like, oh, and I the number of times that Zoolander scene pops into into my head lately. Okay. Really interesting study. They were looking at active engagement in the classroom versus more passive. So essentially, the very passive teaching, which was still, I mean, it's still teaching, you know. There was a lecture. The teacher goes through explanations of this problem. I think it was like a physics or a math class. They went through like some problems, explanations, demonstration of the concept. And then the teacher did the problem on the board, like worked through it in front of the students while explaining what they're doing and why. The active learning group, the first thing you did, you got some handouts and everything, but you work on the problem in small groups while your teacher is roaming around the room answering answering questions. And then at the very end, the teacher gives you the solution and an explanation. So the students in the active learning classroom learned more. On a test later of those concepts, they do better. Their feeling of learning, their metacognition, their judgment, they thought they did worse.

0:15:58 Thomas: Computer. Interesting.

0:17:13 Jenessa: They don't think they learned anything. And the students in the lecture are like, I understand this great. And compared to the active learning group.

0:17:23 Thomas: Interesting. That's almost like to quote another war criminal. No, I don't. It's like the wise man is wise and knows nothing. What is that? The wise man just knows he knows nothing? I don't know. What am I thinking of? The hell? Plato? Some shit? I don't know. Some other war criminal?

0:17:35 Jenessa: Damn it. I what you're talking about.

0:17:39 Thomas: It was, it was, oh man, I had the perfect joke. I can't, oh, man, who's the fucking guy who just died somewhat recently? That was the, you the, know, the diplomat that, am I thinking of? Yeah, Yeah, it Henry Kissinger.

0:17:49 Jenessa: Kissinger?

0:17:55 Thomas: Turns out every good metacognition quote is by a war criminal. It's the weirdest thing. Just turned out that way.

0:18:00 Jenessa: I guess they're just thinking real hard about this. I don't know. All the things they never knew and then they died. Personally, I have noticed the same thing when I teach and I go out of my way to use these active learning styles because I know my students learn better. They express to me, why am I even here? I'm just teaching myself. What am I doing in the classroom. And I'm like, oh my God, you have no, yeah, right? And I am like, you have no idea the work I put in to set you up for success right now. And you get like bad student evaluations. Of course, like college professors do not want to do it because their students think

0:18:06 Thomas: What would you say you do here, Jenessa? Yeah. That sucks.

0:18:45 Jenessa: They did not learn anything, but then they did. Have always gotten great evals because I teach fun stuff and that is also practically cheating at this point. But But like, it's really disincentivized because my students hate it, but I know they're learning better. So I think what might be happening there's this, the sensation of ease of learning in the lecture is being falsely perceived as more learning. Might be that's where the metacognition issue is coming in. Like when I tell you stuff you feel like you learned it so easily yeah

0:19:23 Thomas: You haven't. It hasn't been put to the test, really. It's like, oh, I got all this reserved. We've been here for 50 minutes. I've learned nothing but facts from you. And then the minute you're like, OK, tell do this thing or whatever. Display. This demonstration that you know something. That's when you're you're like, oh, no. Turns out I was listening and kind of not really learning exactly.

0:19:42 Thomas: Whereas if you're already doing that process, I think that makes you feel like, Oh, man, there's so much I don't know.

0:19:42 Jenessa: Mm-huh. Yeah right right and and it can be really frustrating for students I don't blame them sitting in a group trying to work through something it's just you who don't know very much and a couple other people who don't Know Very Much and I come over and you ask me questions and I do that thing where I smile and and I don't give you the answer. I just, I, I know.

0:20:06 Thomas: That's such a classic teacher move. Oh, you get to do those classic teacher moves. I'm jealous.

0:20:13 Jenessa: I mean, it is, it's fun, especially when they are engaged and when they aren't, they're not willing to play along with me. And I'm like, okay, what do you think? Like, what I'll give you a little hint. And if they're willing to pay along, like they learn it's for both of us. Um it can be really frustrating and I don't blame people for it, but it is a better way to learn. So we got to combine these two things. We've got metacognition, which is understanding it and not just understanding the material, but understanding whether or not you learned it properly. And the other concept that has, again, it's been around long before AI, cognitive offloading. You might've been hearing this a lot lately with AI research.

0:20:14 Thomas: It is fun. Yeah. Yeah, yeah.

0:20:57 Jenessa: This too, old concept. I pulled, I'll share with everybody, a study from 2016, a meta -analysis. So that's before any of that stuff hit the market, just just walking through what we already knew. Even is it? Very simply taking a cognitive process and putting it onto either your physical body or items in the world to take away the cognitive effort. So that could be something as simple as running your finger along the text to keep your eyes focused. Was cognitive effort, now I offloaded that onto my body. Something down, I am trying to keep all this stuff in memory, now I could put in substantial cognitive effort, or maybe it's impossible to keep it all in my memory, or I could offload that onto a piece of paper, a computer's memory drive, whatever. So might sound bad. It's not inherently a bad thing. Clearly, it's neutral. It is taking a cognitive task and getting it out of your brain and onto a paper or something like that. Going to relate to both your actual limitations and your metacognition so sometimes we set reminders because I couldn't possibly remember all of these things we do have studies that sometimes people set reminders porque they don't trust themselves to remember all of these and we can demonstrate they're well within their memory tolerance limit but they just like do it anyway and it's kind of a metac cognition problem um I'm writing all this down because I think I can't remember it all. So sometimes the offloading is a literal limitation. It's easier, or it might even be impossible to do this without some offloadings. Sometimes it's a metacognition thing of like, I think, I can do this, without technology, or you know, counting on my fingers, whatever, and actually I can.

0:23:00 Jenessa: This one was interesting. They did a memory test. They gave people a bunch of trivia information and type it onto the computer while I'm saying it really quick. Already miserable. I hate this.

0:23:10 Thomas: I might. Yeah I'm stressed just hearing hearing about this oh

0:23:12 Jenessa: Half of them get told, you're going to type up everything I'm telling you, but I'm going to delete the document immediately after. And the other half don't realize the doc...

0:23:21 Thomas: To me pretending to type the whole time what's the fucking point you're deleting it I don't why would I do this nope

0:23:29 Jenessa: Neither of us could ever be in any of these studies.

0:23:31 Thomas: I I'm the word I just I want here's my new skit show just different experiments where I'm an asshole and I can't the the scientists just euthanize me huh

0:23:42 Jenessa: So the group that knew in advance, I'm going to type this all up, but you're going to take it away from me. They had better recall. So there might be a sense in which cognitive offloading could reduce performance.

0:23:51 Thomas: Okay so they're yeah yeah I mean that makes a ton of sense to me

0:24:00 Jenessa: Yeah, you don't bother to remember it because you think it's stored somewhere else. It's almost obvious.

0:24:03 Thomas: I sort of worry about, you know, that's kind of, I can't remember. God, this, I've just got, it's so meta on my meta, meta, metacognition. I can't remember where the fuck I heard this, but I do worry that with my ADHD, that I'm enabling myself with all my ways of remembering stuff. Cause here's what happened in the year 2009.

0:24:25 Thomas: I tried to remember a four or five digit number and it was like, I can't remember it. And something, you know, with activation code, whatever. And then from then I'm like, well, fuck this. I can't do that. And so I've never tried since I never try. If I'm like, all right, activation code. It doesn't matter if I just got the text. I, I, fortunately the phone lets you do it. But now when I, when I know I can, like the phone isn't going to do it, I have to go back to it and like, say it out loud a couple of times, the six digit number. But I really wonder if it's like, damn, did I just let that part of my brain atrophy because I just decided I can't do it anymore?

0:25:03 Jenessa: It is possible that might be a thing that's going on. I don't have any evidence in front of me that these things things are permanent, that you could never do it again. The atrophy idea of that was a skill you had, you haven't used it in so long that it is weakened, entirely possible because it's offloaded onto... the way I often approach that is anytime looking at a skill where I suspect I might be doing that. Do I care?

0:25:24 Thomas: It. Yeah. Yeah, I offloaded my caring. I know I'm not sure about that either.

0:25:37 Jenessa: Oops. Like on the one hand, that one I think is evidence of like, yeah, if you think that it's almost obvious, if you think the thing is stored somewhere, you're not going to remember it.

0:25:47 Thomas: Not gonna bother.

0:25:49 Jenessa: On other hand, they had people type up a list of words. One group knows they can save it and the other group can't and then they learn a second list of words the group that could save the first list had better memory for the second list so it might be if you can offload some of these memories it's easier to remember other stuff because you don't have your brain all cluttered up um is important to note this is not about like long-term memory humans have a basically basically unlimited capacity with a long-term like you're never gonna hit like I learned too much and I just I have too much biology in my brain and I cannot learn chemistry now like that's not gonna happen you're not gonna meet too many people and now it's just all over for you'll never learn a new name again

0:26:35 Thomas: A question and this This might not be obviously something you could answer now, but I do get a feeling once again, as the kind of person I am who takes on a lot of different new hobbies and whatever. I do have a feeling of like, awesome. I'm into this new thing that I've gotten really into. And then I look at the, whatever the thing I was doing like last week, I'm like, I don't even recognize that anymore. Like, what the fuck is that? What is that. Um the question specifically I do want to ask though is is there any connection between the learning the new thing or is it just the time that goes by that would do it you know what I mean oh

0:27:09 Jenessa: There's a couple of factors there one is just forgetting is an inevitable part of all of this. From the process of going from sensory memory to short-term memory, there is forgetting. Going from short-term to long-term, there is forgetting, then there's a couple of different memory effects that happen where, for example, if you learn very similar information, if you get a new locker combination, it can retroactively interfere with the old locker combination um that one's killer

0:27:36 Thomas: Jesus christ tell you what I reached a limit of zip codes I lived in too many places and the zip codes are too similar because I'm they're all in california roughly you know they're all like so I got to a point the other day where I had to ask my kid our address because I was like I can't I forgot like it's too many it's been like the fifth place I've lived this fucking god how many am I supposed to remember I don't remember any of the other ones either I just I got at the point where that what made me think of it is the similarness like because they all got kind of similar and I learned too many of them I can't remember like like which one went to what house and what, you know what I mean?

0:28:16 Jenessa: Mm-hmm. Yeah the fancy term for that is interference um there's I always mix this up there's retroactive and proactive interference and one is like the past interfering with the future and one isthe future interfering withthe past I never remember which one is which um figure um yeah so that kind of thing can happen a lot it's like in the moment of recall you're trying to pull up one thing you accidentally pull up another thing and then it's like oh I'm never gonna get the one I was looking for now because I just keep getting this other thing um like the general concept of like um they've done this before with like uh people who came from like a small high school class they can use the yearbook to be like hey just tell me how many names you know, free recall, and then test out with the free recall and then with like matching and stuff. There does not seem to be an age limit. Like you don't just have to forget stuff. It'll just stick around. And they've done capacity tests where it's like, learn this many pictures. And obviously you were never going to get 10,000 pictures. I can't test all 10,000. That's crazy. But I'm just going to test a subset. And like, I person learns a set of 100. This person learns the set of 1000. This person learns a set of 10,000 pictures. And when I test you on a subset of 50, everybody's at like 75% accuracy. So that kind of means you got 75% of 10,000 pictures. Like there just doesn't seem to to be a cap on your capacity

0:29:46 Jenessa: For memory. Yeah, human brains are exceptional.

0:29:47 Thomas: Go us. It's insane.

0:29:52 Jenessa: Wattage of what it takes to run a shitty light bulb, you are the best computer in the world. It's an ongoing mystery of neuroscience. It truly exceptional. Okay, coming are you doing when you're offloading stuff? At this metacognition and your knowledge, um of the things they looked at uh this was the big hot question back then was oh my god google is google ruining your brain maybe

0:30:18 Thomas: Yeah I mean kind of yeah. Not for the reasons you think.

0:30:23 Jenessa: Right so um looked to people who used google while they were studying something and they found that they overestimated how much they knew about another topic they hadn't even used Google to research. It seems like when they used Google to research topic X and they got so much information and they felt like such a smarty pants, when I asked them how much they knew about topic Y, they were like, I know so much. I'm going to do great on this test. Even when you tell me I'm not going to be able to Google anything, I'm like, I've got this. I'm going to be great. And of course, that does not pan out, in fact, when you then test them. So liked this quote. The author said, these results suggest that participating in a human internet transactive memory system can lead individuals to blur the distinction between what they know and what the internet knows, scare quotes. Very interesting because on the one hand, that was like too high a confidence. But then in another experiment, they asked participants, in I'm going to ask you a whole bunch of questions, just facts and whatever. In one group. If you say I don't know, I'll just move on to the next question. In the other group, If you Say I Don't Know, I Will Let You Google It. Group who was allowed to Google it was way more likely to say, I don't know, and report lower confidence, lower levels of that feeling of knowing metacognition. So feeling of Knowing is that one of, well, I, I don't have it on the top of my head, but I know if I look it up, I'm going to recognize it. They actually had lower levels of that. And they gave fewer overall correct answers, which indicates they were saying, I don' t know, when they did. They were less likely to give incorrect answers. So like Like their accuracy score, if you compare correct and incorrect, they did do better. They were overall underestimating themselves. They were skipping a time when they could have said, I got this and given an answer. They just said, I don't know, and looked it up instead. In some contexts, that's probably a good thing.

0:32:38 Thomas: Yeah but it's more of that thinking you have this because yeah Yeah. Cause there is a, yeah, there is a resource out there where you can just look it up. Yeah.

0:32:43 Jenessa: They're underestimading. Right. So people use it and then afterwards think they're a genius. And then on the other hand, knowing you could look it up and makes you think you don't know anything in advance, like it is messing with you. Okay, this is all stuff. We already knew this, all of that about how metacognition works and about cognitive offloading and the way that using tools, I mean, obviously it can help. You can't remember infinity things in the short term and your memory is not perfect accuracy. The way that like, if I open a Word document and a single character is different from last time, like something has gone terribly wrong. Of And course, my memory is Not Like That. So there are, of course advantages to using the tool, but there might be these disadvantages this why when we get to the ai stuff a lot of this was not super surprising um interesting paper on this uh generative ai without guardrails can harm learning evidence from high school mathematics so they did here three groups groups one studies without ai one has essentially just base access to chat gpt and one is going to get access to a modified tutor so they not going to just answer your question it's going to give like hints first and it also has notes from the teacher like common mistakes mistakes students tend to make. It also knows the right answer, so it won't accidentally give you the wrong answer on anything. So everybody goes through step one. They get a lecture on the topic. So right off the bat, these are high school students. They've had math exposure before, and then they get a lecturer on the topic. If the chat GPT like taught you nothing or hurt your learning, we shouldn't expect you to get like a zero on the test, right? You've learned, You know some math. Two, you're going to do practice problems. Group one is just completely unassisted. Now, this is a complaint I do have about this study is they said nobody got teacher assistance. And to me, that does not replicate what a classroom, like in a classroom if it was like no AI and you're all sitting there doing practice problems, your teacher would be wandering around answering questions,

0:35:13 Jenessa: You know? So it's a little weird to me that the...

0:35:13 Thomas: Or coyly not answering them.

0:35:18 Jenessa: Yes, exactly. So the no AI group also got no... Well, everybody got no teacher. So to me, that group... It's a control group, but it's not a realistic classroom environment. I don't love that. Me, it biases it in favor of finding that like, oh, the AI helped me so much. And it's like, like, well, literally anything helped you so much, but whatever. They do the practice problems either all by themselves with the base model chat GPT or with the specialized tutor. And then step three, closed book test, everybody. During the practice session, you do these practice problems. How'd you do? You had these little math questions. What kind of answers did you get? There was a small effect size for chat GPT base to improve performance and a moderate effect size for the tutor. It was doing pretty good. Granted that the other students had nothing to help them. On the test, the students with the base model chat GPTE did significantly worse, about 17%. It's not terrible, but it's not great.

0:35:30 Thomas: Yeah gotta pay for the premium. That's fine. No, I'm just kidding.

0:36:31 Jenessa: The ones who got the specialized tutor, they were not different from the controls. So the controls who just worked through practice problems with no feedback from their teacher or anything did the same as the students who practiced with a specialized tutor that had the teacher's notes and would not give them the answer, would only give them hints. They just did the same on the test.

0:36:55 Thomas: Sorry, really stupid question. Is it a tutor like a real tutor or a, or an AI that's designed to be a perfect tutor? Okay. Just making sure.

0:37:01 Jenessa: AI tutor, no human. Yeah, yeah. So the really interesting part to me is the feeling of learning that they asked them to rate, how much do you feel like you learned? And then after the test, how do you think you did on the test? The students who were just using ChatGPT base model, they rated themselves as learning just as well and performing just as well as what the controls said about themselves, even though they performed worse. And the students who trained with the tutor they thought they learned more and performed better, even thought they had not. So, that probably the biggest concern I'm seeing people raise is that it's It's not even so much that it's necessarily going to hurt your learning. There's a little bit of a problem of if Google the answers, if you just chat GPT the answers. That's not learning. That's somebody telling you the answer. Your teacher wouldn't just tell you the answer, a good teacher. Also, if you feel like you learned and you didn't, that is bad news. That's probably worse for education. It's one thing if you didn' didn't learn the information. It's a whole other can of worms. If you are in a position where you feel like you learned stuff, but

0:38:20 Jenessa: You didn't, now we have a deeper problem about learning. Know, it makes me feel like I've heard people say like, oh, you know, I used to struggle so much in school. And like now with this, I learned so great. And I'm like, Oh no. Do you, or do you feel like you learned. Oh no.

0:38:37 Thomas: I mean, it's sort of related to the, to the the Dunning-Kruger idea, which I now can't remember if we debunked on here or not. Does that still hold? That's a long time ago.

0:38:45 Jenessa: Yeah. The, the general issue is, um, public perception. The Dunning-Kruger is not stupid people think they're smart. Dunning Kruger is any you are, all people, when you are asked to estimate your skill at at something you are not terribly skilled at, you will overestimate yourself. You don't tend to estimate yourself as like an absolute expert, but you will underestimate yourself. And the, tend to slightly underestimate themselves, which is probably that, um, syndrome kicking in.

0:39:20 Thomas: It's not the classic thing, which is like you have to if you learn a certain amount about a thing, it makes you think, you know, too much because I feel like that's how it's presented online.

0:39:29 Jenessa: Kind of that. So like, If you know a little bit, you tend to think you know a lot. The Dunning-Kruger is not going to be, I believe, that because I watched one documentary, I know just as much as a NASA engineer. Most people will accurately describe, well, obviously, I don't think I'm going to go take a physics class and get a 100 on every test because I watched one space documentary. But they will, like if their actual performance on the test was going to be a 60, they'll be like, I don't know, probably like a 75. And then the people who got a 100 are like, I don't know, probably like 90.

0:40:12 Thomas: Gotcha. Well, my point was going to be sort of related to that, but the amount of harm that would would happen if let's say we've got just for fun 50 of the population that's gonna not know anything just just hypothetically if that population is like I'm an idiot I don't know anything they love I them they cause no harm there's nothing wrong with that there's anything wrong with not knowing something about a certain thing but if that same 50 is like no I got this then all of a sudden our society is ruined and you have RFK Jr. in a position over our health so big difference there we're worrisome and that happened pre-AI really all that like man

0:40:31 Jenessa: Pretty much. Exactly. Yeah. Yeah? Exactly. Yes. Right. So if this can make that even worse, we have a huge problem. We need people to know what they don't know and to accurately assess that. And yeah, it's not good.

0:41:03 Thomas: That's a that's really interesting like statement of our entire problem as a society like it's a I haven't quite thought about it in exactly those terms like normally Normally, you're just like, we just need to know more stuff or whatever. But it's like, no, we really need to have a more accurate idea of what we don't know.

0:41:23 Jenessa: Yeah. Yeah. Like I just, I don't know how my car works. I'm just going to the mechanic. And honestly, I know when I go to the mechanics, I'm probably getting raked over the coals.

0:41:32 Thomas: The American way.

0:41:33 Jenessa: So I I have a computer friend that I ask these questions to because I just I'm like, I can't read all this. I don't know, man. I got to just ask somebody like it's know, I don't want to act like I'm perfect or anything, but it's an important thing to do is like probably the most important thing you can do is embrace that like ignorance is the natural state of a human. I there's just so much I don't know. And it oh boy, I'm worried about this interfering with that very deeply. Um is uh some room for some hope here like mechanisms so this one I found really interesting um

0:42:24 Thomas: Another AI bot to tell you you're an idiot the whole time.

0:42:28 Jenessa: You know it might really help there is actually a brief aside there was one about sycophantic ai um was basically the more people interact with a a particularly sycophantic AI, the more that interacting with humans becomes exhausting because humans are not going to just tell you how amazing you are all the time.

0:42:49 Thomas: I have such the opposite. I have much the opposite of what you're saying. I was going to say, I fucking can't stand interacting with a human or AI that would be like that. And my comfort zone is the appropriate amount of disregard for me.

0:43:04 Jenessa: Yeah. This is of course all on average, but one of the findings was like the more people interact with that, the More that they find human interaction to be like exhausting because it's You're not agreeing with me on everything.

0:43:18 Thomas: They for sure know that that's the theory I mean I believe you but it's I know I know I'm a weirdo but my feeling would be like god I don't want to interact with anyone because I'm tired of this bot agree with me and what if someone else does that too is this what interaction is like fuck this no I think you're probably right

0:43:31 Jenessa: I'll have to deeper into that one. Yeah, I have not gone all the way through that one yet, but it's concerning. One This sets us up for some ideas about more responsible things we could do. The number one thing I'll say is this one was a study of college students, and we are not talking about children, so that helps. This is Learner's AI Dependence and Critical Thinking, the Psychological Mechanism of Fatigue and Social Buffering Role of AI Literacy. That's a whole lot of words. We got to define some stuff in here. Cognitive fatigue is the state of mental exhaustion characterized by diminished attention, slower cognitive processing, and reduced motivation during sustained mental effort. Effort so anybody who works a very intellectual job will have experienced this you don't have to go do physical labor to walk away tired like it is genuinely physically exhausting to sit there and do intellectual labor all day um this is measurable I've definitely had times where I'm like I need a break from my break I'm yeah

0:44:39 Thomas: I mean yeah podcasting and editing all day I'm sitting here. I'm not doing anything else, but holy shit. You know, like I can be exhausted.

0:44:48 Jenessa: You'll be like I needs a nap I'm done

0:44:49 Thomas: Well, and look at, you know, professional, uh, or poker players, you know, they're just sitting, but Holy shit. That's exhausting.

0:44:57 Jenessa: Yeah. It consumes like a ton of literal energy from your brain running. So, okay. Two possible issues. One is when you're getting so much information so easily, you're under -stimulated. It's just like, this is too easy. It's like boring. So cognitive underload, a term I had never heard before. Disinterested and fatigued. Other option is the sheer volume of information leading to overload, which is also fatiguing. So both directions, there's an optimal amount of information to have in front of you to not get overloaded or underloaded. Then we've got information literacy, your capacity to locate, evaluate, and apply information effectively. Effectively. They're also going to talk about critical thinking. Important to note here from the way they are talking about it in this study, critical thinking is not like an ability you have. I am a critical thinker the way that I like am a tall person. Critical thinking is a task you engage in. So we're not separating out like people who are critical thinkers and people who are not. We're talking about a state. I am in a state where I am engaged in critical thinking, or I have fallen into a state, where I'm not engaged in the best critical thinking because, possibly, I am cognitively fatigued. I'm exhausted. I have no energy left, and so my critical thinking is deeply diminished. Important to note that we're talking about different people. We are talking about people in different states. So, okay, what happens here? They first measured, this is super interesting to me, AI dependence, which they framed almost like addiction. They took a scale of social media addiction and reframed it for AI. It was like the amount of AI use people were engaging in and even like classic addiction -style questions of like, I have tried to reduce my AI use and found that I could not. That's like Addiction 101,

0:47:11 Jenessa: The like, Like I tried to reduce my alcohol use and found that I was struggling. They have this scale of how dependent somebody is or feels that they are on using the AI. And I think to some extent, it's just a metric to get how much are you using it? Because otherwise it's like really hard to measure. But also a measure of how much they feel like they must, which can measuring a lot of different things at once, right? Right? Are you using it, but also do you have the confidence to keep doing your tasks without it, that kind of thing. Measure that, they measure, they have a scale for information literacy. So essentially the people who are better at going through and figuring out this is a good source is a bad source. Or, you know, in this case, hopefully the AI gave me something and I'm like, that looks like bullshit, or I need to fact check that one, I do need to fact check this, I don't need to check that. Of the things that makes this complicated, they were like, okay, so on the one hand, you would expect that the people with more information literacy should just overall do better with this, that if you are able to use the AI and get the information and be better at knowing what's good information, what's bad information, that's great, right? But what if using this tool actually becomes exhausting when you have higher information literacy? Because if you're the kind of person who doesn't do that, and you're just like, whatever, everything it says is great. You don't put in cognitive effort.

0:48:48 Jenessa: Fully offloading. You're just like yep, everything you said is wonderful. And if you are getting hit with with a high volume of information, and you are really a stickler about information literacy, I'm going to check every source, that's overwhelming. That's exhausting, high cognitive fatigue. And then if the cognitive fatigue hits you, my critical thinking is going to go down, which means my information literacy is goingto go down and these are all going to interact, right? Essentially, that was the theory here. What can we find? Sure enough, the higher the AI AI dependency, the more critical thinking was reduced. So if people felt like they had to use it all the time, probably I think what's going on is low confidence that they don't feel good about doing their work without it. And so the critical thinking went down.

0:49:41 Thomas: There no, I was just going to check real quick. Was there no clarification of what they're using it for or was this for a specific thing? Right

0:49:49 Jenessa: These were all college students in China. They very specifically sought out university students um, at different levels.

0:49:57 Thomas: But are they chatting with it as a girlfriend or are they using it for science research you know oh

0:50:01 Jenessa: Good question. Let me, I think they were specifically focused on their use in school, but, um, here we go. 74 % of respondents indicated that their primary purpose for using AI was to support academic or work -related tasks, reflecting the growing integration of AI into students' learning routines. So 74 % say that's the primary purpose. Hope the other 26 % are not like, yes, my primary purpose is a girlfriend. Also like other things in their life. 74 % are saying primarily they're using it for academic purposes. Yeah, obviously the information literacy would be a little less important in some contexts than others. Generally we're finding, okay, The more they were on the AI, the more their critical thinking was going down. Interestingly, it was also increasing cognitive fatigue the more they were using it, which is decreasing the critical thinking. So we've got a little spiral happening. When they included cognitive fatigue as a mediator, the effect on critical thinking, the effect of the effect was reduced. The effect on thinking was reduced, but not eliminated. So essentially what they're saying is, why does AI reduce critical thinking? At least some of it is because of the cognitive fatigue. You're using it, you're getting overloaded with information, it's exhausting, and that is causing the critical thinking to go down. Information literacy had a buffering effect, so it goes the other way around. The AI dependence was less likely to reduce critical thinking for the individuals who report this higher information literacy.

0:51:45 Jenessa: So higher information literacy is reducing the cognitive fatigue, which is reducing the likelihood of lower critical thinking and reducing the dependence. And so how do we bring this all together? First of all, do you think your third grader has high information literacy? Not. That is a huge mediating factor to reduce the fatigue, to therefore reduce the problems with critical thinking. Thinking, and a lot of little kids, all them, kids don't have that. That's not a thing. So we have to teach that skill first, number one. And really, that skill needs to be strong first. Regardless, high dependence is probably not good for learning, period. It's going to produce cognitive fatigue, and cognitive fatigue in any context reduces your critical thinking skills. So just any time you're getting overwhelmed with information, you need to go take a break, period. There's more and less responsible ways to be using this. Getting constant use, constant overload is just going to make your brain not work as good. You're going to be in a less efficient energy state, essentially. Eventually, which lines up with a lot of what we would already suspect about needing breaks anyway from all information taking a break. To me, that lines up really well with the idea that New York, you know, granted, it's only New York City, but yeah, K through eight does not need to be doing this. They need the information literacy skill first to get anywhere near this. Just Just some studies. There's plenty more coming out every day. Guess I would say the way I approach a lot of this, there's really three, and this would apply to any tool use. It doesn't have to be an AI. There are three ways you could think about it. Do I use a calculator? I actually discovered the other day, I tried to do long division on a piece of paper,

0:53:48 Jenessa: And I think I kind of can't do it anymore. It's kind of gone. I really struggled for a minute, it. And I like learned that in elementary school. So why would I still use a calculator? I don't care. That's fine. Can lose that skill. That is okay.

0:53:48 Thomas: That's funny it's tricky gonna be my answer to a lot of things yeah

0:54:04 Jenessa: It is all right if it goes away. So with any tool, if it is a skill you don't mind losing or a skill that you have no interest in acquiring, yeah, then why are you worried about the skill being affected? Like, like, whatever, do what you want. Do I type instead of handwriting everything? Handwriting is actually the kind of skill that would be really, really hard to lose. There's no amount of typing where I'm going to pick up a pen someday and go, how does one write the letter A? It's the sort of muscle memory thing that I'm just going to do it. I'm gonna be okay. I might lose a little fluency, but I'll pick it right back up. Important thing to know that is a scientific question you don't just get to declare that you know you'll be fine and I'll never lose that skill that's actually a question for science to answer whether it's the type of skill that can or cannot be lost um so don't juste talk yourself into it but one example I'll use the tool because I can't lose the skill and the third one would be why do I I use a hammer to hammer in nails. There is a way for me to put a hammer into a piece of wood just using my body, I don't want to know about it.

0:55:18 Jenessa: So if you literally can't do it without the tool, well, then you learn it with the tool. And the first and only way you learn it is with the toolbox. So again, it's easier. That's not, well, I can't do it without the tool, but other people can, that's the kind of thing where you'd have to say, I need to acquire the skill first or choose not to acquire the skill. But if it truly cannot be done without the tool, of course. You want to acquire the skill without the tool, you have to start learning without the tool. There's no amount of learning it with the tool that is going to eventually teach you to do it without the tool. That's just not how it works. You're gonna have to do it without it so I hopefully that framework is helpful obviously there is a time and a place for tool use but it can absolutely impact your ability to do a skill without the tool and we need to think very carefully about that especially with the little ones bringing it into the classroom yeah

0:56:35 Thomas: Well this was a lot so I think I need to try to maybe get the roundup of all of it um overall I mean we went through a lot lot of the downsides potentially of using tools and and how much of the studies that you're referencing were actual ai versus like the more generalized one or does it matter like sorry generalized technology quite or does that matter hmm

0:56:55 Jenessa: The last two we talked about were very very specifically testing when somebody uses LLM specifically, they were looking at. Because, vague term. They were very specifically looking at modified versions of ChatGPT that they prompted specifically in the one study. And then in the other one, it was AI use, kind of self -defined by the user, but AI was the terminology used as opposed to like any old tool use obviously like using excel in your stats class is not what they're talking about yeah yes

0:57:28 Thomas: No I there's just some like general principles or you talked about in terms of tool use I wasn't sure how much of that was like the studies or if you were just using examples yeah

0:57:36 Jenessa: Yeah it's those last two specifically are two examples of the uh absolute deluge of however you say that word of uh studies that are coming out looking at the impacts of ai on learning and mental health and how we interact with each other and oh man there's like a really really cool and on creativity that it like sort of helps in the short term, but then falls apart later. It's very interesting what's going on. Yeah. So I would keep an eye on it and I don't know, just keep your kids away from everything as far as I can tell.

0:57:43 Thomas: That's right the

0:58:10 Jenessa: Just no screens and no, no social media and no people. I don't know.

0:58:10 Thomas: Amish or or whatever. Yeah. I mean, is that, so is that the overall judgment? I mean you don't want this stuff in classrooms. Was there, I mean have they, and I don't know, you know, if these studies will be done by big AI or whatever, but have they found like the version of it that could be okay or something? I know it's early obviously, but a lot of these feel a little bit general where it's like you have it and it's doing this or not, you're learning with it or not. It's like, okay, OK, what if there was one specific task it was doing to help learning? Maybe like could that be something that would be good at? Are we just, you kind of not quite there yet in terms of the research or what do you think?

0:58:23 Jenessa: For now at least. Yeah yeah. I guess I guess I would say, well, obviously it's going to depend. We need more research. Definitely all of that for sure. The takeaways I have right now are information literacy is a huge mediating factor to prevent some of the bad outcomes. And I struggle to imagine young children somehow acquiring that. That's just not a thing they show up with until later in life. They And it takes a long time to learn when comes to the things like I think that classroom study with the tutor is the most useful that the very specific, like tailored tutor compared to the tutor bot. But yes, the tutor bot compared to students who just did practice problems all by themselves with no teacher, they same, but they rated themselves as knowing more and performing better. So really, I think the most important thing would be education on metacognition.

0:59:34 Thomas: The tutor bot. Right. Yeah.

0:59:54 Jenessa: Teaching kids to accurately assess their own learning could seriously help with, and adults, we could all use this to better understand

0:59:54 Thomas: Yeah, yeah. We can't teach adults anything. So this is all the kids is where we're starting. We're starting for the next generation.

1:00:05 Jenessa: Well yeah what would help them to know did I actually learn this or did the robot tell me something and I feel like I learned it they need education on that to be better about using it.

1:00:18 Thomas: Now my kid's school. They do some amount of that. Like, and I, you know, we've had some form of this forever where it's the software where it's like, all right, you do, we're going to have this part of the class think it's, they do Some reading on it. And they answer questions. Sometimes it's just a little software that's like different school. I mean, they got them there for six hours. They got to find crap to do. But, but when you brought up the thing, you just talked about that study, it almost feels like that would apply to that really. I don't know why that would be any any better, but I guess it depends on if it makes you, does the tutor make you specifically think you know more versus if it's just like you're little software thingy that's like, you you or doing math problems at you, you know? Like, the basic software version of that more like the kids doing the problems by themselves or more like the tutor thing probably?

1:01:10 Jenessa: I think it's the overall problem of like, when you have to fight for it, you accurately assess your learning, or even sometimes you under, like, your confidence. You think like, oh, I didn't learn anything, as we saw from that one study of like the active learning, right? You kind of got fed the information or you got hints, you're overestimating yourself. So I think that part is more inherent to the AI or Google. We saw it with that Google study, like training how to to understand the difference between being given information versus actually like fighting for it so that it's stuck in your head now. Don't have an answer on how to teach that skill. I wish I did.

1:01:58 Thomas: I guess that would then invite the question of like, Like, OK, sounds like we need a better tutor bot that actually just quizzes you more and, you know, makes you work for the knowledge a little more. Does that come down to the methodology of tutor bot?

1:02:10 Jenessa: Maybe like the feedback of it, like a way to better tell you you don't understand. Just because I told you the answer does not mean you get it.

1:02:18 Thomas: That's it. We just need roast bot in every classroom,

1:02:21 Jenessa: Oh, probably not.

1:02:21 Thomas: Roasting all the kids saying they don't know shit.

1:02:24 Jenessa: Probably not.

1:02:25 Thomas: No, I'm just kidding. All right. Well, that is a lot to think about. Any other takeaway messages from this that you wanted to get across?

1:02:29 Jenessa: I think just overall we need more research on this and um probably yeah not research funded by the person who has a financial interest in you using it great

1:02:42 Thomas: All right, I will have ChatGPT. She runs some experiments. Good the war crimes that's a solid quote right