WEBVTT 00:00:00.000 --> 00:00:02.319 Welcome to the Deep Dive. We're here to cut through 00:00:02.319 --> 00:00:04.620 the noise around the latest research and find 00:00:04.620 --> 00:00:06.799 the insights that, well, actually matter for 00:00:06.799 --> 00:00:08.740 you. Today, we're digging into something that's 00:00:08.740 --> 00:00:12.919 really been making waves this July 2025. It's 00:00:12.919 --> 00:00:16.039 being called the MIT article. It's a new preprint 00:00:16.039 --> 00:00:19.260 on rxev .org. Right. That's the platform where 00:00:19.260 --> 00:00:21.699 researchers put papers out early before the whole 00:00:21.699 --> 00:00:24.120 peer review thing. Exactly. And our goal today 00:00:24.120 --> 00:00:26.160 isn't just to give you the summary. We want to 00:00:26.160 --> 00:00:28.739 explore what this paper really found. especially 00:00:28.739 --> 00:00:32.460 this kind of overlooked idea about how we learn 00:00:32.460 --> 00:00:34.880 and how we really own what we know. We're going 00:00:34.880 --> 00:00:37.060 past the headlines here. Mm -hmm. Looking for 00:00:37.060 --> 00:00:39.579 what's truly useful. So let's kick things off. 00:00:39.759 --> 00:00:42.200 This MIT article, what's the basic setup? It 00:00:42.200 --> 00:00:44.340 compares essays, right? Written using different 00:00:44.340 --> 00:00:46.719 tools. That's the core of it. People wrote essays 00:00:46.719 --> 00:00:49.460 using either chat GPT or just regular web search 00:00:49.460 --> 00:00:51.579 or, you know, only their own brains. It's a pretty 00:00:51.579 --> 00:00:54.420 long paper, but definitely interesting. Oh, absolutely. 00:00:55.159 --> 00:00:57.460 And it's obviously got educators talking. It 00:00:57.460 --> 00:01:00.700 flags some cautions about using LLM's large language 00:01:00.700 --> 00:01:04.099 models like chat GPT in the classroom. But it's 00:01:04.099 --> 00:01:06.280 also fair to say the study has its limitations 00:01:06.280 --> 00:01:08.959 scope -wise. OK, so methodology, how do they 00:01:08.959 --> 00:01:11.540 actually do it? They had 54 adults. Yep, 54 adults 00:01:11.540 --> 00:01:13.900 each wrote three essays switching up the tools. 00:01:14.340 --> 00:01:17.299 And there was this clever bit in the design wasn't 00:01:17.299 --> 00:01:19.849 there, about the fourth essay. Right. So if you 00:01:19.849 --> 00:01:22.069 started without tech for your first essays, you 00:01:22.069 --> 00:01:24.730 then rewrote one of those earlier essays, but 00:01:24.730 --> 00:01:27.810 this time with ChatGBT. Oh, OK. And the reverse. 00:01:28.030 --> 00:01:30.349 Exactly. If you started with the LLMs, your fourth 00:01:30.349 --> 00:01:33.170 essay was done using only your brain, revisiting 00:01:33.170 --> 00:01:35.730 a topic you'd used AI for before. And they had 00:01:35.730 --> 00:01:39.450 humans and AI grading these essays. And brainwave 00:01:39.450 --> 00:01:41.650 measurements? Correct. Both judged the essays. 00:01:42.150 --> 00:01:44.189 And yeah, the brainwave data is actually a huge 00:01:44.189 --> 00:01:46.599 chunk of the paper. We're talking like, Many 00:01:46.599 --> 00:01:48.799 many pages of graphs the source even calls it 00:01:48.799 --> 00:01:51.079 fascinating reading fascinating. Yeah because 00:01:51.079 --> 00:01:53.760 it shows Visually how our brains are working 00:01:53.760 --> 00:01:56.239 right the neural activity changing with the tools 00:01:56.239 --> 00:01:59.099 precisely That's often where researchers expect 00:01:59.099 --> 00:02:02.319 the big aha moments in that kind of data You'd 00:02:02.319 --> 00:02:04.620 think so all those graphs all that detailed brain 00:02:04.620 --> 00:02:08.639 activity that must be the main story But digging 00:02:08.639 --> 00:02:11.960 into it we found something well pretty surprising. 00:02:12.039 --> 00:02:14.240 Mm -hmm. It seems the most important finding 00:02:14.240 --> 00:02:16.759 the real kicker wasn't in those brain scans. 00:02:17.340 --> 00:02:19.419 He was actually given the least space in the 00:02:19.419 --> 00:02:22.050 paper. Here's where it gets really interesting. 00:02:22.069 --> 00:02:25.030 It really is. Because that crucial insight, it 00:02:25.030 --> 00:02:26.509 came from the short interviews they did with 00:02:26.509 --> 00:02:29.370 the writers, specifically asking about their 00:02:29.370 --> 00:02:31.849 sense of ownership over the essays they produced. 00:02:31.949 --> 00:02:34.050 Just from brief interviews, not the scans. Just 00:02:34.050 --> 00:02:36.330 the interviews. And this idea, this ownership 00:02:36.330 --> 00:02:38.310 piece, even though it gets minimal ink in the 00:02:38.310 --> 00:02:40.930 actual paper, our sources point to it as maybe 00:02:40.930 --> 00:02:43.250 the most vital thing we can foster in students, 00:02:43.509 --> 00:02:46.810 or really in anyone learning anything. Wow. That's 00:02:46.810 --> 00:02:48.889 quite a twist for a scientific paper, isn't it? 00:02:49.870 --> 00:02:52.509 If the brainwaves weren't the main story and 00:02:52.509 --> 00:02:55.110 it was these interviews about ownership, what 00:02:55.110 --> 00:02:57.490 does that actually mean? For you, the listener, 00:02:58.009 --> 00:03:00.810 why is this idea of knowledge ownership so critical? 00:03:01.129 --> 00:03:04.250 Well, let's unpack that. In education, a major 00:03:04.250 --> 00:03:06.370 goal is always transfer. You learn something 00:03:06.370 --> 00:03:08.550 in class. Right, like a math concept or a writing 00:03:08.550 --> 00:03:11.810 skill. Exactly. And transfer means you can actually 00:03:11.810 --> 00:03:14.590 use that knowledge or skill outside the classroom. 00:03:14.710 --> 00:03:17.189 In a different situation, maybe at work, maybe 00:03:17.189 --> 00:03:20.520 just in life. For that transfer to happen effectively, 00:03:20.900 --> 00:03:23.659 our sources say you need two things, competence 00:03:23.659 --> 00:03:26.099 and ownership. Okay, let's take confidence first. 00:03:26.340 --> 00:03:28.719 That sounds fairly straightforward. It mostly 00:03:28.719 --> 00:03:30.759 is. Competence is basically knowing what you 00:03:30.759 --> 00:03:32.759 need to know and being able to do what you need 00:03:32.759 --> 00:03:34.900 to do. Like knowing the facts, being able to 00:03:34.900 --> 00:03:37.800 perform the steps. Pretty much. It's a brain 00:03:37.800 --> 00:03:39.819 activity, sure, but your environment matters 00:03:39.819 --> 00:03:42.479 too. And it grows, you know, as you learn more, 00:03:42.659 --> 00:03:44.719 gather more info. Got it. So that's competence. 00:03:45.340 --> 00:03:47.180 What about the other piece, the ownership part? 00:03:47.180 --> 00:03:50.719 That sounds... Deeper. It is deeper. Ownership 00:03:50.719 --> 00:03:52.280 is when that knowledge isn't just something you 00:03:52.280 --> 00:03:55.280 can recall, but it's truly deeply integrated 00:03:55.280 --> 00:03:58.199 into your thinking. It literally changes how 00:03:58.199 --> 00:04:00.719 you perceive the world differently. You start 00:04:00.719 --> 00:04:02.599 seeing connections you didn't see before. So 00:04:02.599 --> 00:04:05.219 it's not just having the information, it's internalizing 00:04:05.219 --> 00:04:08.419 it, making it part of you. Exactly. When you 00:04:08.419 --> 00:04:10.500 own knowledge, you become critical of it. You 00:04:10.500 --> 00:04:12.759 can judge if it's enough to solve a problem. 00:04:13.039 --> 00:04:15.960 You can adapt it. You can even generate new knowledge 00:04:15.960 --> 00:04:18.720 based on what you already own. It's active. It's 00:04:18.720 --> 00:04:21.579 transformative. It's not just storage and retrieval. 00:04:21.639 --> 00:04:23.560 OK, that makes sense. It's like the difference 00:04:23.560 --> 00:04:25.879 between knowing the recipe and truly understanding 00:04:25.879 --> 00:04:28.540 cooking. That's a great analogy. And when you 00:04:28.540 --> 00:04:30.620 look at generative AI through this lens, this 00:04:30.620 --> 00:04:34.779 ownership lens, it becomes clear that AI is no 00:04:34.779 --> 00:04:37.339 different from any other instructional tool or 00:04:37.339 --> 00:04:39.860 method. Wait, say that again. No different. That's 00:04:39.860 --> 00:04:42.959 a big claim. People are treating AI as this completely 00:04:42.959 --> 00:04:46.199 revolutionary, maybe dangerous thing in education. 00:04:46.439 --> 00:04:48.980 I understand why, but from the perspective of 00:04:48.980 --> 00:04:51.839 ownership, the tool itself isn't inherently good 00:04:51.839 --> 00:04:54.680 or bad. So the focus shifts. Completely. It's 00:04:54.680 --> 00:04:58.019 not about banning chat GPT or requiring it. It's 00:04:58.019 --> 00:05:00.600 about how it affects that sense of ownership. 00:05:00.939 --> 00:05:03.839 Precisely. Can generative AI be used to whip 00:05:03.839 --> 00:05:06.139 up an essay where the student feels zero ownership 00:05:06.139 --> 00:05:09.319 and learns very little? Yeah, we see that. But, 00:05:09.540 --> 00:05:11.240 and this is the crucial point, that same lack 00:05:11.240 --> 00:05:13.939 of ownership can happen just as easily with, 00:05:13.939 --> 00:05:16.720 say, traditional digital information sources. 00:05:17.329 --> 00:05:20.370 Think copy pasting from Wikipedia without really 00:05:20.370 --> 00:05:22.629 understanding it. Right, or just stringing together 00:05:22.629 --> 00:05:25.209 quotes. Or even, as the source points out, with 00:05:25.209 --> 00:05:28.250 learning done with a brain only. How so? Think 00:05:28.250 --> 00:05:31.490 about cramming for an exam. You memorize facts. 00:05:31.769 --> 00:05:34.649 You might even ace the test. Yeah. Competence, 00:05:34.769 --> 00:05:37.120 right? Uh -huh. But a week later, It's gone. 00:05:37.319 --> 00:05:39.300 You never owned it. It wasn't integrated. It 00:05:39.300 --> 00:05:41.839 was just temporary recall. OK, I see. So the 00:05:41.839 --> 00:05:44.040 tool isn't the issue, the outcome, the ownership 00:05:44.040 --> 00:05:47.639 is. That's the core takeaway. If students, or 00:05:47.639 --> 00:05:50.600 again, anyone learning, leaves a course or finishes 00:05:50.600 --> 00:05:53.000 a learning experience without a sense of ownership 00:05:53.000 --> 00:05:55.100 of the curriculum, then they have not learned 00:05:55.100 --> 00:05:57.579 what they needed to learn. Regardless of the 00:05:57.579 --> 00:06:00.319 grade, regardless of the tools used. Exactly. 00:06:00.480 --> 00:06:03.240 The technology is secondary to that deeper goal 00:06:03.240 --> 00:06:06.029 of creating owned knowledge, which Really, this 00:06:06.029 --> 00:06:07.870 raises an important question, I think, for everyone 00:06:07.870 --> 00:06:10.490 listening. If the real measure of learning isn't 00:06:10.490 --> 00:06:13.110 just competence, but this deeper sense of ownership, 00:06:13.870 --> 00:06:15.850 how does that challenge your own approach? How 00:06:15.850 --> 00:06:18.709 you learn things or even how you figure out if 00:06:18.709 --> 00:06:20.550 you've really learned something. That's definitely 00:06:20.550 --> 00:06:22.470 something that you want as you go through your 00:06:22.470 --> 00:06:25.410 day, taking in information from us, from articles, 00:06:25.449 --> 00:06:28.170 whatever. Think about that. What does it actually 00:06:28.170 --> 00:06:30.790 take to make knowledge your own, not just to 00:06:30.790 --> 00:06:33.410 know it? What really stands out to way from this 00:06:33.410 --> 00:06:34.110 deep dive today?