1 00:00:00,040 --> 00:00:04,040 I really am beginning to dislike the term AI tutors. 2 00:00:04,440 --> 00:00:07,600 My problem is that we say human in the loop. 3 00:00:09,000 --> 00:00:12,360 There's a human slowing down that feedback loop because one 4 00:00:12,360 --> 00:00:17,680 of the advantages I see that AI tutors could have is that 5 00:00:17,680 --> 00:00:20,000 closing this feedback loop to pupils. 6 00:00:20,400 --> 00:00:24,400 We have a problem with our education system where you just 7 00:00:24,400 --> 00:00:27,120 go through the school years and what are the gaps? 8 00:00:27,120 --> 00:00:29,920 The gaps just get wider and wider and wider. 9 00:00:30,240 --> 00:00:34,080 And if there is a hope, or if there's a way that we, each 10 00:00:34,880 --> 00:00:40,240 pupil, can have access to a pedagogically rich AI tutor that 11 00:00:40,240 --> 00:00:43,960 supports them in that consolidation, well, surely 12 00:00:43,960 --> 00:00:45,320 that's something we should explore. 13 00:00:45,960 --> 00:00:48,120 I think that what makes it so difficult is that we just don't 14 00:00:48,200 --> 00:00:53,040 know the proportion of errors that teachers make. 15 00:01:13,950 --> 00:01:17,070 Welcome to Thinking Deeply about AI for Schools. 16 00:01:17,070 --> 00:01:20,790 I'm James Rabin, and joining me, as always, is Neil Almond. 17 00:01:20,790 --> 00:01:23,790 We've got a fascinating topic to dive into today. 18 00:01:24,390 --> 00:01:26,590 Absolutely. And for those just joining us, 19 00:01:26,590 --> 00:01:30,070 this is the podcast where we explored the rapidly evolving 20 00:01:30,070 --> 00:01:33,430 intersection of artificial intelligence and education. 21 00:01:34,040 --> 00:01:37,040 Every month we're going to pose the big questions, examine 22 00:01:37,040 --> 00:01:39,720 what's changing in the AI and education landscape, and just 23 00:01:39,720 --> 00:01:42,200 try to make sense of it all and what that means for education, 24 00:01:42,200 --> 00:01:44,160 the pupils, and the future of learning. 25 00:01:44,160 --> 00:01:46,200 And. We're not here to give you all 26 00:01:46,280 --> 00:01:48,880 of the answers because honestly, we're just trying to figure this 27 00:01:48,880 --> 00:01:51,200 out together. Instead, we want to create a 28 00:01:51,200 --> 00:01:54,320 space where we can think critically, question 29 00:01:54,320 --> 00:01:57,400 assumptions, and explore those possibilities. 30 00:01:57,480 --> 00:02:01,080 Each episode we explore one topic, 1 research article or one 31 00:02:01,120 --> 00:02:04,080 AI update and ask the ultimate question, what does this mean 32 00:02:04,080 --> 00:02:06,600 for education? So this week we're going to go 33 00:02:06,600 --> 00:02:10,520 into this world that's very moving very quickly from product 34 00:02:10,520 --> 00:02:14,920 demos to policy conversations, and that is about AI tutors. 35 00:02:15,760 --> 00:02:18,280 And the key thing here is, are AI tutors generally in 36 00:02:18,280 --> 00:02:21,560 breakthroughs and learning or are we moving faster than this 37 00:02:21,720 --> 00:02:24,200 evidence? So in this episode, which I can 38 00:02:24,680 --> 00:02:28,080 envisage already that will become several episodes, we'll 39 00:02:28,080 --> 00:02:32,480 explore what AI tutors are, why there is so much excitement 40 00:02:32,560 --> 00:02:35,680 about them and whether actually this is a hype or whether they 41 00:02:36,000 --> 00:02:40,840 there is research out there and examples that this is concrete. 42 00:02:41,240 --> 00:02:44,280 But we'll also consider some of those deeper concerns around 43 00:02:44,280 --> 00:02:47,840 buyer safeguarding, cheating, the cognitive offloading and 44 00:02:47,840 --> 00:02:50,040 what this happens to 1 to 1 support. 45 00:02:50,320 --> 00:02:53,520 And because this is not just a story about shiny new tools, it 46 00:02:53,920 --> 00:02:57,360 is really a story about what we think tutoring is, what good 47 00:02:57,360 --> 00:03:00,680 learning looks like, and whether AI can genuinely support people 48 00:03:00,680 --> 00:03:03,600 to think more deeply, rather than just helping them get to 49 00:03:03,600 --> 00:03:07,280 that answer faster. Now, one of the things that I've 50 00:03:07,280 --> 00:03:09,640 been listening to lots of podcasts and my journeys up and 51 00:03:09,640 --> 00:03:14,600 down the M5 and M6, I think I did the whole of the M5 the 52 00:03:14,600 --> 00:03:18,800 other day, is this concept of what an AIT 2 is. 53 00:03:18,800 --> 00:03:21,760 I wouldn't throw in this span this term around. 54 00:03:21,760 --> 00:03:25,280 And I think it's important to distinguish what it could be 55 00:03:25,720 --> 00:03:29,800 because with all the research we're discussing and we'll 56 00:03:29,800 --> 00:03:32,960 share, bits and pieces have come out recently around Harford and 57 00:03:32,960 --> 00:03:36,280 others. One of the things you may see is 58 00:03:36,280 --> 00:03:41,200 like it's a almost a GPT on ChatGPT where you're using an 59 00:03:41,240 --> 00:03:46,720 open source product and you're asking it questions. 60 00:03:47,600 --> 00:03:49,640 So that's one idea. You've got the open large 61 00:03:49,640 --> 00:03:53,880 language models such as Gemini or ChatGPT using pupils for 62 00:03:53,880 --> 00:03:57,320 homework, for help, not much education to sign behind them. 63 00:03:57,360 --> 00:04:00,240 The other end, you kind of got these old intelligent tutoring 64 00:04:00,240 --> 00:04:02,960 systems that very subject specific, structured and signed 65 00:04:02,960 --> 00:04:06,080 to adapt to questions the student use. 66 00:04:06,080 --> 00:04:09,920 But also you've almost got this middle ground as well where 67 00:04:09,920 --> 00:04:13,880 they're purpose built. So things like Google who are 68 00:04:13,880 --> 00:04:18,800 looking at this learn LM or what Khan Academy have done as well. 69 00:04:19,040 --> 00:04:24,640 And even some of these models are being used with a human in 70 00:04:24,640 --> 00:04:28,000 the loop coined the phrase of that almost. 71 00:04:28,000 --> 00:04:32,800 So when we say AI tutor, you've got to have quite an open mind 72 00:04:32,800 --> 00:04:36,080 of what that looks like in practicality or especially when 73 00:04:36,080 --> 00:04:41,440 you're deploying them or using them as well because they are a 74 00:04:41,440 --> 00:04:44,800 massive range. It's quite a fast concept behind 75 00:04:44,800 --> 00:04:46,480 it. So it could be quite a carefully 76 00:04:46,680 --> 00:04:49,480 designed Socratic learning tool that's particular for maths. 77 00:04:49,480 --> 00:04:51,600 So it's got the whole knowledge graph behind it. 78 00:04:51,920 --> 00:04:54,720 Or sometimes it could be people thinking, oh, it's just a 79 00:04:54,720 --> 00:04:57,080 channel jackpot that people are using online. 80 00:04:58,120 --> 00:05:01,120 Or it could be something that's actually just very structured, 81 00:05:01,120 --> 00:05:05,720 very controlled inputs where a pupil actually doesn't chat to 82 00:05:05,720 --> 00:05:08,560 it. It they answer, they ask, it 83 00:05:08,560 --> 00:05:12,360 asks the students questions, they answer them and it will 84 00:05:12,600 --> 00:05:15,760 then adapt in the background. The AI is almost in the 85 00:05:15,760 --> 00:05:17,640 background adapting to it as well. 86 00:05:17,640 --> 00:05:21,320 So we're not in this episode. We're kind of not going into 87 00:05:21,320 --> 00:05:24,000 those, but just be aware that's what AI tutoring is. 88 00:05:24,000 --> 00:05:27,080 And we'll probably say that at some point. 89 00:05:27,080 --> 00:05:30,360 And I know Neil Stone and podcasts done an article on this 90 00:05:30,360 --> 00:05:33,920 before around what makes a good AI tutor. 91 00:05:33,920 --> 00:05:37,320 And I think that distinguish between what an AI tutor is and 92 00:05:37,880 --> 00:05:41,800 that technical kind of element of it is is important. 93 00:05:41,920 --> 00:05:46,360 I really am beginning to dislike the term AI tutors for the 94 00:05:46,360 --> 00:05:53,120 reasons that you have said. And you will see, yeah, a 95 00:05:53,120 --> 00:05:58,640 research paper or one of the tech AI Bros on Twitter or 96 00:05:58,640 --> 00:06:02,280 LinkedIn, whatever it is and it will say, you know, AI tutor 97 00:06:02,280 --> 00:06:03,680 fundamentally changes everything. 98 00:06:03,680 --> 00:06:07,320 And then you sort of you have to dig a little bit deeper, as you 99 00:06:07,320 --> 00:06:10,240 say, just to kind of work out or what are we talking about when 100 00:06:10,240 --> 00:06:15,640 we're talking about an AI tutor here, which is difficult because 101 00:06:15,640 --> 00:06:17,800 I think there are some that are going to be pretty useful. 102 00:06:17,800 --> 00:06:21,720 And I think there are some that are or some uses of the term of 103 00:06:21,720 --> 00:06:26,280 AI tutor when it is more just the right kids, like we trust 104 00:06:26,280 --> 00:06:30,720 you to use ChatGPT as a tutor going to be less helpful. 105 00:06:30,720 --> 00:06:36,480 So I am all for really kind of distinguishing these different 106 00:06:36,480 --> 00:06:40,320 types of tutors that use technology to support them, 107 00:06:40,320 --> 00:06:44,560 whether that be sort of, you know, complete generative AI 108 00:06:44,560 --> 00:06:48,000 chat bot free. You rely on the the pupil to do 109 00:06:48,000 --> 00:06:51,160 the right thing versus ones that have far more sort of guard 110 00:06:51,160 --> 00:06:56,920 rails that won't let you deviate the conversation much through to 111 00:06:56,920 --> 00:07:00,120 ones, as you say, where perhaps the learner does have a little 112 00:07:00,120 --> 00:07:04,200 bit more control about. This isn't like Google learn LM 113 00:07:04,200 --> 00:07:07,560 where you write, you know, you say the kind of things that 114 00:07:07,560 --> 00:07:10,960 you're interested in and it will kind of create the problems 115 00:07:10,960 --> 00:07:14,360 based on those things. Whole other conversation to have 116 00:07:14,360 --> 00:07:17,520 whether that is the right thing to do for kids or what not. 117 00:07:17,520 --> 00:07:22,040 But I think we really sort of need more specific terms to help 118 00:07:22,040 --> 00:07:24,400 us sort of really clarify what we mean when we're saying AI 119 00:07:24,400 --> 00:07:26,760 tutors because they say for every, we're going to talk about 120 00:07:26,760 --> 00:07:28,960 one bit of research now. But for every bit of research 121 00:07:28,960 --> 00:07:33,760 that I see around AI tutors and them being effective, I can 122 00:07:33,760 --> 00:07:36,600 point to you with another research article that says these 123 00:07:36,720 --> 00:07:41,360 AI AI tutors are not effective, but you should never give them 124 00:07:41,360 --> 00:07:45,400 to to children. And then you sort of have to go 125 00:07:45,400 --> 00:07:47,360 in to say, you have to look at the difference between what we 126 00:07:47,360 --> 00:07:51,120 mean when we say AI tutors to really sort of get that, dare I 127 00:07:51,120 --> 00:07:55,240 say nuance of what is what's happening in these studies. 128 00:07:56,160 --> 00:07:57,840 Yeah. And one of those studies you 129 00:07:57,840 --> 00:08:03,160 were talking about there was the Google Learn LM study with Ed as 130 00:08:03,160 --> 00:08:06,280 well and it was an RPT. So you know, the quality of this 131 00:08:06,280 --> 00:08:08,920 is important as well. And I think when you're looking 132 00:08:08,920 --> 00:08:12,280 at this research, we're we're professionals in education, but 133 00:08:12,280 --> 00:08:15,560 we don't not necessarily research specialists as well. 134 00:08:16,080 --> 00:08:21,160 And so the quality that we look at this, these research and how 135 00:08:21,160 --> 00:08:25,640 they measure are important are really, really key elements 136 00:08:25,640 --> 00:08:28,360 behind this. And I know charter college and 137 00:08:28,800 --> 00:08:33,000 the EEF do put guidance out there around actually what what 138 00:08:33,000 --> 00:08:35,960 quality research looks like. But the Google one's interesting 139 00:08:35,960 --> 00:08:38,400 because it looked at chat based maths tutoring. 140 00:08:39,200 --> 00:08:42,360 Now, mass tutoring I think is a really interesting one to pick 141 00:08:42,360 --> 00:08:46,200 on, mainly because the subject matter is hierarchical. 142 00:08:46,640 --> 00:08:49,640 You need to know X before you can go to Y and move it down. 143 00:08:49,640 --> 00:08:52,480 So if you don't have understanding of basic place 144 00:08:52,480 --> 00:08:56,280 value and number and then multiplication and division, 145 00:08:56,720 --> 00:09:00,040 actually some of the fraction work that you will get to down 146 00:09:00,040 --> 00:09:02,800 the line and algebra will actually there will be gaps in 147 00:09:02,800 --> 00:09:04,120 the knowledge if you don't have that. 148 00:09:04,120 --> 00:09:06,240 So it's a really interesting way of looking at it. 149 00:09:07,080 --> 00:09:11,040 But essentially the the research, and I'll cipher some 150 00:09:11,040 --> 00:09:13,520 of a bit, some of it often put some of the key points out 151 00:09:13,520 --> 00:09:15,800 there, but feel free to go out and have a look at yourself. 152 00:09:15,800 --> 00:09:20,400 But it's this kind of idea where pupils receive short, short 153 00:09:20,400 --> 00:09:24,520 sessions with the learn LM and it said there were 5.5 154 00:09:24,520 --> 00:09:28,000 percentage points more likely to solve novel problems than pupils 155 00:09:28,000 --> 00:09:30,200 who worked with human tutors alone. 156 00:09:30,680 --> 00:09:33,840 Now with this model, what was really important and with this 157 00:09:33,840 --> 00:09:39,920 research is that every message that was sent out by the AI was 158 00:09:39,920 --> 00:09:43,840 then checked by a human. So therefore there were reported 159 00:09:43,840 --> 00:09:47,720 very low factual errors, nought .1% of messages. 160 00:09:47,960 --> 00:09:51,640 So it was in terms of quite reliable and not all very 161 00:09:51,640 --> 00:09:54,560 reliable you can say, and we'll probably come on to that debate 162 00:09:54,560 --> 00:09:56,120 later. We've already talked about it 163 00:09:56,320 --> 00:09:59,240 before, we've started recording, but it makes it one and more 164 00:09:59,240 --> 00:10:01,240 practical and policy related studies. 165 00:10:01,240 --> 00:10:03,360 Now those in the UK, this would be really interesting because it 166 00:10:03,360 --> 00:10:07,920 was 165 students across 5 secondary schools between years 167 00:10:07,920 --> 00:10:10,680 9:00 and 10:00. And they want to replicate this 168 00:10:10,680 --> 00:10:15,360 and want to do this again this year I think in the US as wow. 169 00:10:16,000 --> 00:10:20,560 And it meant that was a really that they approved these tutors, 170 00:10:20,560 --> 00:10:24,560 these human tutors approved 76.4% of the messages with 171 00:10:24,640 --> 00:10:28,760 basically no edits. That's a huge point of view 172 00:10:29,040 --> 00:10:32,240 behind it. But I think my problem with 173 00:10:32,360 --> 00:10:38,880 almost this is that we say human in the loop, but it's a human 174 00:10:38,880 --> 00:10:42,760 slowing down that feedback loop. Because one of the advantages I 175 00:10:42,760 --> 00:10:48,440 see that AI tutors could have is that closing this feedback loop 176 00:10:48,440 --> 00:10:52,320 to pupils. And if you are adding that human 177 00:10:52,320 --> 00:10:55,280 in, actually are you more efficient? 178 00:10:56,720 --> 00:10:59,560 And it's a quandary I kind of question all the time. 179 00:10:59,560 --> 00:11:01,640 I don't know. Thinking that maybe right now it 180 00:11:01,640 --> 00:11:06,480 probably is a sort of a trade off potentially worth having. 181 00:11:07,320 --> 00:11:10,600 That's the sort of gut feeling that I have based on a, you 182 00:11:10,600 --> 00:11:17,120 know, this sort of technology or this, you know, generative 183 00:11:17,520 --> 00:11:19,760 artificial intelligence and tutoring. 184 00:11:19,760 --> 00:11:21,680 It is certainly new. And I think it's, you know, the 185 00:11:21,680 --> 00:11:24,720 right thing that a human safeguard right now is at that 186 00:11:24,720 --> 00:11:29,600 point where they can intervene if need be. 187 00:11:29,600 --> 00:11:31,520 Obviously, there's no point in closing that feedback loop 188 00:11:32,000 --> 00:11:35,520 quicker if the the feedback is wrong, That's you could then 189 00:11:35,520 --> 00:11:38,920 potentially sort of multiply the errors rather than fix any 190 00:11:38,920 --> 00:11:44,880 errors without human oversight. I also think it's again, we 191 00:11:45,160 --> 00:11:46,760 obviously weren't involved in the study at all. 192 00:11:46,760 --> 00:11:49,920 We might have to get Craig on to talk about it. 193 00:11:49,920 --> 00:11:52,320 Maybe he'll he'll want to do that after we've been on his 194 00:11:52,320 --> 00:11:55,480 podcast a few times. Maybe he'll come come. 195 00:11:55,800 --> 00:11:57,520 It's an open invitation, creative you're listening, 196 00:11:57,600 --> 00:12:01,760 you're free to come on. But I would imagine like it's I 197 00:12:01,760 --> 00:12:08,240 can't imagine that the feedback that the language, large 198 00:12:08,240 --> 00:12:11,520 language model, whatever they might be using is going to, you 199 00:12:11,520 --> 00:12:18,200 know, provide so much recent feedback that, you know, a human 200 00:12:18,200 --> 00:12:21,440 and to be part of the to work with Edie, you know, you need to 201 00:12:21,440 --> 00:12:25,280 be able to teach mathematics up to, you know, secondary GCSE 202 00:12:25,400 --> 00:12:27,840 comfortably. So we're talking about not as 203 00:12:27,840 --> 00:12:30,960 humans, but experts in their subject domain as well. 204 00:12:31,240 --> 00:12:36,760 I can't imagine they would take too long to cipher through those 205 00:12:36,760 --> 00:12:41,080 messages and flag them whether they're right or wrong or, you 206 00:12:41,080 --> 00:12:45,280 know, having to correct them in any way, shape or form. 207 00:12:45,280 --> 00:12:48,160 So I think right now it's it's the right thing to do. 208 00:12:48,160 --> 00:12:54,560 Will we need it all the time? Probably not, but I think it 209 00:12:54,560 --> 00:12:56,320 might be one of those things going back to what you said. 210 00:12:56,320 --> 00:13:00,840 And when you look at the history of of tutors, I'm going to call 211 00:13:00,840 --> 00:13:05,200 it tutors with through technology because it's always 212 00:13:05,200 --> 00:13:07,120 maths, right? You go back to like the old 213 00:13:07,120 --> 00:13:09,600 intelligent tutoring systems from like the 80s. 214 00:13:10,200 --> 00:13:14,640 Like I think about like Alex is 1 like famous one that's been 215 00:13:14,640 --> 00:13:17,840 out there for quite some time. I think like McGraw Gill or 216 00:13:17,840 --> 00:13:20,880 something own it now, but it's always maths. 217 00:13:20,880 --> 00:13:22,200 Why? Because it's hierarchical. 218 00:13:22,200 --> 00:13:26,320 It's very simple. If they get a, they can go to BI 219 00:13:26,320 --> 00:13:28,080 think. We then sort of look at more 220 00:13:28,840 --> 00:13:34,040 different subjects like humanities, English literature, 221 00:13:34,920 --> 00:13:37,160 where sort of we've got that more sort of, you know, it's 222 00:13:37,640 --> 00:13:40,920 wider rather than deeper in terms of that knowledge 223 00:13:40,920 --> 00:13:43,120 structure. I would think you would need far 224 00:13:43,120 --> 00:13:48,400 more humans overlooking what that chat bot was providing just 225 00:13:48,400 --> 00:13:52,040 because there's so much more margin for error I would have 226 00:13:52,040 --> 00:13:54,600 thought based on the training data and the way it could 227 00:13:54,600 --> 00:14:01,800 potentially could go. It comes back to this point that 228 00:14:02,280 --> 00:14:05,520 it needs to know the curriculum you're teaching and it needs to 229 00:14:05,520 --> 00:14:11,640 know where it goes. So Math Academy is a popular one 230 00:14:11,640 --> 00:14:16,560 in the US at the moment, and they've recently released a book 231 00:14:17,600 --> 00:14:20,280 that I found APDF on LinkedIn about which is all about the 232 00:14:20,280 --> 00:14:22,680 cognitive science. They've gone through it all and 233 00:14:22,680 --> 00:14:26,000 they've got their hierarchical document, but it's all about 234 00:14:26,000 --> 00:14:29,960 this adaptive diagnosis where that child is that hierarchical 235 00:14:29,960 --> 00:14:32,440 knowledge. Is it it's against the Common 236 00:14:32,440 --> 00:14:36,120 Core in the US? That's fine, which is great. 237 00:14:36,120 --> 00:14:38,760 So it knows where it is. But I think there's another 238 00:14:38,760 --> 00:14:43,080 question going completely off tangent with any of these AI 239 00:14:43,080 --> 00:14:50,720 tutors is the fact that let's say I was in year 4 as a year 4 240 00:14:50,720 --> 00:14:54,600 as a student. This may more work more in 241 00:14:54,600 --> 00:14:57,320 secondary context actually. So let's say I'm a year a 242 00:14:58,120 --> 00:15:03,880 student and I've been given a tutor like this as well as can 243 00:15:03,880 --> 00:15:08,440 high quality teaching in class. But this tutor does this 244 00:15:08,440 --> 00:15:11,040 diagnostic. And most of these platforms do 245 00:15:11,160 --> 00:15:14,160 have some kind of diagnostic and any intervention. 246 00:15:14,200 --> 00:15:17,160 Same thing we do that within schools, we do very much a human 247 00:15:17,760 --> 00:15:19,840 aid approach of it. That's, that's great. 248 00:15:19,840 --> 00:15:22,000 That's what we need to know where that student is. 249 00:15:22,360 --> 00:15:25,800 Is there a danger then that if it finds that I've got content 250 00:15:25,800 --> 00:15:30,840 knowledge from the year 4 year, five year 6 curriculum, well, is 251 00:15:30,840 --> 00:15:33,600 it going to take me all the way back there, fill in all those 252 00:15:33,600 --> 00:15:36,520 gaps and build it up? Or is it going to work alongside 253 00:15:36,520 --> 00:15:40,480 and support what I'm being taught in class and then almost 254 00:15:40,480 --> 00:15:42,600 consolidate that as well as fill the gap? 255 00:15:42,600 --> 00:15:45,240 So I don't, it's like there's a balance here as well. 256 00:15:45,440 --> 00:15:48,400 There's not only just with what knowledge do I need to fill the 257 00:15:48,400 --> 00:15:51,520 gap so I can work with it. And mass is an easy one with 258 00:15:51,520 --> 00:15:55,560 this because hierarchical, but there's also a how is it 259 00:15:56,320 --> 00:15:59,720 approached from a child's perspective? 260 00:16:00,160 --> 00:16:04,000 Because one of the advantages of technology and AI and where 261 00:16:04,000 --> 00:16:11,000 we're seeing a lot with pupils talking to these AI toys or 262 00:16:11,200 --> 00:16:14,960 things online, it's the fact that they don't want to talk to 263 00:16:14,960 --> 00:16:20,480 us, to their teachers. They actually get this feedback 264 00:16:20,680 --> 00:16:24,000 back from something that's inanimate, but they feel like 265 00:16:24,000 --> 00:16:26,040 it's a human. It's a lot softer, it's a lot 266 00:16:26,040 --> 00:16:27,880 kinder. It's kind of, they're happy to 267 00:16:27,880 --> 00:16:31,160 listen to that before they go back into class and try it. 268 00:16:31,160 --> 00:16:34,640 So I think there's a there's a moral side of this of how does 269 00:16:34,640 --> 00:16:39,560 it feel like from there pupil and student perspective. 270 00:16:39,800 --> 00:16:46,200 And I wonder, and I haven't seen it yet, too much evidence from 271 00:16:47,160 --> 00:16:50,840 what are pupils viewpoints on this as well? 272 00:16:51,240 --> 00:16:53,120 No. It's been a really interesting 273 00:16:53,440 --> 00:16:57,760 bit of data together about how they feel about it and what they 274 00:16:57,760 --> 00:16:59,840 think is best. And because you're getting from 275 00:16:59,840 --> 00:17:05,480 the couple of experiences that I know, you know, it's it's really 276 00:17:05,480 --> 00:17:08,560 mixed. And, you know, adults are as you 277 00:17:08,560 --> 00:17:13,760 know, yeah, there are some who are, you know, vastly pro, some 278 00:17:13,760 --> 00:17:17,760 who are vastly against it and vast majority of us in the 279 00:17:17,760 --> 00:17:19,119 middle. And I, from what I hear, it's 280 00:17:19,119 --> 00:17:22,720 the same with pupils and that use around AI as well. 281 00:17:22,720 --> 00:17:25,680 So it'd be really interesting to see whether, you know, are we 282 00:17:26,440 --> 00:17:28,400 there are, you know, the concept of schooling, we're forcing 283 00:17:28,440 --> 00:17:30,680 children to be in somewhere that they don't want to be anyway for 284 00:17:30,680 --> 00:17:32,760 the vast majority of the time. But we know it's good for them. 285 00:17:33,440 --> 00:17:42,560 But does that change when their experience of that would be, you 286 00:17:42,560 --> 00:17:44,840 know, right, You will have your own sort of personal 287 00:17:44,840 --> 00:17:48,120 workstations and get the headphones on. 288 00:17:48,120 --> 00:17:51,600 And your maths lesson is just, you know, wherever it is that 289 00:17:51,600 --> 00:17:56,600 you may be in whatever intelligence tutoring system, AI 290 00:17:56,600 --> 00:18:00,720 tutoring tool that rears its head and that schools are able 291 00:18:00,720 --> 00:18:07,480 to to afford or deploy in their, you know, infrastructure. 292 00:18:08,520 --> 00:18:10,560 Yeah, no, it's, it's a weird 1. And I think, you know, we've 293 00:18:11,160 --> 00:18:13,800 lots have been said about sort of belonging and all of that 294 00:18:13,800 --> 00:18:17,240 stuff. And you know, if the, I, I think 295 00:18:17,240 --> 00:18:22,600 I maybe struggle with school a little bit more than I did if my 296 00:18:22,600 --> 00:18:28,240 experience of it was, you know, come in and log in and just do 297 00:18:28,240 --> 00:18:32,960 what is what you need to do. And with a teacher teaching 298 00:18:32,960 --> 00:18:35,320 assistant wandering around and just being that extra bit of 299 00:18:35,320 --> 00:18:37,160 support for me if I needed it, I don't know. 300 00:18:37,360 --> 00:18:40,680 But no getting there, getting Jordan's thoughts about all of 301 00:18:40,680 --> 00:18:41,720 this would be really interesting. 302 00:18:41,720 --> 00:18:43,760 And I think that you mentioned that another interesting point 303 00:18:43,760 --> 00:18:46,560 there as well around sort of, you know, how best would we use 304 00:18:46,560 --> 00:18:51,440 these in terms of, is it just a, a consolidation tool or do you 305 00:18:51,440 --> 00:18:55,160 give teachers the, the pupils the choice to go, Hey, you know, 306 00:18:55,160 --> 00:18:58,960 if you're smashing it and you're in year 4 and you want to move 307 00:18:58,960 --> 00:19:01,520 on to the Year 5 stuff, you know, crack on because then, you 308 00:19:01,520 --> 00:19:07,160 know, you're potentially, you know, widening at a gap within 309 00:19:07,160 --> 00:19:11,880 your own sort of microcosm of your own classroom, which if 310 00:19:11,880 --> 00:19:16,160 anyone's, you know, taught any maths at primary school upper 311 00:19:16,160 --> 00:19:18,320 towards the upper levels. And you know, the lowest, you 312 00:19:18,320 --> 00:19:20,840 know, you've still got people who struggle to number bond to 313 00:19:20,840 --> 00:19:23,240 five and those who are ready to do some interest, more 314 00:19:23,240 --> 00:19:26,560 interesting, complex stuff. And if you're if you're just 315 00:19:26,560 --> 00:19:28,840 exasperating that, that would make any sort of poll class 316 00:19:28,840 --> 00:19:31,280 teaching that you're doing harder. 317 00:19:32,560 --> 00:19:35,440 But again, you know, this conversation is not it's not 318 00:19:35,440 --> 00:19:36,520 new. We've been having this sort of 319 00:19:36,560 --> 00:19:39,600 conversation of, you know, do you give your quote UN quote 320 00:19:39,600 --> 00:19:43,600 rapid graspers, which is a term I absolutely lower than detest. 321 00:19:44,760 --> 00:19:46,160 Do you give your what you're doing? 322 00:19:46,160 --> 00:19:49,000 If you're rapid graspers, they clearly do it know it. 323 00:19:49,000 --> 00:19:50,720 So what you're doing to move them on quickly. 324 00:19:50,720 --> 00:19:53,320 And then, you know, you get to year 6 and you wonder, you know, 325 00:19:53,800 --> 00:19:58,080 it's how did, how did the gap go from, you know, a couple of 326 00:19:58,080 --> 00:20:02,360 months in reception all the way then through to several, several 327 00:20:02,360 --> 00:20:05,800 years by year 6? Because we've been widening 328 00:20:05,800 --> 00:20:07,800 those gaps as part of our teaching because we had to, you 329 00:20:07,800 --> 00:20:13,360 know, get those rapid graspers through more interesting stuff 330 00:20:13,360 --> 00:20:18,280 because that's what teachers are some members of SLT wanted in 331 00:20:18,280 --> 00:20:20,960 some schools. So I think I'd be more 332 00:20:20,960 --> 00:20:24,400 comfortable with it as a. Consolidation tool. 333 00:20:24,400 --> 00:20:28,400 I quite like the idea of it just being like 2 weeks behind, like 334 00:20:28,400 --> 00:20:33,160 what you're doing in your like lessons coupled with a bit more 335 00:20:33,200 --> 00:20:36,960 sort of like spaced retrieval from some fancy algorithm. 336 00:20:36,960 --> 00:20:40,720 Or I could even tell the system what we've done and what I want 337 00:20:40,720 --> 00:20:44,720 them to do. And the system could use all its 338 00:20:44,720 --> 00:20:49,200 knowledge that it knows that peoples have questions right and 339 00:20:49,200 --> 00:20:53,000 wrong and whatnot. And it can do that to sort of 340 00:20:53,000 --> 00:20:56,240 give them their own sort of variation or their own sort of 341 00:20:56,240 --> 00:20:59,840 real specific space practice quiz based to them and then go 342 00:20:59,840 --> 00:21:01,400 through that content that they all had. 343 00:21:02,920 --> 00:21:06,280 You know, I think for me, that's a better place to start with 344 00:21:06,280 --> 00:21:09,560 rather than just, you know, here's a, here's a system. 345 00:21:09,560 --> 00:21:12,960 Go wild, kids. Yeah, In some ways, this whole 346 00:21:12,960 --> 00:21:15,680 idea of spatial treatment, I've always thought that technology 347 00:21:15,680 --> 00:21:19,840 is perfectly suited to it, putting up the right points for 348 00:21:19,840 --> 00:21:23,280 pupils at the right time. Like as teachers we try and put 349 00:21:23,280 --> 00:21:28,480 this through like a do now activity in the in the classroom 350 00:21:28,840 --> 00:21:30,520 before we go to the main bit of learning. 351 00:21:30,520 --> 00:21:34,800 We try and do it, but I don't think it's a necessarily of a 352 00:21:34,800 --> 00:21:37,440 clever way of doing it. We're manually doing it and 353 00:21:37,440 --> 00:21:41,560 every child in our class will need to practice things more 354 00:21:41,560 --> 00:21:45,280 times than their peers next to them or less to make sure it's 355 00:21:45,280 --> 00:21:48,280 consolidated and it has encoded into a long term memory. 356 00:21:48,280 --> 00:21:53,440 So I think technology's got a really interesting way of being 357 00:21:53,440 --> 00:21:55,640 able to do that really, really well. 358 00:21:55,720 --> 00:21:58,960 And I think the AI tutors, if they consolidate the learning 359 00:21:58,960 --> 00:22:02,200 that's happening in the classroom, having that space 360 00:22:02,200 --> 00:22:05,440 retrieval aspects into it. So they really mastered it. 361 00:22:06,040 --> 00:22:09,840 It would ensure that it would honor that mastery element that 362 00:22:09,840 --> 00:22:13,080 we talk about so often. They say you need to grasp it 363 00:22:13,080 --> 00:22:15,920 before we can move on and make sure those prerequisites are 364 00:22:15,920 --> 00:22:19,120 secure. But I feel like it goes into a 365 00:22:19,120 --> 00:22:22,680 bigger conversation that we won't get into because we'll be 366 00:22:22,680 --> 00:22:25,840 here all day and it's not a job need of us want probably is. 367 00:22:26,760 --> 00:22:30,800 We have a problem with our education system where you just 368 00:22:30,800 --> 00:22:34,680 go through the school years and what are the gaps? 369 00:22:34,680 --> 00:22:37,480 The gaps just get wider and wider and wider. 370 00:22:37,800 --> 00:22:41,640 And if there is a hope or if there's a way that we, each 371 00:22:42,440 --> 00:22:47,800 pupil can have access to a pedagogically rich AI tutor that 372 00:22:47,800 --> 00:22:51,520 supports them in that consolidation, well surely 373 00:22:51,520 --> 00:22:54,680 that's something we should explore because it seems like 374 00:22:54,680 --> 00:22:58,440 the benefits of it could, could be huge. 375 00:22:58,440 --> 00:23:00,760 We don't know. And then it goes on to the other 376 00:23:00,760 --> 00:23:02,480 thing. I think you you said earlier and 377 00:23:02,480 --> 00:23:07,000 it was a came from, I think it was a new news article as well, 378 00:23:07,800 --> 00:23:12,960 where a dad at North created AAI chat bot. 379 00:23:12,960 --> 00:23:16,520 It was a couple of years ago before the AI tutors were a 380 00:23:16,520 --> 00:23:21,600 thing, really a news ChatGPT and the and the GPT custom GPT in 381 00:23:21,600 --> 00:23:26,440 there, but actually personalized tutoring to that child. 382 00:23:26,440 --> 00:23:29,600 So whatever their interests were, they kind of then applied 383 00:23:29,600 --> 00:23:32,520 the maths or their English, whatever they were doing to that 384 00:23:32,520 --> 00:23:35,800 context. I don't know if I'm comfortable 385 00:23:35,800 --> 00:23:37,440 with that. I don't know if I'm comfortable 386 00:23:37,440 --> 00:23:42,760 with this idea of personalizing learning to that context of that 387 00:23:42,760 --> 00:23:46,040 child. I, I see the benefits of it, but 388 00:23:46,040 --> 00:23:48,960 I don't, it feel makes me uncomfortable and I'm not sure 389 00:23:48,960 --> 00:23:52,200 why. Do you see what I mean or? 390 00:23:53,240 --> 00:23:58,800 Yeah, it's, I think it's longer term. 391 00:23:58,800 --> 00:24:05,760 I think it's just that message of a you can't do this unless 392 00:24:05,760 --> 00:24:11,400 it's you're interested in it, in which case that's blatantly not 393 00:24:11,400 --> 00:24:15,000 true because eventually we want them to transfer this knowledge 394 00:24:15,000 --> 00:24:17,880 to things that they're certainly not interested in. 395 00:24:17,920 --> 00:24:21,040 Like, you know, you hear all the time, don't you, about, you 396 00:24:21,080 --> 00:24:24,600 know, the, the, the ever updated list of things that we should be 397 00:24:24,600 --> 00:24:26,840 teaching kids throughout their, their schooling and, you know, 398 00:24:26,840 --> 00:24:29,240 mortgages and personal finance and all of that stuff. 399 00:24:29,240 --> 00:24:32,360 You know, no 14 year old is interested in tax returns and, 400 00:24:32,400 --> 00:24:38,160 and mortgages, but we've certainly given the maths to be 401 00:24:38,160 --> 00:24:39,280 able to sort of apply it to that. 402 00:24:39,280 --> 00:24:42,800 So the idea that we would then give them information that it 403 00:24:42,840 --> 00:24:44,760 could be about football, rugby or whatever. 404 00:24:44,760 --> 00:24:46,800 And then, you know, that's some issues right there as well, 405 00:24:46,800 --> 00:24:49,400 instead of potentially, you know, reinforcing those gender 406 00:24:49,400 --> 00:24:52,080 stereotypes as well, which, you know, is not needed. 407 00:24:52,200 --> 00:24:55,920 They'll then be able to, you know, after only sort of working 408 00:24:55,920 --> 00:24:59,440 with rugby and football or, you know, whatever it might be that 409 00:24:59,440 --> 00:25:02,240 they can then magically transfer this to mortgages and tax 410 00:25:02,240 --> 00:25:05,080 returns or, you know, whatever it is, basic percentages when 411 00:25:05,080 --> 00:25:06,600 they're out shopping or whatever. 412 00:25:07,600 --> 00:25:11,040 I don't think necessarily is going to be helpful for them. 413 00:25:11,040 --> 00:25:12,480 So I think there's like a transfer issue. 414 00:25:12,840 --> 00:25:16,280 I think there's sort of a long term sort of motivation issue in 415 00:25:16,280 --> 00:25:17,840 the terms that, you know, it signals. 416 00:25:17,840 --> 00:25:20,400 Well, don't worry, you know, you don't have to worry about things 417 00:25:20,400 --> 00:25:22,160 that aren't interesting to you right now. 418 00:25:22,160 --> 00:25:26,720 But you know, you know, you introducing something 419 00:25:26,720 --> 00:25:29,640 interesting in mathematics, you know, whatever it might be, 420 00:25:30,080 --> 00:25:35,160 might give them a new interest or spur some interest that 421 00:25:35,160 --> 00:25:38,880 wasn't there before. And so I think the idea and it's 422 00:25:38,880 --> 00:25:41,320 also, you know, 30 kids, right? You know, yes, we can probably 423 00:25:41,320 --> 00:25:45,720 do this a little bit quicker now because of ChatGPT and and 424 00:25:45,720 --> 00:25:48,000 whatnot. But you know, 30 different kids 425 00:25:48,000 --> 00:25:51,960 with 30 different interests. That's a lot of yeah, that's a 426 00:25:51,960 --> 00:25:54,000 lot of worksheets or a lot of questions that. 427 00:25:54,040 --> 00:25:58,000 And obviously, if you're doing 1 to 30 teaching in your in, you 428 00:25:58,000 --> 00:26:01,280 know, in class, I'll say, you know, in an analogue style, you 429 00:26:01,280 --> 00:26:03,760 know, you're not going to be able to to do it like that. 430 00:26:03,760 --> 00:26:06,840 And then you just go into, you know, you think as a teacher, 431 00:26:06,840 --> 00:26:10,160 you're thinking about the wrong things as well. 432 00:26:10,160 --> 00:26:12,640 You're not thinking about how am I going to teach us in a way 433 00:26:12,640 --> 00:26:15,640 that is going to means that the vast majority of the class doing 434 00:26:15,680 --> 00:26:19,320 it, you're thinking, oh, what was Sandy's favorite thing that 435 00:26:19,320 --> 00:26:23,080 I need to be putting this what that I should wrap this question 436 00:26:23,080 --> 00:26:25,360 in? And so I think you know, big 437 00:26:25,360 --> 00:26:29,520 issues when it comes when it comes to that for me I think as 438 00:26:29,520 --> 00:26:33,160 well, I think those are my 3 reasons why I sort of don't like 439 00:26:33,160 --> 00:26:35,320 it. Long term motivation. 440 00:26:35,440 --> 00:26:39,160 I think there's a transfer issue and in terms of I think it just 441 00:26:39,160 --> 00:26:42,600 increases teachers workload with importantly sort of very little 442 00:26:42,600 --> 00:26:46,560 evidence that it actually improves, improves anything. 443 00:26:47,040 --> 00:26:50,880 I think there will be use in terms of applying contacts in 444 00:26:50,880 --> 00:26:53,200 certain situations. I don't think it's a complete. 445 00:26:53,200 --> 00:26:58,080 Let's not personalize anything, but I do wonder whether the when 446 00:26:58,080 --> 00:27:05,000 these tools get more important and more capable down the line, 447 00:27:05,000 --> 00:27:07,520 let's say, and you've got your own quick and you've got your 448 00:27:07,520 --> 00:27:10,240 own pedagogical model. It's Socrative and it's it is 449 00:27:10,240 --> 00:27:13,680 way of working. There could be things like 450 00:27:14,080 --> 00:27:17,600 financial literacy. It could be one of those painful 451 00:27:17,600 --> 00:27:21,200 lessons on a Friday afternoon, the graveyard shift, where 452 00:27:21,680 --> 00:27:25,560 actually let's put it into context of a project that 453 00:27:26,120 --> 00:27:28,120 they're creating a business or something. 454 00:27:28,400 --> 00:27:31,240 Actually, they want to learn those skills and they need to 455 00:27:31,240 --> 00:27:34,680 learn those skills to be able to apply it in the right way. 456 00:27:34,680 --> 00:27:38,680 And I think that could be an interesting aspect on it, 457 00:27:41,520 --> 00:27:45,040 Something you were saying around the role of the teacher and 458 00:27:45,040 --> 00:27:47,800 where these AI tutors come in place. 459 00:27:47,800 --> 00:27:53,280 And Harvard released ARCT as well, and they came out with 460 00:27:53,280 --> 00:27:55,240 some interesting things that compared. 461 00:27:55,840 --> 00:27:59,560 They compared a custom AI tutor to active in class learning. 462 00:28:00,320 --> 00:28:06,280 And they found that they, through the AI tutor, they 463 00:28:06,280 --> 00:28:09,800 learned significantly more in less time. 464 00:28:09,920 --> 00:28:12,800 And then students felt more engaged, They felt more 465 00:28:13,040 --> 00:28:15,600 motivated. And it was like pedagogically 466 00:28:16,160 --> 00:28:20,240 following the best practices. But I think what it for me, what 467 00:28:20,240 --> 00:28:24,160 I took away from that report more than anything, because it's 468 00:28:24,160 --> 00:28:30,360 about 194 undergraduate physics and this one sort of science is 469 00:28:31,360 --> 00:28:35,640 the design of how you use these tutors are important because 470 00:28:35,640 --> 00:28:39,280 what it comes back to is if you're consolidating learning at 471 00:28:39,280 --> 00:28:43,600 home and supporting it, it's that what do you do in the 472 00:28:43,600 --> 00:28:46,000 classroom? Does your shift in the classroom 473 00:28:46,000 --> 00:28:50,760 that yes, we will really important guided in practice 474 00:28:51,040 --> 00:28:53,200 that we're going to do regarding instruction and we're going to 475 00:28:53,200 --> 00:28:54,560 do some really purposeful practice. 476 00:28:54,840 --> 00:28:58,120 We'll do some low stakes quizzing behind it. 477 00:28:58,120 --> 00:29:00,200 Do a lots of check for understanding that you know that 478 00:29:00,280 --> 00:29:02,880 you are confident you're clear with this in this application. 479 00:29:03,440 --> 00:29:06,040 But some of that extra consolidation work or something 480 00:29:06,040 --> 00:29:08,200 else can be done outside the class. 481 00:29:08,240 --> 00:29:11,440 And I think that whole shift of what happens in school, what 482 00:29:11,440 --> 00:29:15,080 happens outside the school and raise the question of homework 483 00:29:15,080 --> 00:29:22,120 again is a really interesting debate that I think education is 484 00:29:22,120 --> 00:29:26,120 going to have very soon because there is every school across the 485 00:29:26,120 --> 00:29:29,400 country and across the world that they will give homework 486 00:29:29,400 --> 00:29:31,240 out. At the moment that will be just 487 00:29:31,240 --> 00:29:36,680 consolidating bits and pieces on paper or using online platforms 488 00:29:36,680 --> 00:29:42,080 like Sparks or what what not. And you can just click the right 489 00:29:42,080 --> 00:29:45,600 answer, get AI to do it for you anyway because you copy and 490 00:29:45,600 --> 00:29:47,000 paste it and they don't learn anything. 491 00:29:47,120 --> 00:29:51,440 So for me, it raises this whole AI tutor thing, almost raises 492 00:29:51,440 --> 00:29:55,200 the question on what is important that we should be 493 00:29:55,200 --> 00:29:59,240 keeping in the classroom and what are the important things? 494 00:29:59,240 --> 00:30:03,520 What are some of the things that we could offload to AI to ensure 495 00:30:03,520 --> 00:30:09,640 that our classroom time is the most powerful and rich and those 496 00:30:09,640 --> 00:30:12,640 relationships are built and we really have an idea of where 497 00:30:12,640 --> 00:30:16,280 those children are and we can guide them to the next steps? 498 00:30:17,120 --> 00:30:19,400 I don't know the answer to it. I just think it's going to be an 499 00:30:19,400 --> 00:30:23,240 interesting debate that AI is going to create for us. 500 00:30:23,680 --> 00:30:26,680 Yeah. I'm not sure how much I agree 501 00:30:26,680 --> 00:30:31,240 with that in the sense of whether AI would be able to sort 502 00:30:31,240 --> 00:30:34,520 of completely take over aspects of that. 503 00:30:34,520 --> 00:30:37,000 And that means that we can focus on different things in the 504 00:30:37,000 --> 00:30:41,960 classroom because I just don't think we know enough about 505 00:30:41,960 --> 00:30:44,480 whether AI will ever get there to that point. 506 00:30:45,160 --> 00:30:49,960 I mean, the Harvard is a useful is an interesting study, but as 507 00:30:49,960 --> 00:30:57,840 you mentioned, I think it was university physics students, you 508 00:30:57,840 --> 00:31:01,720 know, I don't it's it's interesting. 509 00:31:01,720 --> 00:31:04,040 Of course it is. But when you go to, you know, 510 00:31:05,560 --> 00:31:08,000 people are on that physics course because they want to 511 00:31:08,080 --> 00:31:11,560 learn about physics and they are, they are paying. 512 00:31:13,080 --> 00:31:16,160 It's a mistake. So, you know, a fair and it's 513 00:31:16,160 --> 00:31:19,720 Harvard, so a fair whack to to, to be there. 514 00:31:19,720 --> 00:31:22,600 And so there is a motivation that that doesn't exist, I think 515 00:31:22,600 --> 00:31:26,280 for the 14 year old and 15 year old who again, is going to 516 00:31:26,280 --> 00:31:30,240 school, because the, the law says that, that, that you, you 517 00:31:30,240 --> 00:31:36,360 need to and have to be there. I also think, you know, if 518 00:31:36,360 --> 00:31:40,400 you're studying physics at university, you have a a whole 519 00:31:40,400 --> 00:31:44,120 host of prior knowledge of which you can use to help you with 520 00:31:44,120 --> 00:31:45,520 your physics. And you have obviously 521 00:31:45,520 --> 00:31:51,040 demonstrated through various sort of entrance examinations, 522 00:31:51,040 --> 00:31:53,800 tests, whatever it might be in America. 523 00:31:53,800 --> 00:31:56,840 So, you know, GPA S and all that sort of fun stuff that you are 524 00:31:56,960 --> 00:32:02,400 capable of studying physics at a very prestigious university at, 525 00:32:02,400 --> 00:32:05,520 you know, this level, which again, goes back to what we were 526 00:32:05,520 --> 00:32:08,960 talking about in, you know, that's not the case where you 527 00:32:08,960 --> 00:32:11,760 can sort of guarantee that everyone is going to be far 528 00:32:11,760 --> 00:32:14,480 closer in terms of their understanding and knowledge 529 00:32:15,080 --> 00:32:16,720 rather than that. And I think, yeah, you could 530 00:32:16,720 --> 00:32:19,200 absolutely. I think, you know, and for 531 00:32:19,560 --> 00:32:24,320 humans, say humans for adults to learn something when you have 532 00:32:24,320 --> 00:32:27,080 that motivation to do so, I could imagine that I would 533 00:32:27,080 --> 00:32:31,480 probably learn a little bit more faster and certainly report 534 00:32:31,480 --> 00:32:33,400 higher engagement motivation about it. 535 00:32:34,840 --> 00:32:37,160 I just don't know whether that's the same for teenagers and 536 00:32:37,160 --> 00:32:39,720 children. Often when I talk about AI or 537 00:32:39,720 --> 00:32:43,840 think about AI and have conversations, I think I always 538 00:32:43,840 --> 00:32:48,760 go back to where shouldn't AI be and where shouldn't we use 539 00:32:48,760 --> 00:32:51,000 technology? And where does the human bit 540 00:32:51,640 --> 00:32:54,320 come in place? And I'm just trying to take 541 00:32:54,320 --> 00:32:59,200 something that I don't think as adults, often we're in that 542 00:32:59,200 --> 00:33:02,800 tricky space of remembering how learning a new thing is 543 00:33:02,800 --> 00:33:07,440 difficult. So let's take something like 544 00:33:07,440 --> 00:33:12,400 physical like golf for example. OK I've got a golf set here and 545 00:33:12,400 --> 00:33:16,640 I am absolutely rubbish because all I'm doing is watching a 546 00:33:16,640 --> 00:33:19,040 YouTube video. I don't know how it feels like 547 00:33:19,040 --> 00:33:21,440 what a good swing does. I didn't want 1 intro. 548 00:33:21,840 --> 00:33:27,040 And yes I can get a camera. Who does AI tutor? 549 00:33:27,040 --> 00:33:28,680 It literally is like an AI tutor. 550 00:33:28,680 --> 00:33:31,200 It will see where I am, put the pinpoints on it saying these are 551 00:33:31,200 --> 00:33:32,560 where I think it's going to work. 552 00:33:32,840 --> 00:33:34,200 This is what you need to work at. 553 00:33:35,080 --> 00:33:38,600 But I need that human, I need that someone to really guide me 554 00:33:38,600 --> 00:33:40,680 and point me in the right direction. 555 00:33:40,760 --> 00:33:42,520 And I think that's important as well. 556 00:33:42,520 --> 00:33:46,760 And it's with all of this, it's I've got a passion to want to 557 00:33:46,760 --> 00:33:48,120 learn on it, but I don't know what I'm doing. 558 00:33:48,120 --> 00:33:51,400 I'm wasting almost wasting my time trying to explore it. 559 00:33:51,400 --> 00:33:56,520 I need AI, need a professional guide or professional helper to 560 00:33:56,520 --> 00:33:59,960 help me get to that initial stage where I can then use 561 00:33:59,960 --> 00:34:05,040 almost a hybrid AIG Turing almost to consolidate what the 562 00:34:05,040 --> 00:34:07,000 teacher saying or the students like. 563 00:34:07,000 --> 00:34:10,840 Well, I'm getting told, and I think that could be replica in 564 00:34:11,360 --> 00:34:14,679 education in some ways. This idea of space retrieval, 565 00:34:14,679 --> 00:34:19,760 this idea of some pupils will need to consolidate and practice 566 00:34:19,760 --> 00:34:24,639 things a lot more than others. I think that's where we've got a 567 00:34:24,639 --> 00:34:29,280 real hope in this use. And when the DfE announced 568 00:34:29,280 --> 00:34:32,679 they're looking into AI tutors, they explicitly aimed at 569 00:34:32,679 --> 00:34:37,679 disadvantaged pupils. And it's from 2 points of view, 570 00:34:37,719 --> 00:34:40,120 one, to close a cap. But two, I think there's an 571 00:34:40,120 --> 00:34:45,719 equity thing here as well because those who have devices 572 00:34:45,840 --> 00:34:50,560 can access free AI online quite easily, quite quickly, whether 573 00:34:50,560 --> 00:34:53,480 it's to do their homework or whether to actually use it 574 00:34:53,480 --> 00:34:56,000 purposefully. I think it's an interesting 575 00:34:56,000 --> 00:35:01,160 scope there as well. I have succumbed and I give open 576 00:35:01,200 --> 00:35:04,640 AI £20 a month for my hard earned money because I was 577 00:35:04,640 --> 00:35:10,840 getting really annoyed with my rate limits being, you know, put 578 00:35:10,840 --> 00:35:14,600 to 0 within sort of, you know, 5-10 minutes of a, of a 579 00:35:14,600 --> 00:35:17,720 conversation. And you know, with all my 20 LB 580 00:35:17,720 --> 00:35:20,200 it's I very rarely ever have that problem. 581 00:35:20,200 --> 00:35:24,720 It's the same with with Claude. And so absolutely, you know, 582 00:35:24,720 --> 00:35:29,160 currently if you are getting your, if you're fortunate enough 583 00:35:29,160 --> 00:35:33,840 to be able to give, have a spare 20 quids to give to Anthropic or 584 00:35:33,840 --> 00:35:39,720 open AI, you know, and you're sitting your 1415 year olds down 585 00:35:39,720 --> 00:35:41,760 and getting them to do things that they are getting more out 586 00:35:41,760 --> 00:35:46,800 of it than the and they probably have again, yeah, not new 587 00:35:46,800 --> 00:35:51,160 conversations around advantage versus disadvantage. 588 00:35:51,320 --> 00:35:53,520 They probably have a set space for them to do it. 589 00:35:53,520 --> 00:35:56,600 They probably have a decent computer that would that be a 590 00:35:56,960 --> 00:35:58,560 laptop. So, you know, at least a sort of 591 00:35:58,560 --> 00:36:00,760 13 inch screen. And for the disadvantaged 592 00:36:00,800 --> 00:36:03,080 pupils, they probably have to try and manage on, you know, 593 00:36:03,160 --> 00:36:06,720 some sort of mobile device, whether that might be some sort 594 00:36:06,720 --> 00:36:11,120 of tablet or a mobile phone, which is just, you know, not 595 00:36:11,560 --> 00:36:14,480 necessarily the the optimum way to go about it. 596 00:36:15,160 --> 00:36:17,080 So yeah, that exists. And I think it's really 597 00:36:17,080 --> 00:36:24,400 interesting to see how the UK government are going to approach 598 00:36:24,400 --> 00:36:29,560 this AI powered 1 to 1 tutoring and whether the well, again, you 599 00:36:29,560 --> 00:36:32,920 know, is the expectation going to be that that's that they do 600 00:36:32,920 --> 00:36:37,400 it in in school, whether that be after school lunchtime or are 601 00:36:37,400 --> 00:36:39,760 they hoping that it would, you know, they'll, you know, these 602 00:36:40,080 --> 00:36:41,360 pupils will magically want to decide. 603 00:36:41,360 --> 00:36:44,280 So now it will potentially do their homework if they didn't do 604 00:36:44,280 --> 00:36:48,200 so because it's AI powered. I think it's going to be really 605 00:36:48,200 --> 00:36:51,720 interesting to see what how they go about this and how they sort 606 00:36:51,720 --> 00:36:53,920 of measure success and and all that stuff. 607 00:36:53,920 --> 00:36:55,400 It's yeah, looking forward to it. 608 00:36:56,040 --> 00:36:59,040 Yeah. And I think with anything with 609 00:36:59,040 --> 00:37:04,440 this, in this AI world, what you, what it is trifecta of 610 00:37:04,840 --> 00:37:07,000 policy needs to be supportive of it. 611 00:37:07,040 --> 00:37:09,480 You need actually companies who can build this, but also you 612 00:37:09,480 --> 00:37:13,560 need the, you need the teachers with the practicalities of the 613 00:37:13,560 --> 00:37:16,960 implementation, but also the cognitive science to build it as 614 00:37:16,960 --> 00:37:19,120 well, as well as the curriculum. So there's a quite a few 615 00:37:19,120 --> 00:37:23,120 different things that need to work together in order to create 616 00:37:23,120 --> 00:37:29,520 an AI tutor or however we use it to be purposeful. 617 00:37:32,240 --> 00:37:35,280 One thing I think we could go on for hours, but I think one thing 618 00:37:35,400 --> 00:37:37,520 I think we should leave on today and then we can probably come 619 00:37:37,520 --> 00:37:41,160 back to this, come back and actually, yeah, see if we can 620 00:37:41,160 --> 00:37:44,600 get Craig on to talk about the Ed report in a little bit more 621 00:37:44,600 --> 00:37:48,120 detail and and see what their plans are for it. 622 00:37:49,800 --> 00:37:53,640 And we talked about it before and you brought it up, but I 623 00:37:53,640 --> 00:37:57,080 think with all of these reports, there's something around 624 00:37:57,080 --> 00:38:02,280 accuracy of AI tutors that comes up time and time and time again. 625 00:38:02,280 --> 00:38:06,880 And you say that, OK, it may be nought .41% or whatever it is 626 00:38:07,280 --> 00:38:09,400 that an AI tutor will get something wrong. 627 00:38:11,320 --> 00:38:13,600 And therefore that that could equate to about a million 628 00:38:13,600 --> 00:38:17,600 questions a year, a day or so if it's for every student. 629 00:38:18,480 --> 00:38:24,960 But are we, is our expectation of AI to be 100% correct or any 630 00:38:24,960 --> 00:38:28,760 technology to be 100% correct 100% of the time too ambitious? 631 00:38:29,080 --> 00:38:33,080 Because I know as teachers, there are lessons that we've all 632 00:38:33,080 --> 00:38:35,040 taught. We've walked away from them and 633 00:38:35,040 --> 00:38:38,520 it's like, why did I do that or completely different? 634 00:38:38,840 --> 00:38:42,000 And I didn't plan to do that. And it just went off on a 635 00:38:42,000 --> 00:38:45,920 tangent. And we've all done that, whether 636 00:38:45,920 --> 00:38:48,320 it's through an ECT or whether we're experienced teachers. 637 00:38:49,120 --> 00:38:54,280 So are we holding AI to this pedestal where actually it will 638 00:38:54,280 --> 00:38:57,760 never reach to our expectations And it's probably a better way 639 00:38:57,760 --> 00:39:00,360 to think about, OK, how can we integrate this into what we're 640 00:39:00,360 --> 00:39:02,880 doing already? Yeah, definitely. 641 00:39:03,920 --> 00:39:06,840 I think that is the approach is how we can integrate into what's 642 00:39:07,000 --> 00:39:10,520 what we're already doing. I do think we, I think that what 643 00:39:10,520 --> 00:39:14,200 makes it so difficult is that we just don't know the proportion 644 00:39:14,240 --> 00:39:18,760 of errors that teachers make on a regular basis. 645 00:39:18,760 --> 00:39:22,240 And thinking of the top head, like the only way you could do 646 00:39:22,240 --> 00:39:27,520 that is if you hired probably thousands of observers to go 647 00:39:27,520 --> 00:39:33,400 into a representative number of schools across the age spectrum, 648 00:39:33,880 --> 00:39:36,440 making sure that you were primarily looking at across a 649 00:39:36,440 --> 00:39:39,480 whole, you know, all the different types of lessons for a 650 00:39:39,480 --> 00:39:43,800 large enough portion of time. And they were just noting down 651 00:39:44,880 --> 00:39:49,040 errors as they sort of noticed them happen. 652 00:39:49,080 --> 00:39:53,480 And maybe then you would get some idea of the proportion of 653 00:39:53,840 --> 00:39:58,760 errors that happen, you know, daily on a, you know, every sort 654 00:39:58,760 --> 00:40:02,160 of day current situation. But then, you know, you've got 655 00:40:02,160 --> 00:40:04,280 problems like, you know, how I'm being observed. 656 00:40:04,280 --> 00:40:06,160 So I'm going to make sure that my lessons are really up to 657 00:40:06,160 --> 00:40:07,800 scratch this week and stuff like that. 658 00:40:07,800 --> 00:40:10,600 So there could still be some issues with that, but it's the 659 00:40:10,600 --> 00:40:12,240 only way that I could think about that you could actually do 660 00:40:12,240 --> 00:40:18,120 it. I would be surprised. 661 00:40:19,520 --> 00:40:23,720 And again, there is no it's just a vibe because humans are 662 00:40:24,360 --> 00:40:28,680 fallible to all sorts of erroneous mistakes and errors. 663 00:40:28,680 --> 00:40:37,840 But I would be amazed if more, if a tutor, if a teacher didn't 664 00:40:37,840 --> 00:40:44,000 make more errors than this sort of, you know, 0.14 665 00:40:44,000 --> 00:40:46,600 hallucinations or whatever that was, if every student in England 666 00:40:46,600 --> 00:40:50,120 used them for two lessons. So this is from that Ed tutor. 667 00:40:50,120 --> 00:40:53,560 So if every student used this EED tutor for two lessons, that 668 00:40:53,560 --> 00:40:56,200 would mean about sort of a million errors across that 669 00:40:56,200 --> 00:40:58,480 sample. And I hate it when, when you 670 00:40:58,480 --> 00:41:00,040 sort of measure things that way as well. 671 00:41:00,040 --> 00:41:03,680 And you know, when there's 700,000 days missed a week or 672 00:41:03,680 --> 00:41:06,200 something like that. Yeah, it's a rubbish way to 673 00:41:06,200 --> 00:41:09,240 report things. But anyway, I digress. 674 00:41:11,160 --> 00:41:15,000 Yeah, I, I think we do. I also think, think about EED as 675 00:41:15,000 --> 00:41:17,480 well. You know, that is a sort of a 676 00:41:17,480 --> 00:41:22,560 one to one situation. So if one child is making that 677 00:41:22,720 --> 00:41:25,040 gets an error wrong in that message, that's just that one 678 00:41:25,040 --> 00:41:27,360 child that's been affected. If a teacher makes an error in a 679 00:41:27,560 --> 00:41:30,160 in a lesson, that's a 1 to 30 error. 680 00:41:30,600 --> 00:41:34,200 And so unless the teacher, the teacher made it realise that 681 00:41:34,200 --> 00:41:37,160 they've made a mistake, in which case they might try and try and 682 00:41:37,160 --> 00:41:40,040 self correct. We've certainly all been there 683 00:41:40,040 --> 00:41:42,640 where we tried to self correct and the people still remember 684 00:41:42,640 --> 00:41:45,280 the wrong thing because that's the way they are because we go, 685 00:41:45,280 --> 00:41:48,400 oh, that was a mistake. I think we also got to think 686 00:41:48,400 --> 00:41:51,920 about like the kinds of areas that we're talking about here I 687 00:41:52,040 --> 00:41:55,720 go into when I certainly did go into many lessons, like, you 688 00:41:55,760 --> 00:41:59,240 know, I would still see work that wasn't necessarily, you 689 00:41:59,240 --> 00:42:01,440 know, text on the board or whatever, that wasn't 690 00:42:01,440 --> 00:42:03,920 necessarily like punctuated correctly. 691 00:42:03,920 --> 00:42:07,240 You know, obviously the pupils aren't noticing that, but they 692 00:42:07,240 --> 00:42:09,800 are still reading texts that isn't punctuated correctly. 693 00:42:09,800 --> 00:42:13,880 So is that an error that we're sort of happy with to say that, 694 00:42:13,880 --> 00:42:18,520 you know, that's a potentially, you know, saying that it could 695 00:42:18,520 --> 00:42:21,280 impact pupils learning because they're reading it and they're 696 00:42:21,280 --> 00:42:24,520 not taking in that orthographic system because they realise, oh, 697 00:42:24,720 --> 00:42:28,120 teacher didn't start that proper noun with a capital letter that 698 00:42:28,120 --> 00:42:29,800 time. And I haven't realised it's a 699 00:42:30,080 --> 00:42:32,400 proper noun, but I'm just going to read on anyway and go, oh, 700 00:42:32,400 --> 00:42:35,520 OK, fine. I, you know, don't need to use 701 00:42:35,520 --> 00:42:37,840 their capital letter for that word because teacher didn't have 702 00:42:37,840 --> 00:42:41,520 to a teacher forgot to, even though it's not a sort of, you 703 00:42:42,000 --> 00:42:44,560 know, outwardly facing whilst I'm doing some teaching sort of 704 00:42:44,560 --> 00:42:48,160 mistake. So, yeah, I think we do hold 705 00:42:48,160 --> 00:42:54,480 technology because we're so used to it sort of maybe we, we sort 706 00:42:55,000 --> 00:42:57,360 of put it on that sort of pedestal of, you know, it's all 707 00:42:57,400 --> 00:43:00,800 it's either right or wrong. And I think that's a big issue 708 00:43:00,800 --> 00:43:05,080 that we have, you know, with it, we do just take it as gospel. 709 00:43:05,080 --> 00:43:08,440 You know, I'm guilty if it's anything you get a confident 710 00:43:08,440 --> 00:43:11,760 message back from ChatGPT or insert whichever one of these 711 00:43:11,760 --> 00:43:14,560 chat bots. And so you do think it's right 712 00:43:14,720 --> 00:43:16,480 because we're so used to thinking that, you know, 713 00:43:17,000 --> 00:43:19,840 technology is sort of somehow superior. 714 00:43:19,840 --> 00:43:22,440 And you know, when you think about a calculator, calculator 715 00:43:22,440 --> 00:43:25,120 will always give you the right answer unless you unless it's a 716 00:43:25,120 --> 00:43:29,320 user error through the input. And I think the kind of 717 00:43:29,320 --> 00:43:32,640 technology that we've had is been that sort of system that 718 00:43:32,640 --> 00:43:35,960 this new technology that does Saffler from this sort of 719 00:43:36,600 --> 00:43:40,560 feature of hallucinations is kind of quite jarring to our 720 00:43:40,560 --> 00:43:43,640 mental models of how we suspect this technology to behave. 721 00:43:44,160 --> 00:43:47,240 And so we aren't necessarily good at sort of updating our 722 00:43:47,240 --> 00:43:49,520 model around that. This could be right, could be 723 00:43:49,520 --> 00:43:52,720 wrong. But that said, it's always 724 00:43:52,720 --> 00:43:55,080 important to remember that yes, we talk about keep the human in 725 00:43:55,080 --> 00:43:57,400 the loop, keep the expert in the loop, absolutely. 726 00:43:59,000 --> 00:44:02,360 But humans are fallible, they make errors. 727 00:44:02,440 --> 00:44:06,800 And I would imagine up and down the country on any day, assuming 728 00:44:06,800 --> 00:44:11,760 100% attendance across the every school in the country, I would 729 00:44:12,160 --> 00:44:16,200 suggest that within two lessons there are probably more than a 730 00:44:16,200 --> 00:44:19,640 million errors, either that being like direct or indirect 731 00:44:20,520 --> 00:44:24,200 being made. And I'd be surprised and shocked 732 00:44:24,240 --> 00:44:29,240 if it wasn't. And I'd call on any researcher 733 00:44:29,240 --> 00:44:34,920 who is listening to this try to find a way to work it out 734 00:44:34,920 --> 00:44:36,880 because I think it would be really interesting. 735 00:44:38,040 --> 00:44:40,560 But even I think it would change the conversation around how we 736 00:44:40,560 --> 00:44:41,880 think about these. Because if we can get the 737 00:44:41,880 --> 00:44:43,000 cheater, that's better than the cube. 738 00:44:43,000 --> 00:44:47,240 Even if you know, let's say for example, it's 0.2, well now you 739 00:44:47,240 --> 00:44:52,400 know, this combination through EED is better. 740 00:44:53,120 --> 00:44:55,200 And I think that would change the conversation a lot. 741 00:44:55,680 --> 00:44:59,560 Yeah, and even in, you wouldn't even need to see what teachers 742 00:44:59,560 --> 00:45:01,160 have written. It's just their language. 743 00:45:01,600 --> 00:45:04,240 Exactly. Across the across the country, 744 00:45:04,240 --> 00:45:08,240 it will be different in terms of how different nouns, phrases, 745 00:45:08,560 --> 00:45:10,680 colloquialisms that are being used as well. 746 00:45:11,720 --> 00:45:15,400 Actually that's not as easy as black and white, right or wrong 747 00:45:16,640 --> 00:45:18,560 as well. So I think that is really 748 00:45:18,560 --> 00:45:22,080 interesting. So I think while you were 749 00:45:22,240 --> 00:45:25,760 talking, I was thinking, we always, we've said to ourselves, 750 00:45:25,760 --> 00:45:27,640 we're healthy skeptics around AI. 751 00:45:28,200 --> 00:45:29,800 Often I think that's the phrase we've used. 752 00:45:30,280 --> 00:45:31,680 But I was like, OK, what are the problems? 753 00:45:31,680 --> 00:45:33,400 What are the hopeful things then? 754 00:45:33,400 --> 00:45:36,200 Because we can come down on this, say, oh, yeah, we don't 755 00:45:36,200 --> 00:45:38,880 think it's going to work. But I think there are some hopes 756 00:45:38,880 --> 00:45:43,960 in this. And I think this AI isn't the 757 00:45:43,960 --> 00:45:47,680 next big thing that's going to be like interactive whiteboards. 758 00:45:47,680 --> 00:45:53,080 I think this could change education or be embedded in. 759 00:45:53,080 --> 00:45:55,640 And I think we've got this, we've got it on a pedestal at 760 00:45:55,640 --> 00:45:56,920 the moment. It's going to change everything. 761 00:45:56,920 --> 00:45:59,560 But I think actually we should think about how it's embedded 762 00:45:59,560 --> 00:46:01,080 in. But the three things I thought 763 00:46:01,080 --> 00:46:05,240 of that I'm hopeful for in some ways is how we can close 764 00:46:05,240 --> 00:46:08,760 feedback loop, whether that's actually just for students 765 00:46:08,760 --> 00:46:12,880 getting feedback so they can Draw Something again or get 766 00:46:13,360 --> 00:46:16,880 immediate feedback on an answer. And we don't build those 767 00:46:16,880 --> 00:46:21,280 misconceptions. There are even there are even 768 00:46:21,560 --> 00:46:24,800 tools out there that are coaching teachers just on that 769 00:46:24,800 --> 00:46:27,520 kind of idea as well. So closing that feedback loop. 770 00:46:28,080 --> 00:46:31,240 I think the other thing with this is scalability. 771 00:46:32,480 --> 00:46:36,640 There's a massive jump to start off with in trying to integrate 772 00:46:36,640 --> 00:46:39,400 these, whether you may need more humans in the loop, but over 773 00:46:39,400 --> 00:46:42,600 time these will be scalable. And I think that's interesting. 774 00:46:43,200 --> 00:46:48,920 And the third thing for me is actually how you can get really 775 00:46:48,920 --> 00:46:53,680 accurate diagnosis of where every child is and where the 776 00:46:53,680 --> 00:46:56,680 gaps are or where the commonalities are for those 777 00:46:56,680 --> 00:46:58,880 pupils. I think that could be really 778 00:46:59,440 --> 00:47:04,840 interesting insights that will help a teacher because I think 779 00:47:04,840 --> 00:47:08,200 learning teaching is difficult because the learning isn't 780 00:47:08,200 --> 00:47:11,240 always visible. And the more data points that we 781 00:47:11,240 --> 00:47:14,480 get for high quality to identify what are those misconceptions 782 00:47:14,480 --> 00:47:18,440 that come in or where those gaps of knowledge that we can address 783 00:47:19,240 --> 00:47:22,360 something really concrete that could be amazing. 784 00:47:22,360 --> 00:47:26,640 And then we address that through space retrieval as well. 785 00:47:27,400 --> 00:47:31,800 That could really help us as teachers, just having that raw 786 00:47:31,800 --> 00:47:35,560 data, being able to action it as well, both in in class and our 787 00:47:35,560 --> 00:47:40,160 teaching, but also through whether it's extra time on the 788 00:47:40,200 --> 00:47:41,480 AI tutor or other things as well. 789 00:47:41,480 --> 00:47:44,280 I think else be important. Are there any hopes from you 790 00:47:44,280 --> 00:47:48,000 where you think AI tutors could be really purposeful? 791 00:47:48,440 --> 00:47:50,720 Just building on that last one around after your diagnosis, 792 00:47:50,720 --> 00:47:54,920 just thinking about how that hand over meeting that you get 793 00:47:54,920 --> 00:47:59,520 sort of certainly in sort of in primary or for the secondary 794 00:47:59,960 --> 00:48:01,880 teaching. If it's a, you know, newest 795 00:48:01,880 --> 00:48:04,760 teaching year 8 or year 9. Like how useful that would be to 796 00:48:04,760 --> 00:48:10,480 have sort of knowledge graph of each child's personal learning 797 00:48:10,480 --> 00:48:13,720 journey. And you know, hopefully and you 798 00:48:13,720 --> 00:48:18,040 know, undoubtedly you would then be able to talk to the system 799 00:48:18,040 --> 00:48:22,440 and it would just tell you class 60s, seventies, 80s, you know, 800 00:48:22,720 --> 00:48:26,040 whatever it is, you know, main sort of knowledge gaps and stuff 801 00:48:26,040 --> 00:48:28,320 like that. So you don't have to do too much 802 00:48:28,320 --> 00:48:32,600 of the sieving through the data yourself. 803 00:48:32,600 --> 00:48:36,480 And I think that's a really sort of interesting bit around that. 804 00:48:38,360 --> 00:48:41,520 I think in terms for me, I think what I'm sort of hopeful about 805 00:48:41,560 --> 00:48:46,720 is that certainly I think the EED study is really powerful 806 00:48:46,720 --> 00:48:49,800 because it just kind of it keeps the human there. 807 00:48:49,800 --> 00:48:53,440 And I think that's something to be really interested in, I think 808 00:48:53,440 --> 00:48:57,280 for, you know, for teachers who potentially might be fearful 809 00:48:57,280 --> 00:48:59,760 that the role of the teacher might change. 810 00:48:59,760 --> 00:49:03,960 I think this is kind of evidence that points towards that, you 811 00:49:03,960 --> 00:49:09,800 know, teacher expertise within your subjects is still important 812 00:49:09,800 --> 00:49:12,600 right now. And this study only worked only 813 00:49:12,600 --> 00:49:16,280 got the results that it did because it had those those 814 00:49:16,280 --> 00:49:19,520 experts there that kept them going, which I think is really, 815 00:49:19,920 --> 00:49:26,040 really useful signal to the profession that keep on keep on 816 00:49:26,040 --> 00:49:28,960 getting better at teaching and keep on improving your subject 817 00:49:28,960 --> 00:49:31,960 knowledge because that's going to, you know, only help pupils 818 00:49:32,840 --> 00:49:36,120 learn more and potentially faster too. 819 00:49:37,560 --> 00:49:43,440 The other thing that I think is potentially quite interesting 820 00:49:43,440 --> 00:49:47,240 for me, I think there is a bit of an, again, we need to see how 821 00:49:47,240 --> 00:49:50,080 it all plays out and stuff, But again, linked to that sort of 822 00:49:50,080 --> 00:49:54,320 scalability of just the cost of it all and if you can get the 823 00:49:54,320 --> 00:49:57,960 conditions of it right. If you are ensuring that you are 824 00:49:58,200 --> 00:50:02,160 providing space for pupils who potentially can't do this at 825 00:50:02,160 --> 00:50:06,760 home or won't do it at home and you're providing space in the 826 00:50:06,760 --> 00:50:09,120 school day, well, that's sort of after school, a bit of a lunch 827 00:50:09,120 --> 00:50:11,800 club, whatever it might be. I know, you know, there's pros 828 00:50:11,800 --> 00:50:15,760 and cons to all of that, but I think we are sort of seeing 829 00:50:15,760 --> 00:50:19,600 that, you know, with every iteration of AI modelling, it 830 00:50:19,600 --> 00:50:24,920 gets cheaper to use and it is cheaper than human capital, you 831 00:50:24,920 --> 00:50:30,560 know, tutors £4045.00 an hour, this, you know, it's £20 a 832 00:50:30,560 --> 00:50:33,440 month. You know, there's an economic 833 00:50:33,440 --> 00:50:36,440 argument as well. I think that this potentially, 834 00:50:36,720 --> 00:50:40,640 again, all other conditions being right, which is a big if 835 00:50:41,200 --> 00:50:45,160 for sure. But I do think, you know, 836 00:50:45,320 --> 00:50:49,000 looking forward, you know, there's an economic argument 837 00:50:49,000 --> 00:50:54,160 that it's a, well, it's a, you know, potentially a useful spend 838 00:50:54,160 --> 00:50:56,920 of people, premium, for example. So I think in terms of your 839 00:50:56,920 --> 00:50:58,440 scales. So it's kind of more just adding 840 00:50:58,600 --> 00:51:01,240 I guess to to yours than having my own. 841 00:51:01,880 --> 00:51:03,440 No, that's good. I'll put you on the spot anyway. 842 00:51:03,440 --> 00:51:06,200 I just as you were talking, I was thinking about it anyway. 843 00:51:06,200 --> 00:51:09,600 But thank you for everyone listening to thinking Deeply 844 00:51:09,600 --> 00:51:12,640 about AI for Schools. I'm James Rabin, and joining me, 845 00:51:12,640 --> 00:51:15,680 as always, is Neil Almond. If you've enjoyed the 846 00:51:15,680 --> 00:51:21,920 conversation, scribe to Neil's newsletter at 4 schools dot AI. 847 00:51:21,920 --> 00:51:27,640 That's four schools dot AI for your weekly dose of AI news 848 00:51:27,640 --> 00:51:29,760 around education around the wider world. 849 00:51:29,760 --> 00:51:34,200 And so often a really good point for you to take and steal. 850 00:51:34,600 --> 00:51:39,120 Thank you. And next time we will be, I'm 851 00:51:39,120 --> 00:51:41,520 not sure what we're going to do next time, Neil, but there are 852 00:51:41,520 --> 00:51:43,560 hundreds of episodes that we could do. 853 00:51:43,640 --> 00:51:47,120 And I think we'll explore some more AI tutors and explore 854 00:51:47,120 --> 00:51:50,360 whether we can get some guests involved as well. 855 00:51:50,520 --> 00:51:53,520 Thank you as always, goodbye. Always a pleasure.