WEBVTT 00:00:08.339 --> 00:00:11.859 Opening AI for language learning explores artificial 00:00:11.859 --> 00:00:14.500 intelligence through informal conversations between 00:00:14.500 --> 00:00:17.019 two longtime colleagues in language education 00:00:17.019 --> 00:00:20.940 and digital technologies, Mat Schulze and Phil 00:00:20.940 --> 00:00:25.359 Hubbard. Language educators, learners, and enthusiasts 00:00:25.359 --> 00:00:28.160 are invited to an informed discussion where emerging 00:00:28.160 --> 00:00:32.280 technology meets human communication. Hello, 00:00:32.299 --> 00:00:34.840 everyone. We're back to recording an episode, 00:00:34.859 --> 00:00:39.359 and this time, We're again using an article as 00:00:39.359 --> 00:00:42.340 the basis for our discussion and exploration. 00:00:42.719 --> 00:00:46.619 So we're back to the old tricks. Yes, we'll be 00:00:46.619 --> 00:00:49.820 doing some archaeology, exploring the inner workings 00:00:49.820 --> 00:00:53.479 of John Searle's Chinese room thought experiment. 00:00:54.899 --> 00:01:00.719 Why? A couple of reasons. I'm still stuck on 00:01:00.719 --> 00:01:02.840 this archaeology. That's an interesting kind 00:01:02.840 --> 00:01:07.010 of metaphor. So we're digging deep. I really 00:01:07.010 --> 00:01:10.230 find this paper by John Searle came out in 1980. 00:01:10.430 --> 00:01:13.750 It's a really fascinating paper. As I said, it 00:01:13.750 --> 00:01:17.549 came out in 1980 in behavioral and brain sciences, 00:01:18.290 --> 00:01:20.909 volume number three, and we're still talking 00:01:20.909 --> 00:01:26.349 about it more than 40 years after. It has particularly 00:01:26.349 --> 00:01:30.930 come up in discussions about generative AI. Initially, 00:01:31.230 --> 00:01:34.290 in that journal, they call it a target article. 00:01:34.810 --> 00:01:39.769 It appeared together with 27 peer commentaries. 00:01:39.890 --> 00:01:43.090 It's almost a volume, basically. They're responding 00:01:43.090 --> 00:01:47.750 all to John Searle. What does he do? He rejects 00:01:47.750 --> 00:01:53.950 what is called strong AI. John Searle, as a philosopher, 00:01:54.349 --> 00:01:58.579 basically says, the computer is not a mind. and 00:01:58.579 --> 00:02:02.239 therefore does not understand. But he accepts 00:02:02.239 --> 00:02:05.260 weak AI and says the computer is a good tool 00:02:05.260 --> 00:02:09.039 to understand the mind. So it can mimic certain 00:02:09.039 --> 00:02:13.719 things in some way. It can simulate understanding. 00:02:14.439 --> 00:02:18.300 And I believe this is of utmost relevance in 00:02:18.300 --> 00:02:21.000 education today, particularly for a lecture. 00:02:21.740 --> 00:02:24.219 So he also rejects the Turing test. We talked 00:02:24.219 --> 00:02:28.120 about the Turing test in episode nine. This was 00:02:28.120 --> 00:02:30.319 AI as the imitation game, because that's what 00:02:30.319 --> 00:02:34.620 Turing calls it. And Searle basically argues 00:02:34.620 --> 00:02:38.360 a conversation is not a good measure for intelligence. 00:02:38.580 --> 00:02:41.539 Just because somebody can hold their own in a 00:02:41.539 --> 00:02:43.199 conversation doesn't mean they're intelligent. 00:02:43.699 --> 00:02:46.900 Same for a machine. And most importantly, linguistic 00:02:46.900 --> 00:02:50.199 argument. He distinguishes between syntax, the 00:02:50.199 --> 00:02:56.020 form, and semantics, the meaning. This is also 00:02:56.020 --> 00:02:59.039 relevant, I would think, in kind of work without 00:02:59.039 --> 00:03:01.580 computers. Think about language assessment, for 00:03:01.580 --> 00:03:06.000 example. The GCSE exams in Britain, certainly 00:03:06.000 --> 00:03:09.960 in the 1990s when I lived there, a lot of these 00:03:09.960 --> 00:03:13.280 kids wrote, learned little dialogues for the 00:03:13.280 --> 00:03:16.379 exam. They would then rehearse them in the exam, 00:03:17.280 --> 00:03:20.900 basically, and the moment the teacher would divert 00:03:20.900 --> 00:03:23.560 from the script, basically, that they had learned, 00:03:24.020 --> 00:03:26.479 they couldn't continue the dialogue, basically. 00:03:26.680 --> 00:03:30.800 They were screwed, literally, in the exam. So 00:03:30.800 --> 00:03:33.900 syntactic fluency through rote learning is not 00:03:33.900 --> 00:03:36.319 necessarily understanding, even for humans, not 00:03:36.319 --> 00:03:41.780 just for a machine. Okay. So what exactly is 00:03:41.780 --> 00:03:45.909 the Chinese Room? Yeah, he calls it the Chinese 00:03:45.909 --> 00:03:48.969 room argument, right? So the Chinese room is 00:03:48.969 --> 00:03:52.870 a room, obviously. John Searle was an American 00:03:52.870 --> 00:03:55.530 philosopher and he taught at Berkeley, at the 00:03:55.530 --> 00:03:59.150 University of California, Berkeley. Major figure 00:03:59.150 --> 00:04:06.080 in philosophy of language and mind. I came across 00:04:06.080 --> 00:04:09.419 his name first in the context of speech act theory. 00:04:09.919 --> 00:04:12.939 Austin and Searle are the two main ones for speech 00:04:12.939 --> 00:04:16.100 act theories. He talked about intentionality 00:04:16.100 --> 00:04:20.000 and things like that. In his famous thought experiment, 00:04:20.120 --> 00:04:22.699 basically, of the Chinese room, his argument 00:04:22.699 --> 00:04:27.800 that he, John Searle, is in what he calls the 00:04:27.800 --> 00:04:31.000 Chinese room. It's a room that doesn't have any 00:04:31.000 --> 00:04:33.139 windows, just one little slot where you can kind 00:04:33.139 --> 00:04:36.220 of slide in pieces of paper basically with notes 00:04:36.220 --> 00:04:41.459 and questions on them. John says of himself that 00:04:41.459 --> 00:04:45.399 he does not speak Chinese. Outside of the room 00:04:45.399 --> 00:04:48.000 are Chinese speakers, first language speakers, 00:04:48.639 --> 00:04:51.300 who ask questions and write them down on little 00:04:51.300 --> 00:04:53.980 notes, throw them in through a kind of letterbox 00:04:53.980 --> 00:04:58.860 like thing. He picks them up. He has all sorts 00:04:58.860 --> 00:05:03.699 of manuals in the Chinese room that help him 00:05:03.699 --> 00:05:08.199 transfer those Chinese characters from the questions 00:05:08.199 --> 00:05:12.439 into Chinese characters for answers. And he does 00:05:12.439 --> 00:05:15.500 that without knowing Chinese, without being able 00:05:15.500 --> 00:05:17.879 to speak or write Chinese. He just uses those 00:05:17.879 --> 00:05:21.360 manuals that help him basically. And then he 00:05:21.360 --> 00:05:23.860 writes them out. The answer is basically what 00:05:23.860 --> 00:05:26.379 he finds in the manuals, puts this back, and 00:05:26.379 --> 00:05:29.850 the Chinese people, speakers outside of the room 00:05:29.850 --> 00:05:33.069 believe there's another Chinese speaker in the 00:05:33.069 --> 00:05:36.569 room, hence the name Chinese Room, basically. 00:05:37.670 --> 00:05:41.269 All right. Yeah, what do you think? I know you've 00:05:41.269 --> 00:05:45.290 read the paper, too. Do you also think it shows 00:05:45.290 --> 00:05:49.250 anything that is important to know? Or is it 00:05:49.250 --> 00:05:52.069 just interesting for language educators? What's 00:05:52.069 --> 00:05:55.110 your take? Well, I think it's still provocative. 00:05:55.529 --> 00:05:59.029 So here's what I think it does well. It makes 00:05:59.029 --> 00:06:02.449 a plausible case for a system to be able to generate 00:06:02.449 --> 00:06:06.730 a useful language interaction by manipulating 00:06:06.730 --> 00:06:09.490 rules without understanding the meaning behind 00:06:09.490 --> 00:06:13.610 them. It's almost like me following a recipe 00:06:13.610 --> 00:06:16.350 for something I've never cooked. I'm actually 00:06:16.350 --> 00:06:19.389 not a very good cook, but I can follow instructions. 00:06:20.410 --> 00:06:23.589 Or maybe like putting together a piece of IKEA 00:06:23.589 --> 00:06:28.250 furniture. where all of these arrows and representations 00:06:28.250 --> 00:06:32.550 of screws and bolts somehow come together and 00:06:32.550 --> 00:06:36.069 the desk I bought turns out to look like a desk. 00:06:37.069 --> 00:06:40.129 That said, I think it falls short for language 00:06:40.129 --> 00:06:44.009 learning rather than language use. And of course, 00:06:44.709 --> 00:06:47.810 there are all sorts of criticisms of it even 00:06:47.810 --> 00:06:52.269 for language use. So I asked my good friend Claude 00:06:52.269 --> 00:06:57.740 Opus 5 about this. And it extruded that, for 00:06:57.740 --> 00:07:01.819 example, working with comprehensible input, which 00:07:01.819 --> 00:07:05.180 is, I think, pretty well established as important 00:07:05.180 --> 00:07:08.860 in language learning, doesn't require a comprehending 00:07:08.860 --> 00:07:13.779 source. So the meaning is constructed in the 00:07:13.779 --> 00:07:16.920 mind of the learner. To me, that means even if 00:07:16.920 --> 00:07:20.300 you accept Searle's claim that face value, that 00:07:20.300 --> 00:07:24.170 there's no intelligence involved, It may not 00:07:24.170 --> 00:07:26.910 matter that much, at least in some use cases. 00:07:27.649 --> 00:07:32.230 I think it comes down to this weak AI sense of 00:07:32.230 --> 00:07:37.430 how good a simulation is it. And our simulations 00:07:37.430 --> 00:07:41.629 are getting better and better. Right. This idea 00:07:41.629 --> 00:07:45.730 of weaving in comprehensible input there and 00:07:45.730 --> 00:07:49.370 basically saying that let's just think of a person. 00:07:49.470 --> 00:07:53.220 It doesn't always have to be machined. uttering 00:07:53.220 --> 00:07:56.079 what becomes the comprehensible input according 00:07:56.079 --> 00:07:58.620 to Krashen basically doesn't have to understand 00:07:58.620 --> 00:08:01.759 themselves what they're saying basically. That's 00:08:01.759 --> 00:08:04.279 pretty much what you said is an interesting idea. 00:08:05.079 --> 00:08:08.360 Yes, there are a couple of, I called it a fascinating 00:08:08.360 --> 00:08:13.100 paper, there are a couple of things that I would 00:08:13.100 --> 00:08:15.550 consider shortcoming too. There's always couple 00:08:15.550 --> 00:08:18.089 of things. Some of them were pointed out already 00:08:18.089 --> 00:08:20.910 in these, I mentioned the 27 commentaries. They 00:08:20.910 --> 00:08:24.889 also targeted, hence the name target article, 00:08:25.050 --> 00:08:28.449 I guess, targeted a number of those shortcomings. 00:08:30.670 --> 00:08:34.070 Talking about learning first, I was always puzzled 00:08:34.070 --> 00:08:37.309 by thinking, gosh, if somebody really were to 00:08:37.309 --> 00:08:41.750 use so many manuals for looking up so many things 00:08:41.750 --> 00:08:44.799 and kind of generating these answers based on 00:08:44.799 --> 00:08:48.919 questions. Just because of the frequency of the 00:08:48.919 --> 00:08:52.039 activity, some learning should probably take 00:08:52.039 --> 00:08:55.460 place. Something is going to stick in some way. 00:08:57.860 --> 00:08:59.960 The other one, and this is the most important 00:08:59.960 --> 00:09:02.139 one that has been made right from the beginning, 00:09:02.580 --> 00:09:05.399 the salt experiment. it's kind of a little bit 00:09:05.399 --> 00:09:09.220 difficult to comprehend in some way. If you think 00:09:09.220 --> 00:09:12.799 of those manuals, which Searle clearly, 1980, 00:09:13.399 --> 00:09:16.779 thinks of as book literal manuals, basically, 00:09:17.899 --> 00:09:21.379 there will be no room big enough to hold all 00:09:21.379 --> 00:09:24.240 those manuals. Searle would have known this because 00:09:24.240 --> 00:09:28.440 we know this since 1957 and Chomsky's review 00:09:28.440 --> 00:09:31.320 of Skinner's book basically, then the number 00:09:31.320 --> 00:09:34.220 of utterances in any language is infinite, right? 00:09:34.220 --> 00:09:37.779 So you would have infinitely many rules of transposing 00:09:37.779 --> 00:09:41.539 those question utterance into an answer utterance 00:09:41.539 --> 00:09:45.500 basically. But it underscores the power of these 00:09:45.500 --> 00:09:47.360 large language models because that's essentially 00:09:47.360 --> 00:09:53.759 what they kind of do in some shape or form. Okay, 00:09:53.779 --> 00:09:56.679 that's a bit about the pros and cons of the paper. 00:09:58.379 --> 00:10:01.120 Before we continue, I did want to make an aside 00:10:01.120 --> 00:10:04.519 for those linguists out there. John Searle is 00:10:04.519 --> 00:10:08.460 also credited with speech act theory, which of 00:10:08.460 --> 00:10:10.860 course is a critical part of moving from just 00:10:10.860 --> 00:10:14.500 structure to pragmatics in language learning 00:10:14.500 --> 00:10:20.720 and language use. Back to the point here, why 00:10:20.720 --> 00:10:25.019 should world language teachers care today about 00:10:25.019 --> 00:10:31.980 this? I think I've talked about teacher reflection 00:10:31.980 --> 00:10:34.639 a number of times. This is also of the seven 00:10:34.639 --> 00:10:40.259 principles that we have in... sustained integrated 00:10:40.259 --> 00:10:42.799 professional development. This is kind of my 00:10:42.799 --> 00:10:46.059 favorite principle basically, this reflective 00:10:46.059 --> 00:10:48.559 element that this is very important. Teacher 00:10:48.559 --> 00:10:52.480 reflection is a very powerful tool. And in general, 00:10:52.879 --> 00:10:56.960 theory is a very good lens for reflection. Now, 00:10:56.980 --> 00:10:59.559 admittedly, John Searle, he's a philosopher of 00:10:59.559 --> 00:11:02.580 language. This is a very theoretical paper. It's 00:11:02.580 --> 00:11:06.100 not directly transferable into the language classroom, 00:11:06.100 --> 00:11:10.669 but it's a very good tool to do some reflection 00:11:10.669 --> 00:11:18.049 in some way. If I compare Alan Turing with the 00:11:18.049 --> 00:11:20.049 Turing test, what we call it now, what he called 00:11:20.049 --> 00:11:23.610 the imitation game, episode nine, remember, and 00:11:23.610 --> 00:11:27.049 John Searle and his Chinese room argument, I 00:11:27.049 --> 00:11:30.570 think John Searle makes it clearer that we are 00:11:30.570 --> 00:11:34.350 dealing with a machine that mainly manipulates 00:11:34.350 --> 00:11:38.129 form. And really that's the main thing to remember 00:11:38.129 --> 00:11:41.210 in all those reflections. If you have that at 00:11:41.210 --> 00:11:43.509 the back of your mind, you can experiment and 00:11:43.509 --> 00:11:49.169 can do things. But if we lose sense of what we're 00:11:49.169 --> 00:11:51.830 interacting with and not who we're interacting 00:11:51.830 --> 00:11:56.690 with, that's often when problems start. So I 00:11:56.690 --> 00:11:58.669 would kind of, it's a little bit polemic, but 00:11:58.669 --> 00:12:02.750 I would say machines are tools which we use and 00:12:02.750 --> 00:12:05.230 they can be very sophisticated. And I do not 00:12:05.230 --> 00:12:08.129 see them as partners. I do not see them as agents, 00:12:08.470 --> 00:12:15.450 basically. OK, well, I'd like to, on a practical 00:12:15.450 --> 00:12:19.289 level, defend the idea of the chat bot as a language 00:12:19.289 --> 00:12:22.590 learning partner. So it seems to me there are 00:12:22.590 --> 00:12:25.809 at least three roles for a chat bot. And the 00:12:25.809 --> 00:12:31.049 one you're, I think, alluding to more in terms 00:12:31.049 --> 00:12:38.210 of the more machine linked part is when it's 00:12:38.210 --> 00:12:43.169 a conversation partner for practice. So the student 00:12:43.169 --> 00:12:47.009 has already learned something in class, has a 00:12:47.009 --> 00:12:50.490 basic vocabulary for dealing with a situation, 00:12:52.750 --> 00:12:55.750 presumably some idea of the grammar structures, 00:12:55.809 --> 00:12:58.470 because this is a formal class. So if you link 00:12:58.470 --> 00:13:02.740 it to coursework, That role means you're trying 00:13:02.740 --> 00:13:05.679 to get better at things you already sort of know. 00:13:05.840 --> 00:13:08.460 You're trying to make them more internalized, 00:13:08.779 --> 00:13:13.000 more automatized, and maybe also figure out some 00:13:13.000 --> 00:13:16.299 things you mislearned with the feedback that 00:13:16.299 --> 00:13:19.379 you might be getting from that conversation partner. 00:13:22.360 --> 00:13:24.960 There's also the role of a conversation partner 00:13:24.960 --> 00:13:27.419 to support more informal learning, and that's 00:13:27.419 --> 00:13:30.529 something I'll be talking about later, I think 00:13:30.529 --> 00:13:33.370 today. And then of course, there's actually the 00:13:33.370 --> 00:13:37.809 role of a tutor where it's replacing the teacher 00:13:37.809 --> 00:13:41.970 or at least augmenting the teacher in this role 00:13:41.970 --> 00:13:45.850 where the chat bot itself has a teaching presence. 00:13:46.610 --> 00:13:49.889 This is very common we see in language, AI language 00:13:49.889 --> 00:13:53.809 apps today. For example, Speakology, which is 00:13:53.809 --> 00:13:57.750 one that we featured in our second episode from 00:13:57.750 --> 00:14:00.870 Calico, the one that won the Launchpad Awards. 00:14:01.350 --> 00:14:09.350 In that one, you have basically a synthetic visual 00:14:09.350 --> 00:14:13.250 presentation of a language tutor and someone 00:14:13.250 --> 00:14:17.409 who's simulating the experience of an online 00:14:17.409 --> 00:14:22.809 tutorial for the learner. So how do you think 00:14:22.809 --> 00:14:25.330 teachers can make sense of this and use it in 00:14:25.330 --> 00:14:29.200 a practical way? Roofing off this practical, 00:14:30.240 --> 00:14:33.980 I 100 % agree with you on the practice elements. 00:14:36.200 --> 00:14:39.379 As you know from our many other discussions, 00:14:39.519 --> 00:14:42.799 I'm still struggling somewhat with this idea 00:14:42.799 --> 00:14:45.179 of a machine also having a teaching presence. 00:14:45.360 --> 00:14:48.980 I see exactly what you mean in the sense of this 00:14:48.980 --> 00:14:53.049 tutorial call, but it's... It's a very fine line, 00:14:53.309 --> 00:14:56.110 right? I'm probably more on the worried side 00:14:56.110 --> 00:14:59.169 that people then go too far in their anthropomorphization. 00:14:59.570 --> 00:15:05.830 Beautiful word, difficult to pronounce. And the 00:15:05.830 --> 00:15:09.529 one thing I'm always thinking about in a way, 00:15:09.549 --> 00:15:13.129 and I think all language educators need to think 00:15:13.129 --> 00:15:17.409 about these days, is not so much only what is 00:15:17.409 --> 00:15:19.370 the role of the machine, but also what's the 00:15:19.370 --> 00:15:24.440 role of the teacher. In the age of generative 00:15:24.440 --> 00:15:27.000 AI, what can the teacher actually do? What should 00:15:27.000 --> 00:15:29.320 they do? What are they contributing? What does 00:15:29.320 --> 00:15:32.580 the teacher contribute in this interaction? And 00:15:32.580 --> 00:15:34.919 again, using a kind of theoretical angle there, 00:15:34.940 --> 00:15:37.519 I would refer to Meryl Swain, one of the kind 00:15:37.519 --> 00:15:42.659 of best known applied linguists. She coined the 00:15:42.659 --> 00:15:46.080 term languaging. What she means by that, and 00:15:46.080 --> 00:15:50.500 here I'm kind of quoting, she says, as I am using 00:15:50.500 --> 00:15:54.039 the term, refers to the process of making meaning 00:15:54.039 --> 00:15:58.320 and shaping knowledge and experience through 00:15:58.320 --> 00:16:02.740 language. Now, I would say this is the ultimate 00:16:02.740 --> 00:16:05.759 goal of language learning because you really 00:16:05.759 --> 00:16:09.059 want them to be part of another community. You're 00:16:09.059 --> 00:16:11.480 getting the key to another language community 00:16:11.480 --> 00:16:13.620 basically through your language learning. You 00:16:13.620 --> 00:16:16.659 also need the shared knowledge, the shared experience. 00:16:17.049 --> 00:16:19.509 in order to be part of that community and to 00:16:19.509 --> 00:16:23.230 use also language in a way as a medium to do 00:16:23.230 --> 00:16:27.629 this, literally as a vehicle to transport that 00:16:27.629 --> 00:16:34.710 meaning. Those things I would say you do best 00:16:34.710 --> 00:16:38.009 in interaction with other humans. Are there other 00:16:38.009 --> 00:16:39.750 ways? Of course we have libraries and things 00:16:39.750 --> 00:16:42.129 like that. We've had them for a while, but the 00:16:42.129 --> 00:16:46.519 best way still is interaction really with other 00:16:46.519 --> 00:16:48.860 people because you do want to be part of that 00:16:48.860 --> 00:16:51.320 group and there's kind of sometimes not that 00:16:51.320 --> 00:16:53.559 many shortcuts to kind of circumvent the group 00:16:53.559 --> 00:17:00.299 all the time. When it comes to practicing, yes, 00:17:00.519 --> 00:17:03.240 incidental vocabulary learning with this informal 00:17:03.240 --> 00:17:05.160 learning that you talked about earlier, Phil, 00:17:06.579 --> 00:17:09.460 is a very good example where the chatbook is 00:17:09.460 --> 00:17:12.329 a brilliant option. especially if used wisely 00:17:12.329 --> 00:17:14.890 and people realize, no, I'm kind of dealing with 00:17:14.890 --> 00:17:18.170 a patient and funny machine that I can manipulate 00:17:18.170 --> 00:17:21.650 and do strange things with. So it's an additional 00:17:21.650 --> 00:17:24.549 language practice that can be very individual. 00:17:25.089 --> 00:17:27.750 All students in however big the classroom is 00:17:27.750 --> 00:17:29.849 can do it at the same time, as long as they have 00:17:29.849 --> 00:17:34.490 a device to interact. The linguistic accuracy 00:17:34.490 --> 00:17:40.539 of the chatbot, synthetic text is very good basically. 00:17:41.480 --> 00:17:45.460 So they're getting pretty good exposure basically 00:17:45.460 --> 00:17:48.779 to a lot of linguistic forms and can one pick 00:17:48.779 --> 00:17:54.720 up things there for sure. Okay well I have another 00:17:54.720 --> 00:17:59.220 practical example that moves us into this realm 00:17:59.220 --> 00:18:03.519 of informal language learning and as you know 00:18:03.519 --> 00:18:06.250 in some of my work I've really push the idea 00:18:06.250 --> 00:18:08.690 of how important it is to get out of the classroom, 00:18:08.849 --> 00:18:12.089 and even to get away from the teacher's influence, 00:18:12.190 --> 00:18:15.490 especially if you want to have things more personalized 00:18:15.490 --> 00:18:18.289 for you as a learner. That doesn't mean we don't 00:18:18.289 --> 00:18:21.549 need teachers, but it changes the teacher's role 00:18:21.549 --> 00:18:25.869 potentially. So I think this one's related to 00:18:25.869 --> 00:18:28.470 Swain's languaging, keeping in mind that that 00:18:28.470 --> 00:18:33.130 concept was presented 20 years ago. or even trans 00:18:33.130 --> 00:18:36.589 -languaging, which is more the way I see my interaction. 00:18:37.490 --> 00:18:41.430 So let me go ahead and do this, but I do want 00:18:41.430 --> 00:18:45.369 to emphasize along your lines that the meaning 00:18:45.369 --> 00:18:48.990 here, the creation of the meaning is only on 00:18:48.990 --> 00:18:51.529 the human side. It's only on the learner's side. 00:18:52.089 --> 00:18:55.190 So even though you're having this simulated interaction 00:18:55.190 --> 00:18:58.470 and it can be linguistically and in some cases 00:18:58.470 --> 00:19:04.019 even culturally rich, because the chatbot has 00:19:04.019 --> 00:19:08.180 access to all this cultural information. It's 00:19:08.180 --> 00:19:11.559 not the same as the human. And I'll try to remember 00:19:11.559 --> 00:19:14.500 to give an example of that right at the end of 00:19:14.500 --> 00:19:18.299 this. So I ran this little experiment just a 00:19:18.299 --> 00:19:23.759 couple of days ago with chat GPT 5 .6. For background, 00:19:24.000 --> 00:19:27.359 I have high school German from audio lingual 00:19:27.359 --> 00:19:32.740 method classes. in language labs. This is my 00:19:32.740 --> 00:19:36.500 base in German. It's mostly been unused and forgotten. 00:19:36.920 --> 00:19:39.980 The first time I had a chance to use it was, 00:19:39.980 --> 00:19:43.960 gosh, almost 30 years after I first learned it. 00:19:45.220 --> 00:19:50.279 Anyway, I asked ChatGPT to play the role of a 00:19:50.279 --> 00:19:54.099 German friend of mine in Bremen, telling me about 00:19:54.099 --> 00:19:56.880 the soccer team there. I know a little bit about 00:19:56.880 --> 00:20:00.190 Werder Bremen. And I know a lot about soccer 00:20:00.190 --> 00:20:03.069 from the player perspective. I played it for 00:20:03.069 --> 00:20:06.470 over 30 years, but I don't really follow the 00:20:06.470 --> 00:20:10.349 teams. And especially, I don't know the German 00:20:10.349 --> 00:20:16.710 terms. So I presented myself as roughly an A2 00:20:16.710 --> 00:20:19.130 learner, but said in the prompt that I would 00:20:19.130 --> 00:20:22.450 be using English sometimes. And then I wanted 00:20:23.450 --> 00:20:29.349 to respond to me in German, but often by recasting 00:20:29.349 --> 00:20:33.430 or or using the correct word. I asked it not 00:20:33.430 --> 00:20:37.250 to correct errors directly. So having done that 00:20:37.250 --> 00:20:41.470 in print, I then switched to GPT live advanced 00:20:41.470 --> 00:20:44.890 voice mode. This is something relatively new. 00:20:45.390 --> 00:20:47.910 The live mode in chat GPT's been around awhile, 00:20:47.910 --> 00:20:50.640 but in this one. There's this little blue ball 00:20:50.640 --> 00:20:53.000 that's throbbing on your screen to let you know 00:20:53.000 --> 00:20:56.140 you're just interacting directly in voice. So 00:20:56.140 --> 00:20:58.799 in roughly two, two and a half minute interaction, 00:20:59.619 --> 00:21:05.119 I learned informally by speaking to the chat 00:21:05.119 --> 00:21:08.220 bot and listening to it. At that point, there 00:21:08.220 --> 00:21:11.380 was no text coming back. It was all oral interaction. 00:21:12.099 --> 00:21:15.559 So I learned sports bar. I used the English word 00:21:15.559 --> 00:21:18.789 sports bar. It came back. with the German sports 00:21:18.789 --> 00:21:24.450 bar, Mittelfeld for midfield. That was easy to 00:21:24.450 --> 00:21:28.890 pick up. Fußballmannschaft, which I just got 00:21:28.890 --> 00:21:31.230 from, I didn't even use the word team. It popped 00:21:31.230 --> 00:21:34.130 up with this and I go, okay, that must be team. 00:21:34.589 --> 00:21:39.589 And of course, pass for pass. I remembered from 00:21:39.589 --> 00:21:43.210 my old German when it's mentioned something. 00:21:43.559 --> 00:21:47.359 related, used langsam, that that meant slow. 00:21:48.039 --> 00:21:51.119 I also learned something new, which is that when 00:21:51.119 --> 00:21:55.339 I use the word gold, it replied with tor. And 00:21:55.339 --> 00:21:58.440 this is where the human side comes in. I wanted 00:21:58.440 --> 00:22:00.880 to make sure that wasn't an AI hallucination 00:22:00.880 --> 00:22:06.119 of some sort. And so I asked Mat, is that indeed 00:22:06.119 --> 00:22:09.559 the word for tor? And then because we were in 00:22:09.559 --> 00:22:14.049 a sports bar, I said, you know, ich wünsche I'm 00:22:14.049 --> 00:22:18.950 beer, I'm Paulaner. And it didn't correct me 00:22:18.950 --> 00:22:22.910 exactly. It just in a friendly way told me, you 00:22:22.910 --> 00:22:27.930 can say, you know, man kann es auch, I'm Paulaner 00:22:27.930 --> 00:22:31.309 bitter. Das ist ganz normal. So that's the regular 00:22:31.309 --> 00:22:35.630 way to say it. So I guess it couldn't help being 00:22:35.630 --> 00:22:38.789 helpful, even when I asked it not to. So the 00:22:38.789 --> 00:22:41.150 point to me is I didn't just practice. Yeah, 00:22:41.150 --> 00:22:44.500 I had some German in there. But I never tried 00:22:44.500 --> 00:22:46.779 to apply it to soccer. I didn't do any readings 00:22:46.779 --> 00:22:49.740 on soccer in German ahead of this. I just did 00:22:49.740 --> 00:22:54.359 it the way I might if I was in the real situation. 00:22:54.660 --> 00:22:59.200 And to me, the fact that I got stuff that, coming 00:22:59.200 --> 00:23:03.759 back to what's learning, hopefully there are 00:23:03.759 --> 00:23:08.440 persistent changes in my brain that happened 00:23:08.440 --> 00:23:11.819 because of this interaction. And those are useful. 00:23:12.000 --> 00:23:17.480 changes, and they do no harm. That's intriguing 00:23:17.480 --> 00:23:21.920 to me. So this is what I keep doing is trying 00:23:21.920 --> 00:23:24.440 to have these little experiments and seeing what 00:23:24.440 --> 00:23:27.779 I get out of them. They don't all work as well 00:23:27.779 --> 00:23:31.920 as this one did. I'm sure it helped that it was 00:23:31.920 --> 00:23:33.579 German. I thought you would never do this. This 00:23:33.579 --> 00:23:37.759 was great. And then soccer in there. I mean, 00:23:37.859 --> 00:23:40.339 what more can you ask for? No, but seriously, 00:23:40.640 --> 00:23:44.480 it's I find it very interesting. I had not thought 00:23:44.480 --> 00:23:50.099 earlier about this incidental vocabulary learning 00:23:50.099 --> 00:23:52.740 aspect in there and kind of picking up things. 00:23:53.019 --> 00:23:55.819 Yes, of course that is learning in an informal 00:23:55.819 --> 00:24:00.240 way. I'm not aware that this has been discussed 00:24:00.240 --> 00:24:04.339 in very comprehensive by now literature about 00:24:04.339 --> 00:24:06.440 generative AI and language learning. I think 00:24:06.440 --> 00:24:11.670 that's very fruitful. avenue, mixed metaphor, 00:24:12.130 --> 00:24:14.750 to kind of go down in further research. I think 00:24:14.750 --> 00:24:16.990 it would be interesting to see how they're actually 00:24:16.990 --> 00:24:20.130 picking up, how students actually pick up things. 00:24:20.190 --> 00:24:22.829 And that's probably, as you said, level -dependent 00:24:22.829 --> 00:24:26.849 and all sorts of things. My main goal, and that's 00:24:26.849 --> 00:24:29.750 why I think I'm sometimes a little more polemic 00:24:29.750 --> 00:24:32.269 in those things. My main goal with this podcast 00:24:32.269 --> 00:24:34.349 is, and that's why I also like to have those 00:24:34.349 --> 00:24:36.980 articles in there and make... hopefully some 00:24:36.980 --> 00:24:39.759 listeners aware of this, is to help to create 00:24:39.759 --> 00:24:42.720 a little more awareness of the role and impact 00:24:42.720 --> 00:24:46.319 of generative AI. Because we all lead busy lives, 00:24:46.559 --> 00:24:49.759 there's a lot going on in the professional life 00:24:49.759 --> 00:24:53.980 and the other life of a teacher, so to have those 00:24:53.980 --> 00:24:56.940 little moments of reflection I believe is very 00:24:56.940 --> 00:25:01.019 useful. We'll certainly put in links in the show 00:25:01.019 --> 00:25:04.049 notes, also the The paper is widely available. 00:25:04.849 --> 00:25:06.890 For a philosophy paper is actually quite readable, 00:25:06.950 --> 00:25:11.710 I would say. The commentaries were really done 00:25:11.710 --> 00:25:15.589 by who is who of official intelligence at the 00:25:15.589 --> 00:25:20.130 time in 1980. But not only, this was also fairly 00:25:20.130 --> 00:25:23.190 many people in cognitive science, philosophy. 00:25:23.549 --> 00:25:25.950 And psychology, which makes this really interesting 00:25:25.950 --> 00:25:28.769 because I think psychology particularly is an 00:25:28.769 --> 00:25:31.230 angle that's very often missing in this whole 00:25:31.230 --> 00:25:33.930 discussion of artificial intelligence, right? 00:25:33.930 --> 00:25:36.910 Because there's a lot of knowledge from there. 00:25:38.890 --> 00:25:42.549 So, you know, just based on this little experience 00:25:42.549 --> 00:25:45.069 and I've done some other things I've also reported 00:25:45.069 --> 00:25:50.349 and I think we'll in later episodes report projects 00:25:50.349 --> 00:25:55.630 I've done along these lines. Try it out with 00:25:55.630 --> 00:26:01.789 your chat bot, whatever it is. And if you're 00:26:01.789 --> 00:26:06.789 using an app that's a language learning focused 00:26:06.789 --> 00:26:11.589 app, try to see not just how it works the way 00:26:11.589 --> 00:26:14.809 it was designed to along some strict line, but 00:26:14.809 --> 00:26:20.950 how you can work with it to make it even better. 00:26:22.400 --> 00:26:25.099 Transfer that information to the students. See 00:26:25.099 --> 00:26:29.680 how it can help. To me, it's this experience. 00:26:29.880 --> 00:26:31.880 It's not reading about it. It's not looking at 00:26:31.880 --> 00:26:35.819 demos. Those are both really important, but you 00:26:35.819 --> 00:26:39.480 got to do it too. Right. Yes, it's about the 00:26:39.480 --> 00:26:43.579 doing and I'll make an additional promise. we'll 00:26:43.579 --> 00:26:46.640 lift off from the 1980s next time when we're 00:26:46.640 --> 00:26:50.000 talking. We won't be stuck in a time warp or 00:26:50.000 --> 00:26:54.319 something like that. But perhaps now you can 00:26:54.319 --> 00:26:57.160 at least this Chinese room argument comes up 00:26:57.160 --> 00:27:00.000 also in small talk sometimes and different conversations. 00:27:00.440 --> 00:27:03.059 Use it as a kind of tidbit whenever you kind 00:27:03.059 --> 00:27:06.319 of talk about those things. And it's a very useful 00:27:06.319 --> 00:27:10.400 tool for our own reflection as language teachers. 00:27:11.339 --> 00:27:13.619 Yeah, you can especially talk about it if you're 00:27:13.619 --> 00:27:16.519 talking to Matt I love talking about things 00:27:16.519 --> 00:27:23.539 like that. And last but not least, on a more 00:27:23.539 --> 00:27:26.460 of a meta note. We're trying to get this podcast 00:27:26.460 --> 00:27:29.240 out to more language teachers, teacher educators, 00:27:29.720 --> 00:27:31.559 but also language learners, language enthusiasts. 00:27:32.160 --> 00:27:35.059 The one thing we're told as novices in the field 00:27:35.059 --> 00:27:39.039 of podcasting that reviews on Apple podcasts 00:27:39.039 --> 00:27:42.539 are very, very helpful to kind of make this particular 00:27:42.539 --> 00:27:46.079 podcast a little more known. Of course, subscriptions 00:27:46.079 --> 00:27:49.430 help greatly. They don't cost you anything. We're 00:27:49.430 --> 00:27:52.609 totally ad free. We're human voice produced and 00:27:52.609 --> 00:27:55.049 we like it that way. We're just chatting amongst 00:27:55.049 --> 00:27:58.309 ourselves and not generating things because we 00:27:58.309 --> 00:28:01.250 believe artificial intelligence is no substitute 00:28:01.250 --> 00:28:06.690 for natural stupidity. See you next time. Bye 00:28:06.690 --> 00:28:17.789 for now. We are grateful for the support for 00:28:17.789 --> 00:28:20.650 opening AI for language learning by the Language 00:28:20.650 --> 00:28:22.990 and Applied Research Center at San Diego State 00:28:22.990 --> 00:28:25.710 University and the Southern Area International 00:28:25.710 --> 00:28:28.809 Languages Network, SAILN, which is part of the 00:28:28.809 --> 00:28:32.069 California World Languages Project. Mari Ocando Finol 00:28:32.069 --> 00:28:34.430 is the production coordinator of OAILL 00:28:34.529 --> 00:28:37.869 Our theme music is by Tillmann Spiegl. Our editor 00:28:37.869 --> 00:28:41.230 is Chris Brown. Live conversations are moderated 00:28:41.230 --> 00:28:44.470 and promoted by me, Shahnaz Ahmadeian. Until next 00:28:44.470 --> 00:28:44.769 time.