From Academic Text to Talk-Show: Deepening Engagement and Understanding with Google NotebookLM

* * * On the Internet * * *

February 2025 — Volume 28, Number 4

https://doi.org/10.55593/ej.28112int

Marie Alina Yeo
SEAMEO RELC, Singapore
<marie.yeorelc.org.sg>

Benjamin Luke Moorhouse
City University of Hong Kong, Hong Kong, China
<Bmoorhouum.cityu.edu.hk>

Yuwei Wan
Hong Kong Baptist University, Hong Kong, China
<yuweiwanlife.hkbu.edu.hk>

Abstract

This paper looks at Google’s NotebookLM, an AI-powered research assistant tool that can represent dense academic content in a range of output modes, like FAQs, timelines, study guides, and, most uniquely, as “Deep Dive” discussions. The discussions mimic a talk-show, where two AI-hosts unpack complex ideas from reading or audio texts, for example, a journal article or lecture, connecting them with related ideas. The talk-show format provides a relaxed, low-pressure way for learners to grasp key concepts. NotebookLM is especially helpful for those who may find academic reading overwhelming, as it transforms the reading process into a more enjoyable, multimodal experience. The paper describes and evaluates the discussion generation function in relation to principles of second language and digital learning. Overall, NotebookLM shows promise in helping learners to understand and engage with academic content more deeply and may be especially valuable for second language learners in English Medium Instruction (EMI) or bilingual education contexts which emphasize content development alongside language learning. However, the tendency of the AI-hosts to extract, embellish and augment the content may lead to misrepresentation of the original source text, raising epistemological concerns about the spread of misinformation.

Keywords: NotebookLM, AI-powered research assistant, AI research tool, multimodal research tool, AI-generated podcast, research management tool, deep-dive discussions

AI-research assistants are changing the way learners engage with academic texts. Among these tools, Google’s NotebookLM reimagines the way learners consume and comprehend complex learning materials. In its “Deep Dive” discussion function, which simulates a talk-show format, two AI hosts unpack ideas from source texts, expanding upon these ideas with related examples and explanations. This paper explores the effectiveness of this innovative feature in supporting learners, especially those in EMI or bilingual education settings. We begin by outlining the functionalities of NotebookLM, focusing on its distinctive audio discussion generation feature. Next, we evaluate this feature in relation to principles of second language and digital learning. We then offer practical suggestions on how to leverage NotebookLM’s discussion function in teaching and research, ending with some cautionary words about the importance of critically analyzing the content of the discussions to mitigate risks of the spread of misinformation.

Description of the Product

Initially released in July 2023, NotebookLM is currently available free of charge at https://notebooklm.google/. The webpage markets it as a “Personalized AI Research Assistant” and highlights four key capabilities: (i) summarizing texts and making connections among topics; (ii) providing instant insights through formats such as study guides, FAQs, and tables of contents; (iii) offering citations with exact quotes; and (iv) converting sources into “Deep Dive” conversations. This paper focuses on the fourth feature – the ability to create AI-hosted “Deep Dive” discussion podcasts between two interlocutors – as this feature is not commonly available in other research assistant tools.

While the website does not state who the target users are, the use cases listed for “power study,” “organize your thinking,” and “spark new ideas” suggest that it can benefit both academic and industry users alike. However, as it is touted as a “personalized AI research assistant,” anyone involved in knowledge management and creation, such as students, researchers, and presenters, seems the most likely potential users of this tool. According to the FAQ page, NotebookLM accepts text and audio input in 35 languages, so users with varying levels of English language proficiency will be able to use it with ease. However, at present, the conversation output is provided only in English, though the language level can be customized to suit the users’ English proficiency level. The website states that as NotebookLM is in its “early testing phase”, there is no access charge “at this time”.

Main Features of NotebookLM

Figure 1. Screenshot of How to Upload Texts and Formats Accepted

NotebookLM is easy and intuitive to use. Without the need for installation, users can simply go to the website and click “Try NotebookLM” to get started. They can then upload a single or multiple text sources in various formats (PDF, .txt, Markdown, Audio), from Google Drive, as a link, or as pasted text (See Figure 1). The current limit is 500,000 words per source.

As shown in Figure 2, once uploaded, users can click on “Notebook Guide” (A) to choose the format and mode of output (B) — FAQ, study guide, timeline, table of contents, and briefing notes— or generate a “Deep Dive” conversation (C) between two AI hosts in a talk-show style. As with other AI research assistant tools such as Elicit, Co-SciSpace, and Co-Pilot, readers can also “chat with” an article by asking the suggested questions (D) or use the chat box in ChatGPT-style to ask questions, request paraphrases or summaries complete with references (E).

Figure 2. Notebook Guide Tab (Bottom Right of Image) and Different Possible Outputs

Evaluation of NotebookLM

While an array of research assistant AI tools, such as those mentioned above, allow readers to “chat with” an article and can generate summaries with reference sources, Google’s NotebookLM, powered by Gemini, introduces a unique feature: the ability to help readers engage with and understand content more deeply by representing information in various output modes. While the FAQs, study guides, tables of contents, timelines, and briefing documents formats are useful in helping learners to distil vast amounts of information, it is the “Deep dive conversation” creation function that is distinctive and potentially powerful in supporting content and language learning especially for second language learners in EMI and bilingual education courses.

Theoretical Basis and Principles of Language Learning

One of the most demanding study tasks faced by EMI learners undertaking university-level undergraduate and graduate courses is reading academic texts. Academic staff and seasoned researchers may similarly find it challenging and time-consuming to capture, organize, distil and express or “CODE” (Forte, 2022) relevant ideas from dense academic texts. Even more difficult is for readers to make connections among multiple sources and derive usable insights for assignments, academic presentations, and publications. In short, the sheer volume of available information has made research and writing overwhelming, leaving many students, teachers, and researchers desperate for assistance in managing the academic reading and research process. NotebookLM can offer such support in four important ways: (i) providing exposure to the content through multiple formats and modes, making the content more accessible and memorable; (ii) adjusting the content’s scope and language level to accommodate users’ interests and levels; (iii) deepening understanding of the content by allowing users to “listen in” as third-party participants; and (iv) helping users make connections across language genres, such as transitioning from written academic text to spoken discussions on specific topics.

Firstly, with the focus on multimodal and digital literacy in language learning (Hafner, 2014), there is a need for information to be represented in multiple modes. While learners have traditionally relied on written sources for information, the rise in online and mobile learning accompanied by the use of various social media platforms (e.g., Facebook, TikTok) in education has created expectations of “anytime, anywhere” learning with knowledge sources provided in a variety of modes (Meniado, 2023). NotebookLM’s podcast function enables written and audio texts to be presented as spoken output that goes beyond mere text-to-voice rendering. Instead, two AI-hosts discuss and extend the content of the source text as a discussion, unpacking difficult concepts expressed in academic language by explaining them in conversational language and exemplifying them with common everyday examples. Experiencing texts multimodally can reduce the linguistic demands of academic reading and create opportunities for language learners to engage in texts beyond their language proficiencies (Archer, 2022).

Second, the customization function (See Figure 3) allows users to specify the scope, topic, and audience. This aligns with second language acquisition (SLA) principles which claim that for learners to develop language, they must receive meaning-based, communicatively embedded comprehensible input (Lichtman & VanPatten, 2021; Nation & Macalister, 2020; Yeo & Newton, 2024). It is also supported by Cai et al.’s (2023) research on how customization features in AI tools can provide scaffolding so that students can operate within the Zone of Proximal Development (ZPD). In the ZPD, the difficulty level of the learning is slightly above learners’ current level, and scaffolding from a “more capable peer” (in this case, NotebookLM’s AI hosts) enables learners to achieve understanding of the content and language.

Figure 3. Customization Function

Third, the talk-show format enables learners to listen to others discussing a topic rather than participating directly in the discussion. Unlike other AI research assistants where users have to “chat with” the tool by asking questions or giving commands, with NotebookLM, learners are onlookers or observers. Taking the backseat, they can listen to AI-hosts unpacking complex academic concepts in a simple, conversational style without the stress of being an interlocutor, which can be demanding for language learners who struggle to produce language output. The importance of managing learners’ negative emotions and creating a stress-free learning environment in language learning (Richards, 2022; White, 2018) and more specifically in EMI situations (Hillman et al., 2023) has been noted. The friendly talk-show style format is familiar to today’s learners who are accustomed to consuming information in such entertainment-like formats. Tsang and Lee’s (2024) research found that students derived great enjoyment from teachers’ use of multimedia, hence the talk-show format is likely to be appealing to learners.

Finally, the transformation of written academic text or academic lectures into a spoken genre is especially helpful for content learning in EMI and bilingual education settings. Richards and Pun (2022) explain that in school settings, students encounter academic language that can be dense and inaccessible due to the presence of complex structures such as nominalizations, passive structures, and complex sentences. By presenting the content in a conversational form with paraphrasing and repetition of key concepts and explication through the use of familiar examples, NotebookLM makes academic texts more accessible, engaging, understandable and memorable. At the same time, users become aware of the differences between academic language and conversational language so that they can learn to use these registers appropriately. Appendix 1 shows examples of how original written text is represented in conversational form within the Google NotebookLM audio summary.

Despite these benefits, NotebookLM poses a range of potential dangers, especially if students rely on the audio discussion in lieu of reading the original texts. This is because the audio discussion is not simply a summary or a paraphrase of the original input, but the content is extracted, embellished and extended by the AI-hosts. Our interactions with the feature suggest that the hosts generally put a positive spin on the content with hyperbolic praise and positive statements to grab the interest of listeners and mimic the genre of talk-shows. In addition, in an attempt to unpack the content to help listeners understand academic and technical concepts, the AI-hosts elaborate on the meanings of key ideas and provide their own examples based on familiar contexts, but these may sometimes be hallucinations.

As shown in Appendix 1, the 179-word abstract is represented as a nearly 1000-word discussion between two AI-hosts. One glaring inaccuracy is that while the source text (abstract) states “What does it mean to write, learn to write, and teach writing” (Yeo, 2023), the audio generated discussion only focuses on how AI impacts writing for students. In addition, expressions such as “seriously cool stuff,” “mind blowing,” “pretty impressive stuff,” “amazing and kind of scary,” and “fascinating content” carry a value judgement or opinion that is being generated by two AI-hosts. Most worrying is that some learners may assume that the content of the discussions is based solely on the source text and may end up misquoting it. They may attribute an idea to the source text when it was actually an extension or example provided by the AI-hosts. There is, therefore, a very real risk of sources being misquoted or incorrect information being reproduced. Any language teachers or learners should be aware of these risks when engaging with this tool, and develop verification processes, if they plan to use the content in academic contexts.

Technical Evaluation

As stated on the website, Google NotebookLM is currently free to use and supports users aged 18+ in over 200 regions (as of 13 Nov 2024). Users can import audio files in more than 70 languages. According to the website, NotebookLM assures privacy, stating that it will not utilize personal data for training. As of December 8, 2024, users could upload up to 20 sources per notebook, with each source containing up to 200,000 words (though this may have changed by the time of publication).

Although NotebookLM offers customization of the audio-generated discussion, the functionalities are limited. For example, when we explored the option to tailor audio summaries for an audience of CEFR B1-level English teachers, we found it worked quite well, with content narrowed down and the language level simplified. However, NotebookLM was unable to meet the prompt for different accents, with both AI-hosts retaining an American accent in the audio output. Given the current attention to Global Englishes and the need for learners to hear a range of accents (Galloway & Rose, 2018), it would be beneficial to add an accent variation feature. Other AI tools like Voki can generate different accents (Karakas, 2023) so it may be easy to integrate this as a future update. Additionally, NotebookLM does not yet generate transcripts of the discussions automatically. While this can be done through another app (e.g., Microsoft 365), having a direct transcript option would benefit learners as research has shown that reading-while-listening offers advantages to second language learners (Conklin et al, 2020).

Practical Applications for Teaching

The discussion function of NotebookLM can be used to support language teaching in several ways. Firstly, in EMI settings, students can be asked to match concepts from the text written in academic terminology with their meanings conversationally, thus raising their awareness about the differences between what Cummins (2015) called Cognitive Academic Language Proficiency (CALP) and Basic Interpersonal Communication Skills (BICS). For example, based on Appendix 1, we can see that “minimal original input” is expressed conversationally as “barely having to do anything” and “widespread” is changed to “already out there”.

Secondly, teachers can use the discussions as examples of conversations between two proficient users of English to help develop aspects of learners’ core speaking skills. To develop phonological awareness, students can be asked to listen to the discussion for isolated sounds as a precursor to practising and producing the sounds. For example, in the transcript of the discussion in Appendix 1, three potentially problematic sounds for specific groups of EMI learners might be: (1) /dʒ/ – as in “just,” “generative,” or “judge”; (2) /tʃ/ – as in “challenge,” “chat” or “teaching”; and (3) /eɪ/ – as in “say,” “amazing” or “change.” In addition to segmentals, students can be asked to notice supra-segmental features such as syllable stress in academic words, use of word stress to emphasize meaning and elision of sounds to increase fluency. Based on haptic approaches, in which students physically represent different aspects of pronunciation (Burri, 2023), while listening to the discussion in the transcript, students could, for example, clap when they hear a stressed syllable (e.g. It’s a fas-ci-na-ting topic” – clap on “fas-”); make a punch movement to show which word(s) in the sentence is/are stressed (e.g. “We have to adapt how we teach” – punch on “adapt” and “teach”); and swing their arms to show linking sounds (e.g. “Think of it like a super student” – swing between “of” and “it”).

Finally, it is possible to develop learners’ critical thinking skills by getting them to compare the academic text with the transcript of the discussion generated by NotebookLM to identify the source and evidence of ideas. For example, based on Appendix 1, learners could be asked to complete an exercise such as the one shown below (Table 1).

Table 1. Sample exercise

Idea	Source (Tick)	Evidence
1. GPT-3 is described as a “super student” who has read every book.	☐Article ☐Discussion ☐Both
2. The use of AI tools risks students becoming overly dependent on them.	☐Article ☐Discussion ☐Both
3. AI tools can paraphrase, summarize, and create original text with minimal input.	☐Article ☐Discussion ☐Both

The ability to identify the provenance of ideas is especially important since, as noted earlier, the AI-hosts may misrepresent ideas from the original source or add ideas from other less reliable sources, causing learners to misattribute the source of ideas.

Practical Applications for Research

NotebookLM can be a powerful assistant for researchers conducting literature reviews. Not only can it generate discussions about a single academic text, but users can upload multiple texts from a variety of sources such as papers or entire books, then “have conversations” with these documents by adding prompts to the chat box. The corresponding original texts are displayed alongside, which helps prevent AI hallucinations by grounding responses in the source material. During the process, researchers can save original excerpts or summaries to their notes, facilitating direct or indirect quotations when writing their literature reviews. For comparative analysis, multiple documents can be uploaded, allowing researchers to compare several sources simultaneously. This helps automate part of the process and speed up the review period (Chubb et al., 2021). For instance, users can prompt, “What are the focus of these articles?” and NotebookLM can read and summarize each article, providing corresponding quotes from the original text. After initial prompts, related questions are suggested for deeper exploration. Researchers can investigate further by selecting specific paragraphs of the original texts and choosing options like “suggest related ideas,” “summarize to note,” or “help me understand the selected note.” Later, users can prompt NotebookLM to generate a literature review, presenting the connections between the articles in a structured manner so that researchers can quickly grasp their interrelations. What makes NotebookLM stand out is that it can then produce an audio overview in which the AI-hosts synthesize ideas from all the documents. By listening to the discussion of the AI-hosts, researchers can deepen and widen their understanding of the uploaded sources. For auditory learners, listening to the synthesis may help them to process and retain the information more effectively (Montero Perez, 2022). Another unique use of NotebookLM might be to get students to submit their own written articles, generate a discussion, and then listen to it to evaluate how well the AI-hosts have represented the main ideas from their writing. This could be part of their review and revision process with the AI-hosts serving the dual role of audience and peer reviewer. Many EMI learners may feel shy and therefore unwilling to share drafts of their writing with a classmate for fear of judgement, so having an audience, albeit an AI-audience, may be motivating and empowering (Wan & Moorhouse, 2024; Kim & Su, 2024).

Conclusion

Reviews on Google’s NotebookLM webpage suggest that users see its release as a “ChatGPT Moment,” indicating its potential beyond typical AI research assistants. Its ability to create customized AI-hosted talk-show style discussions make it a potentially effective tool for engaging with and understanding dense academic texts. This feature could be especially beneficial for EMI learners who struggle with complex academic terminology and structures. Content presented in the talk-show format is not only easy to understand but also less intimidating, meeting SLA principles by providing meaning-focused comprehensible input and offering scaffolding to learners in developing content and language proficiency. However, because the AI-hosts do not simply summarize or paraphrase but also embellish and extend the content, there are dangers of misrepresentation of the source text. Users should engage with it critically and, as just as they would with any AI tool, develop processes to verify the accuracy of the content (Moorhouse, 2024). The tool should not replace the need to engage with original texts, but it can help facilitate language learners’ engagement with and comprehension of the texts.

Like other AI tools, Google’s NotebookLM is neither good nor bad. Rather, it is the way that we use it that determines its morality. As with ChatGPT, only with wider adoption and prolonged usage will we truly be able to reflect on its benefits, drawbacks, and impact on knowledge creation and curation in an age of human-AI co-intelligence.

About the Authors

Marie Alina YEO is a Senior Language Specialist at SEAMEO RELC. She conducts teacher development courses in language assessment, Content and Language Integrated Learning (CLIL) and teaching methodology. Marie is Co-Editor of the RELC Journal and has published articles on academic integrity and GenAI, CLIL and formative assessment. ORCID ID: https://orcid.org/0000-0003-0136-2532

Benjamin Luke MOORHOUSE is an Associate Professor in the Department of English, City University of Hong Kong, China. He studies the lived experiences of language teachers and learners. Recently he has been exploring the impact of GenAI on language teachers’ professional competencies and practices. ORCID ID: https://orcid.org/0000-0002-3913-5194

Yuwei WAN is a PhD candidate in the Department of Education Studies at Hong Kong Baptist University, China. Her research interests include teachers’ AI literacy, technology in language teaching and learning, and professional development for language teachers. ORCID ID: https://orcid.org/0000-0002-6377-7904

To Cite this Article

Yeo, M. A., Moorhouse, B. L., & Wan, Y. (2025). From academic text to talk-show: Deepening engagement and understanding with Google NotebookLM. Teaching English as a Second Language Electronic Journal (TESL-EJ), 28(4). https://doi.org/10.55593/ej.28112int1

References

Archer, A. (2022). A multimodal approach to English for academic purposes in contexts of diversity. World Englishes, 41(4), 545–553. https://doi.org/10.1111/weng.12600

Burri, M. (2023). Making the Case for Brain-friendly Pronunciation Instruction. RELC Journal, 54(3), 864-872. https://doi.org/10.1177/00336882211035519

Chubb, J., Cowling, P. & Reed, D. (2022). Speeding up to keep up: exploring the use of AI in the research process. AI & Soc 37, 1439–1457. https://doi.org/10.1007/s00146-021-01259-0

Conklin, K., Alotaibi, S., Pellicer-Sánchez, A., & Vilkaitė-Lozdienė, L. (2020). What eye-tracking tells us about reading-only and reading-while-listening in a first and second language. Second Language Research, 36(3), 257-276. https://doi.org/10.1177/0267658320921496

Cummins, J. (2016). BICS and CALP: Empirical and Theoretical Status of the Distinction. In B.V. Street & S. May (Eds.), Literacies and Language Education (pp. 1–13). Springer International Publishing. https://doi.org/10.1007/978-3-319-02321-2_6-1

Forte, T. (2022). Building a second brain: A proven method to organize your digital life and unlock your creative potential. Simon and Schuster.

Galloway, N., & Rose, H. (2018). Incorporating Global Englishes into the ELT classroom. ELT Journal, 72(1), 3-14. https://doi.org/10.1093/elt/ccx010

Hafner, C. A. (2014). Embedding digital literacies in English language teaching: Students’ digital video projects as multimodal ensembles. TESOL Quarterly, 48(4), 655-685. https://doi.org/10.1002/tesq.138

Hillman, S., Li, W., Green-Eneix, C., & De Costa, P. I. (2023). The emotional landscape of English medium instruction (EMI) in higher education. Linguistics and Education, 75, 101178. https://doi.org/10.1016/j.linged.2023.101178

Karakaş, A. (2023). Supporting the Teaching of Global Englishes Language Teaching (GELT) Through Voki. RELC Journal, 0(0). https://doi.org/10.1177/00336882231199515

Kim, A., & Su, Y. (2024). How implementing an AI chatbot impacts Korean as a foreign language learners’ willingness to communicate in Korean. System, 122, 103256. https://doi.org/10.1016/j.system.2024.103256

Lichtman, K., & VanPatten, B. (2021). Was Krashen right? Forty years later. Foreign Language Annals, 54(2), 283-305. https://doi.org/10.1111/flan.12552

Meniado, J. C. (2023). Digital Language Teaching 5.0: Technologies, Trends and Competencies. RELC Journal, 54(2), 461-473. https://doi.org/10.1177/00336882231160610

Montero Perez, M. (2022). Second or foreign language learning through watching audio-visual input and the role of on-screen text. Language Teaching, 55(2), 163-192. https://doi.org/10.1017/S0261444821000501

Moorhouse, B.L. (2024). Generative artificial intelligence and ELT, ELT Journal, 7 (4), 378–392, https://doi.org/10.1093/elt/ccae032

Nation, I. S. P., & Macalister, J. (2020). Using the four strands to plan a reading or writing course. In Teaching ESL/EFL Reading and Writing (pp. 15-29). Routledge.

Richards, J. C., & Pun, J. (2022). Teaching and learning in English medium instruction: An introduction. Routledge.

Tsang, A., & Lee, K. F. K. (2024). What is Most and Least Enjoyable in the EFL Classroom? A Large-scale Inductive Qualitative Study. RELC Journal, 0(0). https://doi.org/10.1177/00336882241292090

Wan, Y., & Moorhouse, B. L. (2024). Using Call Annie as a Generative Artificial Intelligence Speaking Partner for Language Learners. RELC Journal, 0(0). https://doi.org/10.1177/00336882231224813

White, C. J. (2018). The emotional turn in applied linguistics and TESOL: Significance, challenges and prospects. Emotions in second language teaching: Theory, research and teacher education (pp.19-34). Springer, Cham.

Yeo, M. (2023). Academic integrity in the age of Artificial Intelligence (AI) authoring apps. TESOL Journal, 14(3), e716. https://doi.org/10.1002/tesj.716

Yeo, M. & Newton, J. M. (2024). The four strands: A simplified approach to SLA for CLIL. TESOL Journal, e887. https://doi.org/10.1002/tesj.887

Appendix 1: Comparison of Original Text and Conversational Form in Google NotebookLM

Original source text (179 words)

Abstract What does it mean to write, learn to write, and teach writing in an age when students can use the latest artificial intelligence (AI) co-authoring tools to produce entire essays without even adding an original idea or composing a single sentence? This article addresses questions of authorship and academic integrity concerning the use of AI writing assistants and the latest GPT-3 (Generative Pre-trained Transformer, Version 3) tools. It begins by problematizing the use of these tools, and then illustrates how students can use these tools to paraphrase, summarize, extend, and even create original texts with minimal original input, raising questions about authorship and academic integrity. The author argues that as these tools become more widespread, teachers must find creative ways to integrate them into the teaching and learning process and offer practical suggestions for classroom practice. The author hopes to raise awareness about threats to academic integrity brought about by the use of the latest AI co-authoring tools and aims to equip teachers with strategies to embrace the use of these new digital technologies in the teaching of writing. (Yeo, 2023)

Transcript of talk-show (1185 words)

Speaker 1: All right, so we’re diving deep today into some seriously cool stuff. Articles you sent over about AI… Speaker 2: Yeah. Speaker 1: …really got me thinking, you know, AI and how it impacts writing. Speaker 2: Uh-huh. Speaker 1: But especially for students and how they write essays. Speaker 2: Right. Speaker 1: I mean. It’s kind of mind blowing to think about tools that can just like paraphrase for you or summarize stuff for you, or even extend your thoughts and like it, create totally original text just from a few little prompts you give it. Speaker 2: Pretty wild. Speaker 1: Yeah, it’s wild. It makes you wonder, though, like, what does this even mean for authorship? You know? What about academic integrity? Like where does that all fit in? Speaker 2: Well, the crazy thing is like these AI tools, a lot of them are powered by this thing called GPT 3. Speaker 1: OK. Speaker 2: They’re already out there and they can write some pretty impressive stuff with the student, like barely having to do anything. Speaker 1: Really? Speaker 2: Yeah, imagine like feeding the AI every essay you wrote in high school. It learns your style. You know how you write all your little quirks, and then it just spits out this brand-new essay. Even if it doesn’t really understand the topic. Speaker 1: So, it’s basically like having a ghostwriter who also happened to be like a super computer or something. That’s amazing and kind of scary all at the same time. Speaker 2: Yeah. Speaker 1: I mean, if this student isn’t the one coming up with the ideas or actually writing the sentences, can we really say it’s their work? Speaker 2: You’ve hit on a huge point there like this whole question of authorship is getting really blurry. But to really get into that, we got to breakdown what GPT 3 actually is. So GPT 3 stands for generative pre trained transformer 3. It’s a type of language model. Speaker 1: OK. Speaker 2: Think of it like a super student who’s read every single book in the library… Speaker 1: Oh wow. Speaker 2: …and can copy any writing style you can imagine. It’s trained on massive amounts of data, but it doesn’t actually think for itself. It just uses patterns to guess what words should come next, and that’s how it makes the text. Speaker 1: OK. So, it’s not like it’s sentient or anything, but doesn’t that still bring us back to that problem of academic integrity? If a student uses one of these tools to write an essay, isn’t that basically plagiarism? Even if the AI is creating unique text? Speaker 2: That’s where things get really complicated. See, a lot of plagiarism detection tools just look for exact matches in the text, but AI often spits out stuff that’s unique which makes it harder to catch. Speaker 1: Oh, I see. Speaker 2: And that raises the question like should we be more worried about the text being original or the ideas being original? Speaker 1: That’s a good point, because the AI might be writing new sentences, but it’s still pulling from this huge database of information that’s already out there, so the student isn’t really showing what they know or like their critical thinking skills. Speaker 2: Exactly. And that’s why this whole conversation is so important. It’s forcing us to rethink what we actually care about in education. Are we just obsessed with the final product, like getting that perfect essay? Or are we more interested in how students learn? You know them wrestling with ideas and making connections and finding their own voice? Speaker 1: It kind of feels like we’re at a crossroads here. We can either try to fight this new technology… Speaker 2: Yeah. Speaker 1: …or figure out how to make it part of our classrooms in a way that helps students learn and doesn’t mess with academic integrity. Speaker 2: I completely agree. Trying to ban these tools would be like trying to ban the Internet. It’s just not going to happen. Speaker 1: Right. Speaker 2: They’re here to stay. The key is figuring out how to adapt. Speaker 1: So, for teachers, that might mean changing how they teach and how they grade stuff instead of just focusing on the essays themselves. Maybe it’s about helping students really understand the subject. Speaker 2: Yeah. Exactly. Speaker 1: So instead of saying, write a 5- paragraph essay on the Civil War, maybe we asked them to use AI to like, analyze some primary sources and then present what they found in a more creative way, like a podcast or a video essay. Speaker 2: Exactly. We can encourage them to use these tools for brainstorming or outlining, or even to get over writer’s block. But then we gotta push them to go further, to build on that with their own insights and critical thinking. Speaker 1: This reminds me of that whole debate about calculators. Speaker2: Oh. Yeah! Speaker 1: People were freaking out that kids would forget basic math. But it turned out calculators actually helped them do more complex stuff. Speaker 2: That’s a great comparison. It shows how AI writing tools can be helpful. They can do those boring tasks for us, like fixing grammar and stuff so students can focus on the bigger picture. Speaker 1: But we have to watch out for the downsides too. Speaker 2: Absolutely. Speaker 1: Like what if students start relying on the AI too much and don’t learn how to write well on their own. Speaker 2: Exactly. That’s why it’s so important to have clear rules and to talk to students openly about these schools. Help them understand like the ethical side of things and encourage them to use the tools responsibly. It’s about making them smart users of technology, not just passive consumers. Speaker 1: So, I guess the big take away here is that we need to adapt. We have to adapt how we teach, how we assess students and even how we think about authorship. All because of these new AI tools. Speaker 2: Absolutely. And the most interesting question of all might be how will these AI writing tools change how we see creativity and authorship in the future? Speaker 1: That’s a big one. Speaker 2: It’s a question we’re all going to have to figure out. Speaker 1: Well, this deep dive has been seriously thought provoking. You’ve given us so much to think about when it comes to writing and education in this whole age of AI. Speaker 2: It’s a fascinating topic. I’m glad we could talk about it Speaker 1: To all our listeners out there, keep exploring this stuff. Experiment with these tools and think critically about what it all means. The future of writing is being shaped right now… Speaker 2: Yeah. Speaker 1: …and your voice is part of that conversation. [back to article]

Copyright of articles rests with the authors. Please cite TESL-EJ appropriately. Editor’s Note: The HTML version contains no page numbers. Please use the PDF version of this article for citations.