ELT and AI for Good

Posted: July 24, 2025 in GenAI, Review
Tags: , , , , , , , , , , , , , ,
4

Cambridge publishes a series of short academic titles called ‘Elements’. There are dozens of them and, for a few weeks only after publication, they can be freely downloaded in pdf format. They vary in quality. Some are very useful. For others, the quality is commensurate with the price you pay.

A recent addition to the series is ‘Generative Artificial Intelligence and Language Teaching’ (Moorhouse & Wong, 2025). In 8 chapters and under 80 pages, the authors rattle through a general introduction to Generative AI (1), the use of Generative AI as a ‘knowledge resource and development tool’, as an assistant for lesson planning and materials production, and for use with assessment and feedback (2-4), how students use AI for language learning (5), ethical and social considerations (6), necessary AI skills (7) and professional development through AI (8). There’s little, if anything, that is new here for people who already know about and use Generative AI in language teaching. For those who don’t, there’s not enough in the way of detailed practical suggestions to make this title useful.

The book claims to promote ‘evidence-informed approaches’. It doesn’t. It is an enthusiastically crude flag-waving exercise for GenAI. Such evidence as is cited is extremely selective, and evidence that might deflate the authors’ hype is ignored. The tone of the book is, perhaps, best summarised by a little poem (in part used to demonstrate how good AI-generated poems are!) that the authors generated with AI to summarise chapter 2:

So teachers rise, embrace the aid,

Of GenAI tools carefully made

Reservations about, say, the problematic reliability of ChatGPT or the bias and stereotyping that is unavoidable in ChatGPT text are always relegated to the end of sections. Very briefly. Some of the suggestions are frankly bizarre. The authors suggest that teachers use ChatGPT to find out what pedagogical approach might be good for them: ‘Language teachers can prompt a conversational AI chatbot to generate a list of methods and approaches, provide key information about a specific approach, or create example activities and lesson plans that illustrate the implementation of the approach in a specific context.’ The example that is provided is of using the Silent Way – not perhaps the most evidence-informed approach. Then, there’s the suggestion to use GenAI to produce an exercise which involves transforming passive verbs to active ones (p.31). It’s hard to imagine what kind of evidence might have informed this. Stilted, unnatural ‘dialogues’ are praised for their authenticity and inclusion of conversational features.

The chapter on ethical and social considerations is especially shallow. The problem of standardised English (produced by GenAI) is, apparently, easily solved: ‘Educators can take proactive steps to balance the benefits of many conversational AI chatbots’ standardised outputs with the promotion of linguistic diversity.’ To address issues of copyright and intellectual property, teachers should explicitly teach students ‘about the importance of copyright laws and the limitations of AI generated content’. Bias is acknowledged as a problem but no solutions are offered.

When it comes to environmental costs, the suggestion is to prioritise ‘specific AI tools that might require less computational power, optimise classroom workflows to reduce unnecessary AI usage, and advocate for greener infrastructure from AI developers’. Yup, it’s as easy as that!

As for the ways that students use GenAI, AI literacy needs to be fostered because without it ‘students may either over-rely on GenAI or fail to leverage its potential in ways that truly benefit their learning’. The first example that is given of how to do this is to give students the following prompt:

I am a high school student learning English and I struggle with verb tenses and phrasal verbs. Please create a quiz with 10 sentences where I have to choose the correct verb tense (past, present, or future), and explain why the answer is correct.

Again, not a lot of evidence I’m aware of would support this approach.

Underlying the whole book is the unexamined assumption that language teaching can be made more efficient by using GenAI: ‘AI tools can enhance teaching by automating routine tasks such as grading or generating lesson plans, allowing teachers to focus more on individualised instruction and critical thinking activities’. There’s no evidence to support this claim, and no indication that there might be evidence that suggests the opposite (see, for example, Selwyn et al., 2025). The claim is probably taken from an OECD report (2023), full of the OECD’s usual techno-solutionism, and lacking any research.

What kind of person would write such a one-sided and evidence-ignored book? Well, one of the authors, B. L. Moorhouse, gave a talk earlier this year, entitled ‘AI for Good: Nurturing critical and positive uses of GenAI in language learning’. What does this choice of title, ‘AI for Good’, tell us?

‘AI for Good’ is a United Nations initiative, established in 2017, with a brief to identify innovative applications of AI to solve global challenges. It organises annual conferences which bring together people who are interested in its mission, and celebrates worthy projects in its ‘AI for Good Impact Awards’. These projects are often very inspiring. There’s a chatbot co-created with refugee communities to strengthen healthcare access and autonomy for refugee women. Or another that uses satellite imagery and machine learning to analyse soil health and land degradation in Ghana.

However, it is worth asking who the main beneficiary of ‘AI for Good’ might be. In the marketing of AI by Microsoft, Apple, Amazon, Alphabet, NVIDIA, Meta and all their business partners, the technology is ‘couched as a global opportunity, a promise at the feet of humankind itself, to extend into a new world space of ease, efficiency’ (Adams, R., 2025: 18). Meanwhile, criticism of AI is growing fast with talk of a new AI backlash (Rogers, 2025). Rather than being a force for good, more voices are articulating the idea that AI ‘deepens poverty, fractures community and social cohesion, and exacerbates divides between people and between groups’ (ibid.:11). Worse still, business and investment voices have started to sound warnings: Goldman Sachs has massively revised down its estimates of the extent to which AI will boost productivity (Bender & Hanna, 2025: 193). Talk of AI for Good serves to distract from thoughts of a world actually made worse by AI. Such distraction is crucial if the big AI companies are to continue to attract investment.

It’s no surprise, then, to find that the major sponsors of the ‘AI for Good’ conference are the big vendors and their associates (Microsoft, pwc, Deloitte, IBM, Shell, Amazon, Samsung, Alibaba, Huawei, Dell, Lenovo, etc.). It’s no surprise either to learn that the event is packed with demos of robots and drones, or that many of the speakers are tech executives and CEOs (Sam Altman of OpenAI was the star speaker last year). This year, one of the invited speakers, Abeba Birhane, was asked, shortly before her talk, to remove some of her slides. The offending slides included anything that mentioned Palestine, the word ‘genocide’ had to be removed, along with another slide referring to illegal data torrenting by Meta (Goudarzi, 2025). Birhane commented:

It feels like when they are claiming AI for social good, it’s only good for AI companies, good for the industry, good for authoritarian governments, and their own appearance. They pride themselves on having tens of thousands in attendance every year, on sponsorships, and on the number of apps that are built, and for me that really is not a good measure of impact. A good measure of impact is actual improvement of lives on the ground. Everywhere you look here feels like any other tech summit rather than a social good summit. So, it’s really disappointing to witness something that is supposed to stand for social good has been completely overtaken by corporate agenda and advancing and accelerating big corporations, especially their interests, rather than doing any meaningful work.

Outside this year’s summit, a group of about 100 activists protested to accuse the major tech companies of complicity in war crimes against Palestinians. The complicity of IBM, Microsoft, Google and Amazon is catalogued in the report by Francesca Albanese for the UN Human Rights Council, ‘From Economy of Occupation to Economy of Genocide’.

Time to return to the Cambridge ‘Element’. I think it is much more deeply informed by the ‘AI for Good’ tropes than it is by any evidence. It is probably driven primarily by naivety and wishful thinking, but its net effect is to contribute to the hype of GenAI and to promote the corporate agenda.

References

Adams, R. (2025) The New Empire of AI. Cambridge: Polity Press

Bender, E. M. & Hanna, A. (2025) The AI Con. London: Bodley Head

Goudarzi, S. (2025) AI for good, with caveats: How a keynote speaker was censored during an international artificial intelligence summit. Bulletin of the Atomic Scientists, July 10, 2025 https://thebulletin.org/2025/07/ai-for-good-with-caveats-how-a-keynote-speaker-was-censored-during-an-international-artificial-intelligence-summit/

Moorhouse, B.L. & Wong, K.M. (2025) Generative Artificial Intelligence and Language Teaching. Cambridge: Cambridge University Press

OECD. 2023. Generative AI in the Classroom: From Hype to Reality. Paris: Organisation for Economic Co-operationand Development. https://one.oecd.org/document/EDU/EDPC(2023)11/en/pdf.

Rogers, R. (2025) The AI Backlash Keeps Growing Stronger. Wired, 28 June 2025 https://www.wired.com/story/generative-ai-backlash/

Selwyn, N., Ljungqvist, M. & Sonesson, A. (2025) When the prompting stops: exploring teachers’ work around the educational frailties of generative AI tools. Learning, Media, and Technology, 23 July 2025 https://www.tandfonline.com/doi/full/10.1080/17439884.2025.2537959#abstract

What’s (not) changed in English Language Teaching?

Posted: July 17, 2025 in innovation
Tags: , , , , , , , , ,
2

The British Council likes innovation. For over 20 years they have been handing out ‘ELTon’ awards to celebrate ‘innovation in English language teaching and learning’. Last year, to mark their 90th anniversary, they organised the ‘ELTons Festival of Innovation’. As part of this, they produced a video, available on YouTube, called ‘Never Stop Innovating’, inspired, perhaps, by their earlier video ‘Innovation is Great’. I’ve written elsewhere about the British Council’s obsession with innovation, so I won’t add to that here.

Last month, the British Council produced a ‘research paper’ called ‘What’s changed in English Language Teaching? A review of change in the teaching and learning of English and in teacher education and development from 2014-2024’ (Motteram & Dawson, 2025). My curiosity was piqued, not least because I wrote a book called ’30 Trends in ELT’ a few years ago.

The authors of the British Council report conducted a review of blogs, journals, publishers’ websites and books (including mine) and compiled a general overview of trends as reflected in these sources. They then carried out an international online survey asking participants (about 1000 people responded) if they thought that the identified trends were very important / important / not important / not sure.

The report begins with predictable introductory guff about how change is exciting, powerful and inevitable, how technology is revolutionising education, etc. The actual findings of the report will come as no surprise to many. The most important trends in the last ten years include:

  • New contexts for English language learning (e.g. EMI, CLIL)
  • Plurilingual practices and a shift away from native-speaker norms
  • Digital and critical literacies
  • Technology related pedagogies (e.g. blended, hybrid, mobile)
  • Social-emotional learning (e.g. resilience, learner agency, well-being)
  • Inclusive practices
  • Assessment for learning
  • 21st century skills

Identifying trends in the literature around ELT is one thing. Making claims about what has changed in ELT (which must be primarily about classroom practices) is something else altogether. The fact that certain topics appear frequently in the pages of ELTJ, TESOL Quarterly and TESOL Journal tells us nothing about what teachers are actually doing. The readers and writers of these journals are overwhelmingly academics, not teachers. It may even be the case that one of the reasons that teachers do not read these journals is that they do not see the subject matter as being relevant to them.

Asking a bunch of teachers whether they think a given trend is important also tells us very little. It is unlikely that the teachers who responded are representative of English language teachers in general. Some countries are disproportionately under or over represented. Primary teachers are under-represented, post-secondary teachers are over-represented. Huge numbers of teachers, like those who teach on the big platforms are probably not represented at all. Does it even make any sense to aggregate the figures in order to paint some sort of picture of ‘ELT’?

There’s also a problem with the way that the questions are framed. There are probably few teachers, for example, who do not think that critical literacies are important. There are probably equally few who do much about it in their English language classrooms. Since we know (e.g. from research into English language teachers’ use of L1) that there can be large discrepancies between what teachers think they are doing and what they are actually doing, we need to be wary of taking their word for it. Genuinely interesting would be insights into classroom practice, not statistics about academic literature and perceptions of topics. Such insights would necessarily be very context-specific. Ironically, the British Council report underlines the ‘central importance’ of teaching and learning contexts … and then ignores them.

Here’s one context that I am very familiar with. It’s a class of mostly 12-year-olds in the second year of an Austrian ‘gymnasium’. In terms of linguistic background, the class is very mixed: only a very small number have German as a home language. Four or five really struggle with German, the school’s language of instruction, but have reasonably good English language skills. The class would seem to be a very strong candidate for a CLIL approach, but there is not a hint of plurilingual practice in any lessons. In fact, the only plurilingual practice is found in English lessons where the children learn bilingual vocabulary sets. All the children received iPads at the start of the first year. They are mostly used to access digital versions of their maths textbooks and, in English lessons, to complete gap-fills and comprehension questions as their regular homework from the digital workbook of their coursebook (‘More’ by Puchta et al.). The syllabus, reflected in the tests, is grammar-oriented and accuracy-focused. Most of the children are now using ChatGPT to do English homework when this involves writing. After two years of weekly digital literacies lessons (which are not integrated into other parts of the curriculum), there has been no attempt to deal with critical digital literacy. In another weekly lesson (also not integrated with the rest of the curriculum), aspects of twenty-first century skills and social-emotional learning are ‘taught’. Racism and homophobia are sadly common undercurrents in the discourse of the children.

In the approach to English learning / teaching in this school, I cannot identify a single element that has changed in the last ten years. Or twenty, for that matter. And very little in the last thirty. It is impossible to say to what extent this context is representative of the wider world of ‘ELT, but my suspicion is that very little has changed in most primary, secondary and tertiary English language classrooms around the world in recent decades. Coursebooks remain ubiquitous.

The British Council, along with ELT journals, publishers’ blogs and so on, competes in the marketplace for attention. They have an image to nurture and a product to sell. But their customers are not primarily classroom teachers, or, at least, not directly. I suspect that for most classroom teachers, the central concerns remain what they always were: how to get and hold the attention of large classes of less than highly motivated students, and how to reduce their own (especially administrative) workload. The trouble with most ‘innovations’ is that they do not offer any solutions to the first concern, and are more likely to add to their workload than to decrease it.

References

Motteram, G., & Dawson, S. (2025). What’s changed in English language teaching? A review of change in the teaching and learning of English and in teacher education and development from 2014–2024. British Council. https://doi.org/10.57884/FPE8-5130

Taking AI literacy seriously

Posted: April 3, 2025 in critical thinking, ed tech
Tags: , , , , , , , , , , ,
2

English language teacher education, says the British Council (Edmett et al. 2024: 21), ‘must include a focus on AI literacy’ and ‘teachers also need to develop their learners AI literacy’. This report is referring to Generative AI (e.g. ChatGPT), the focus of my discussion here.

There are basically two kinds of AI literacy relevant to language learners. The first concerns a knowledge of the AI-powered tools that are available to them as learners. The internet is awash with things like ‘Top 10 AI tools for planning EFL lessons’ or ‘Generative AI for teachers: free idea pack!’, there are courses available, and language teaching conferences are full of the same. Many people like this sort of thing. Myself, I’m a bit more meh, for two reasons.

First of all, because the tools date so quickly in terms of functionality, accessibility (e.g. cost) and range. Over a year ago, there were already 3 million AI bots available on OpenAI’s GPT store. It is impossible to count how many of these are useful or designed for language learning and teaching. My second reason is that, although many of these tools have the potential to help language learners, that potential is only likely to be realised if the learner is autonomous and self-regulating. That is the main challenge of language learning, and it is one that we are far from providing answers to. The choice of this app or that app is unlikely to make much difference. It has been argued that Generative AI can promote learner autonomy, but there is no empirical evidence that this is the case (Szabó, & Szoke, 2024), and there are legitimate fears that the technology may have the opposite effect. The history of edtech suggests that it is unlikely that Generative AI will be any different from any other technology in promoting autonomy (Watters, 2021).

I find the second kind of AI literacy more interesting. This concerns critical thinking and ethical awareness, and is reflected in the subtitle of a newly published book ‘AI Literacy in the Language Classroom: Facilitating critical, ethical and responsible use’ (Menyhei & Szoke, 2025). Definitions of critical AI literacy vary, but here is the Open University’s (2025) version:

[It] highlights the influence of power in digital spaces. It explores how knowledge, identities and relationships are shaped to privilege some while marginalising others. […] Critical AI Literacy applies the lens of equality, diversity, inclusion and accessibility (EDIA) to the use of AI.

There can be little debate about the importance of critical and ethical AI literacy. However, is it possible that, like other desiderata (learner autonomy, growth mindset, grit, for example), it cannot easily be taught?

Critical AI literacy is a particular form of critical digital literacy, which, in turn, is a particular form of critical thinking. Critical digital literacy as an object of instruction has been around for almost forty years (Pangrazio, 2016), critical thinking for far longer. But the impact of instruction on critical digital literacy and critical thinking has been limited, to say the least. In fact, the very frequency of current calls for AI literacy is, in part, a result of our collective failure to teach critical digital literacy in the past. Audrey Watters, the Cassandra of edtech, wrote in a recent post:

We’ve totally failed with almost all ‘digital literacy’ efforts – to such an extent that it seems naïve to take seriously all the calls for some sort of new ‘AI literacy’.

So what is the problem? Well, the first big issue is that critical thinking is domain specific, and the same holds true for critical digital literacy and critical AI literacy. That is to say that background knowledge of a particular domain is needed (Willingham, 2007) before we can think critically within that domain. The ability to think critically about sustainable development goals, for example, does not readily transfer to an ability to think critically about systemic white privilege … or anything else, for that matter. The use of Generative AI by language learners probably ranges across so many domains that we have a huge problem if we want to teach critical AI literacy. Where do we even start?

Big issue #2: critical thinking can only be taught if learners have a disposition to think critically (Facione, 2000). Shaping dispositions is notoriously hard! If a learner is emotionally predisposed to be positive about Generative AI (because, for example, they have discovered that it can help them achieve better grades), it will be an uphill struggle to persuade them to adopt a more critical approach.

Big issue #3: a failure to think critically about Generative AI (or SDGs or systemic white privilege or whatever) is not primarily a problem of critical digital literacy. It is much, much broader than that. Bulger and Davidson (2018) argue that:

The extent to which media literacy can combat the problematic news environment is an open question. Is denying the existence of climate change a media literacy problem? Is believing that a presidential candidate was running a sex-trafficking ring out of a pizza shop a media literacy problem? Can media literacy combat the intentionally opaque systems of serving news on social media platforms? Or intentional campaigns of disinformation?

Big issue #4: evidence for any effectiveness in critical digital literacy instruction is in very short supply. We simply do not know what kind of instruction works (Guess et al., 2020, Huguet et al., 2019), and effectiveness (if there is any at all) is probably very context-dependent. If and when it is effective, the instruction will probably be long-term and cross-curricular. We should not expect occasional doses of critical digital literacy (or critical AI literacy) training in the language classroom to have any meaningful impact.

Big issue #5: is it even possible to imagine a critically informed and ethical use of Generative AI? For a technology that (1) is based on massive copyright theft, and (2) is so detrimental to the environment, it is not difficult to make a case that the only critical and ethical use is zero use.

If, however, we are prepared to put aside big issue #5, there is, perhaps, a way of avoiding the problem of domain-specificity. There is one domain which of relevance to all language learners, where dispositions may not be such a problem, where critical digital literacy may be achievable, and where we can reasonably expect some positive impact. That domain is the domain of language learning.

In Pegrum et al (2022), there are two activities that address AI literacy. The activities involve getting learners to experiment with AI-powered tools and then reflecting on how useful they found these tools to their own language learning. True, this is a very limited form of AI literacy that is being promoted, but it stands a reasonable chance of promoting metacognition and enhancing language learning.

It may well be the case that some of the growing number of activities intended to help learners recognise the algorithmic bias and the limitations of Generative AI, or to raise awareness of the environmental impact of the technology have a value in terms of the affordances for language learning that they offer. Sadly, though, I think they are unlikely to have much impact on critical AI literacy.

References

Bulger, M. & Davidson, P. (2018). The Promises, Challenges and Futures of Media Literacy. Data and Society. https://datasociety.net/pubs/oh/DataAndSociety_Media_Literacy_2018.pdf

Edmett, A., Ichaporia, N., Crompton, H. & Crichton, R. (2024) Artificial intelligence and English language teaching: Preparing for the future 2nd edition. British Council

Facione, P. A. (2000). The disposition toward critical thinking: Its character, measurement, and relation to critical thinking skill. Informal Logic, 20(1), 61–84.

Guess, A. M., Lerner, M., Lyons, B., Montgomery, J. M., Nyhan, N., Reifler, J. & Sircar, N. (2020). A digital media literacy intervention increases discernment between mainstream and false news in the United States and India. Proceedings of the National Academy of Sciences Jul 2020, 117 (27) 15536-15545; DOI: 10.1073/pnas.1920498117

Huguet, A., Kavanagh, J., Baker, G. & Blumenthal, M. S. (2019). Exploring Media Literacy Education as a Tool for Mitigating Truth Decay. RAND Corporation, https://www.rand.org/content/dam/rand/pubs/research_reports/RR3000/RR3050/RAND_RR3050.pdf

Menyhei, Z. & Szoke, J. (2025) AI Literacy in the Language Classroom: Facilitating critical, ethical and responsible use. DELTA Publishing

Open University (2025) A framework for the Learning and Teaching of Critical AI Literacy skills. The Open University

Pangrazio, L. (2016) Reconceptualising critical digital literacy. Discourse: Studies in the Cultural Politics of Education, 37 (2): 163-174, DOI: 10.1080/01596306.2014.942836

Pegrum, M., Hockly, N. & Dudeney, G. (2022) Digital Literacies 2nd edition. Abingdon: Routledge

Szabó, F. & Szoke, J. (2024) How does generative AI promote autonomy and inclusivity in language teaching? ELT Journal, 78 (4): 478 – 488 https://academic.oup.com/eltj/article/78/4/478/7784519#

Watters, A. (2021) Teaching Machines. MIT Press

Willingham, D. T. (2007). Critical Thinking: Why Is It So Hard to Teach? American Educator Summer 2007: pp. 8 – 19

Stochastic parrots and the metaphors of AI in ELT

Posted: March 14, 2025 in Discourse, ed tech
Tags: , , , , , , ,
1

Much of the language we use is metaphorical. These metaphors both shape and reflect the way we think. Often, as in the preceding sentence, they give agency (a human attribute) to non-human entities, they give architectural form to abstractions and make them appear visible. We struggle to function without metaphors: they help us to describe what otherwise would be indescribable. They give us an illusion of understanding something. Learning and teaching, for instance (Thornbury, 1991). Or artificial intelligence.

The word ‘intelligence’ in AI is itself metaphorical. Intelligence is typically an attribute of humans or other living beings. By using the word, we anthropomorphize a technological system and this, says Sam Altman (founder of OpenAI’s ChatGPT), is something that we should try to avoid. But this is very hard. Anthropomorphic metaphors are ubiquitous because their use is probably a fundamental part of human psychology. In the case of AI, it’s even harder to avoid because (1) the most important related terms (machine learning, neural networks, etc.) also use anthropomorphic metaphors, (2) so many AI interfaces (e.g. Alexa) are given pseudo-human form and (3) one of the goals of General AI is to become indistinguishable from human cognitive function. Do a Google image search for ‘AI’ and you’ll get little besides humanoid robots or illuminated human brains. The term ‘artificial intelligence’, along with its associated metaphors and images, is often used very loosely, and often as little more than a marketing construct (Rudolph et al., 2025).

Advertisement for a training course offered by Dr Nellie Deutsch

But if we don’t anthropomorphize AI (ruling out ideas like ‘companion’ or ‘coach’), what better metaphors should we use? A tool (Altman’s preference)? A cultural technology (comparable to a library)? A stochastic parrot or an autocomplete on steroids? (Mitchell, 2024) Our choice of metaphorical frame matters because it is likely to affect how we react to and use the technology (especially in our choice of prompts), and also how the technology interacts with us (Eliot, 2025)!

Unsurprisingly, researchers have turned their attention to technological metaphors in educational discourse. There’s a whole book (Weller, 2022) called ‘Metaphors of Ed Tech’, freely downloadable from the Athabasca University Press website. It’s a quirky and entertaining read, but I personally found an older academic article more interesting. This is ‘A Critical Metaphor Analysis of Educational Technology Research in the Social Studies’ (Mason, 2018), inspired (like Scott Thornbury’s article on metaphors in EFL) by the classic work on metaphor analysis by Lakoff & Johnson (1980).

Mason identifies four common categories of metaphor in discourse about ed tech. These are:

  • Manual labour metaphors – especially the words ‘tool’, ‘impact’, ‘leverage’ and ‘scaffold’
  • Mechanistic metaphors – especially the word ‘integration’
  • Journey metaphors – words like ‘access’, ‘route’ and ‘moving beyond’
  • Biological organism / agent metaphors – words like ‘evolution’, ‘next generation’, ‘adopt’, ‘reinforce’ or ‘deepen’

Mason’s taxonomy holds good, too, for specifically ELT contexts. I took a close look at one text, a blog post from Pearson, ‘The potential of AI in English language teaching’ (5 March, 2024), which I found fairly typical of the discourse about AI directed at English language teachers.

The content of this piece is predictable: a broad introduction to ways that AI is being used in schools, a list of six ways that AI can ‘enhance’ teaching, a further seven vague tips for ‘integrating’ AI, and, finally, a mention of the dangers of cheating and plagiarism, coupled with the reassuring (but unsubstantiated) claim that AI can provide ‘educators with robust mechanisms to ensure the integrity of academic work’. It’s an infomercial, written by a ‘web content producer’, whose job is to reinforce the Pearson brand message – transforming language learning through innovative technology (or something like that). The choice of metaphor reinforces the brand message further.

AI is presented as a ‘tool’ (13 times in this short piece), sometimes with adjectives like ‘useful’ or ‘invaluable’, which can be ‘harnessed’ or ‘leveraged’ to ‘enable’ or ‘enhance’ aspects of learning. The manual labour metaphor is particularly striking in one section where the writer describes how AI ‘can take on the laborious task of grading’. It’s a mechanistic world with lots of ‘integration’ (4 times) and increased ‘efficiency’ (2 mentions). The overall effect is to reduce ‘the complexity of teaching and learning by conceptualizing it in mechanistic terms’, making ‘classrooms seem more amenable to technological solutions’ (Mason, 2018: 550).

The journey metaphor is present, too. AI tools ‘pave the way’ with bespoke ‘pathways’ to desirable destinations with five-star travel.

But it’s the biological organism / agent metaphors that are most represented in this piece. AI is not just a human-like agent: it’s a generous, benevolent one that ‘offers’ (7 times), ‘enhances’ (8 times), ‘ensures’ (4 times) and ‘aids’, ‘enables’ or ‘facilitates’. It’s the gift that keeps on giving.

It is not just commercial texts of this kind that use AI metaphors in this way. See, for example Cogo et al. (2024), with its ‘leveraging’ and ‘offering’ and so on. I don’t mean any criticism of Cogo et al by this. It’s an editorial article which must have required them to read huge amounts of stuff about AI and language learning, plus any amount of related material. It’s inevitable that they should mirror, to some extent, the metaphorical language they were exposed to. Language output is, after all, primarily a function of language input. For any given topic in any given genre, the chunks, including the metaphors, are very predictable. Our language choices are, to a large extent, probabilistically determined. For this reason, it is unsurprising that text that is produced by Generative AI itself – on the topic of AI in education – is the richest source of examples from Mason’s metaphorical menu. So, here’s a challenge. Next time you’re reading social media posts or conference abstracts for the forthcoming IATEFL conference, look out for those about AI. What does the choice of metaphor tell you about how commercial the text is, how critically reflective the writer is, and how likely it is that the text was actually written by AI?

References

Cogo, A., Patsko, L. & Szoke, J. Generative artificial intelligence and ELT, ELT Journal, Volume 78, Issue 4, October 2024, Pages 373–377, https://doi.org/10.1093/elt/ccae051

Eliot, L. (2025) Why Our Metaphors About AI Shape How AI Thinks About Us. Forbes, 14 Feb, 2025. https://www.forbes.com/sites/lanceeliot/2025/02/14/why-our-metaphors-about-ai-shape-how-ai-thinks-about-us/

Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago, IL: The University of Chicago Press.

Mason, L. (2018). A critical metaphor analysis of educational technology research in the social studies. Contemporary Issues in Technology and Teacher Education, 18(3), 538-555. https://www.researchgate.net/publication/327560850_A_Critical_Metaphor_Analysis_of_Educational_Technology_Research_in_the_Social_Studies

Mitchell, M. (2024) The metaphors of artificial intelligence. Science, Vol. 386, Issue 6723 https://www.science.org/doi/10.1126/science.adt6140

Rudolph, J., Ismail, F., Tan, S. & Seah, P. (2025) Don’t believe the hype. AI myths and the need for a critical approach in higher education. Journal of Applied Learning & Teaching, 8 / 1: 6 – 27 https://journals.sfu.ca/jalt/index.php/jalt/article/view/2879/983

Thornbury, S. (1991) Metaphors we work by: EFL and its metaphors, ELT Journal, Volume 45, Issue 3, July 1991, Pages 193–200, https://doi.org/10.1093/elt/45.3.193

Weller, M. (2022) Metaphors of Ed Tech. Athabasca: AU Press. https://www.aupress.ca/app/uploads/120309_Weller_2022-Metaphors_of_Ed_Tech.pdf

AI and ELT jobs

Posted: February 27, 2025 in big data, ed tech, investment, learning outcomes, Online learning, politics
Tags: , , , , , , , , , , , , , , ,
3

In the British Council’s recently commissioned report on AI and English language teaching, the ‘Executive Summary’ suggests that, ‘when it comes to AI replacing humans, the majority view is that AI will not replace the need for human teachers any time soon’ (Edmett et al., 2024: 11). This is based on the opinion of 1112 English language teachers from around the world, the majority of whom teach children or teenagers in either face-to-face or in blended scenarios. The most common reason for this belief is that AI ‘cannot substitute the unique human touch’ (ibid., 39). Although not mentioned in this report, these teachers are also almost certainly aware that one of the core functions of teachers – the supervision of young people during the working day – cannot easily be replaced by technology, as we discovered during the COVID pandemic.

The ‘majority view’ referred to in the British Council’s ‘Executive Summary’ is, however, not really much of a majority at all. Later in the report, we see that 24% of teachers surveyed agree or strongly agree with the statement ‘by 2035, AI will be able to teach English without a teacher’ and 26% expressed no opinion. This leaves 50% who are reasonably confident that their jobs are not threatened. The ‘Executive Summary’ could have read ‘when it comes to AI replacing humans, only half of the respondents were confident that AI will not replace the need for human teachers any time soon’. Perhaps the authors of the report wanted to offer a reassuring message. In any case, given the relatively small sample size and the restricted demographics of respondents, the Council’s survey on this question tells us very little.

There are obvious reasons why teachers might feel their jobs are under threat. The press is full of alarming / alarmist headlines like ‘Smart robots could soon steal your job’ (Kottasova, 2016), echoing pronouncements from powerful players. These include a British university vice-chancellor (Seldon, 2017) who predicted that robots will replace teachers in the classroom by 2017, the World Economic Forum (2019) – ‘Professor Robot – why AI could soon be teaching in university classrooms’, and, very recently, Elon Musk, who believes that AI can do a better job than teachers.

Still, in the world of English language teaching, it seems that most commentators agree that AI will not replace teachers. Such voices include publishers like Cambridge English (‘AI enhances English learning, but will not replace the teacher’), academics (e.g. Ahn et al., 2024), and employers like Preply (‘AI will Fire up, not Fire our teachers’) (Voloshyn, 2024). Not only are jobs not threatened, apparently, but we will be empowered as educators, and AI will make us even better teachers, by providing learner support, inclusive and accessible content, and teacher support in lesson design, content creation, test creation, and assistance with feedback and evaluation (Cogo et al., 2024). If we can just sort out a few environmental and ethical concerns, everything will be fine and dandy.

So, who’s right? The neoliberal gang of Musk, the World Economic Forum and Duolingo founder, Louis von Ahn (Nieva, 2024) who recently said that AI ‘may put one-on-one human tutors out of business. I understand that’? Or the more liberal voices of ELT?

Two key facts are worth bearing in mind. First, the costs of education are currently driven primarily by teachers’ wages, which, currently in the EU, account for 64% of education budgets. Second, the education sector is increasingly becoming privatized in various ways (OECD, 2021). If you’re looking to improve profit margins, one of the first places to turn is the wage bill. If, at the same time, your costs increase through spending on, say, an AI licence, your wage bill comes under further pressure. The whole point of privatized, technologized education is to generate profit. As Audrey Watters (2023) has documented, entrepreneurs and technologists have spent the last century trying to devise ways to replace teachers with machines. Talk of improved learning outcomes is mostly a smokescreen to cover financial gains (Saltman, 2022). This does not necessarily lead to immediate job losses, but it would be very surprising if it did not lead to a lowering in pay and conditions. At some point, jobs will go, but it’s not machines that will take them away, it’s the entrepreneurs and executives (investing in those machines) who are looking to increase profit (Merchant, 2023).

It is common to read that, by liberating teachers from tasks like marking, lesson planning and testing, AI will ‘free up’ teachers to focus on ‘higher-level teaching tasks that require more human input’ (Voloshyn, 2024). However, the inescapable logic is that this means that teachers will be deskilled to a greater or lesser extent, they will have less work to do, and can, therefore, be paid less. And so they can be forced to work more.

Teachers in state sectors are, for the time being, relatively protected against salary cuts, even though their salaries have been falling in real terms (in most countries) in the last ten years. It is in the growing private sector where AI will inevitably hit hardest. There are huge numbers of online English tutors. Preply alone (and it is far from the biggest provider) claims that learners can ‘choose from 33281 online English tutors’. The average price of a one-hour lesson on Preply is between $15 and $17. Preply’s commission rate varies from 33% to 18% – you can do the maths! Teachers compete directly with each other to get the lessons on offer. It’s part of the platformized teaching gig economy, where market forces drive down wages as low as they will go. The threat to this precariat comes from the rapidly growing number of fully automated, GPT-powered, always available AI tutors. There’s https://talkpal.ai/ at $6.25 a month if you sign up for two years. There’s https://languatalk.com/ai-language-tutor at 14.90 Eur a month if you sign up for a year. There’s https://yourteacher.ai/ at $12.25 a month if you sign up for a year. There’s Duolingo Max at $14 a month, if you sign up for a year. And countless others. AI won’t replace teachers …. really?

Meanwhile, last year, a grade school in Arizona received permission ‘to offer a fully online school experience for grades four through eight that gives students just two hours of academic instruction per day — taught entirely by artificial intelligence’ (Prada, 2024). The experiment may fail, but it’s a safe bet that others will try it out.

I remain unconvinced that AI will lead to significant improvements in language teaching. There is potential, sure, but, as with all educational technologies, it is unlikely to be realised. Far more likely, is that the pressures on (at least, some) teachers’ jobs and wages will get worse.

References

Ahn, J., Lee, J & Son, M. (2024) ChatGPT in ELT: disruptor? Or well-trained teaching assistant? ELT Journal 78 / 3: 345 – 355

Cogo, A., Patsko, L. & Szoke, J. (2024) Generative artificial intelligence and ELT. ELT Journal, 78 / 4: 373–377

Edmett, A., Ichaporia, N., Crompton, H., & Crichton, R. (2024). Artificial intelligence and English language teaching: Preparing for the future (Second edition). British Council. https://doi.org/10.57884/78EA-3C69

Kottasova, I. (2016) Smart robots could soon steal your job. CNN Business January 15, 2016 https://money.cnn.com/2016/01/15/news/economy/smart-robots-stealing-jobs-davos

Merchant, B. (2023) Blood in the Machine. New York: Little Brown

Nieva, R. (2024) Duolingo’s Billionaire Founder Is All In On AI. Forbes, 24 September 2024. https://www.forbes.com/sites/richardnieva/2024/09/24/duolingo-luis-von-ahn-billionaire-ai-tutor/

OECD (2021) Education at a Glance 2021: OECD Indicators. OECD Publishing, Paris, https://doi.org/10.1787/b35a14e5-en

Prada, L. (2024) This Grade School Offers AI-Only Classes, No Teachers Involved. Vice, 20 December, 2024. https://www.vice.com/en/article/grade-school-ai-only-classes-no-teachers/

Saltman, K. J. (2022) The Alienation of Fact: Digital Educational Privatization, AI, and the False Promise of Bodies and Numbers. Cambridge, Mass.: MIT Press

Seldon, A. (2017) Robots will replace teachers in the next ten years, Vice-Chancellor reveals. The University of Buckingham News 12 September 2017 https://www.buckingham.ac.uk/news/robots-will-replace-teachers-in-the-next-ten-years-vice-chancellor-reveals/

Voloshyn, D. (2024) AI will Fire up, not Fire our teachers. Preply blog, April 4, 2024. https://preply.com/en/blog/ai-will-fire-up-not-fire-our-teachers/

Watters, A. (2023) Teaching Machines. Cambridge, Mass.: MIT Press

World Economic Forum (2019) Professor Robot – why AI could soon be teaching in university classrooms. World Economic Forum blog April 13, 2019 https://www.weforum.org/stories/2019/04/what-robots-and-ai-may-mean-for-university-lecturers-and-students/

Does ChatGPT enhance learning (or does it make you stupid)?

Posted: December 20, 2024 in critical thinking, ed tech, research
Tags: , , , , , , , , , ,
1

Forgive me if you think these are badly-framed questions. Asking whether ChatGPT enhances learning is no different from asking whether technology more generally will enhance learning (a question that is fundamentally unanswerable), something that I had hoped we had moved on from some time ago. But please bear with me. Such questions matter (not because of any answers they may or may not provide), but because their framing is indicative of a techno-determinist mindset, which is itself inseparable from the marketing of edtech products.

Does ChatGPT enhance learning?

This question forms the title of a recently published (today!) systematic review / meta-analysis by Deng et al. (2024). Like so many meta-analyses, it has its problems (mostly acknowledged by the authors). The studies they aggregate were predominantly of relatively small size (very few with over 100 participants), of short duration (typically under 10 weeks, but many under 4 weeks), overwhelmingly with university learners in a variety of academic subjects in Asian settings. It is also clear that ChatGPT was used in a wide variety of ways. Experimental design was not always up to scratch in some of the studies: in some, ChatGPT was not allowed in final evaluation, in others, it was. I’m not sure that anything useful or actionable can be learnt from such a study, but the authors’ main finding is that ‘ChatGPT enhances academic performance’.

‘Academic performance’, of course, is not the same as learning, although it may be used as a proxy for it, and the authors mistakenly equate the two. The use of ChatGPT may well enhance academic performance, in the same way that a line of coke may enhance sporting performance. A quick, measurable boost during the event, but not sustainable afterwards. And there may be detrimental effects, too.

Performance over the very short term, therefore, tells us nothing. The authors also found that ChatGPT boosts affective-motivational states (although not nearly as much as performance), which would be a good thing for learning. Except that these positive affective-motivational states may be entirely due to novelty effects, and we cannot be sure that they were actually enhanced, since they were measured by self-reporting. The authors also found that higher-order thinking propensities (i.e. critical thinking) were also improved, but much less so than performance or affective-motivational states. You have to wonder what has happened to account for the fact that performance improves much more than critical thinking.

On the downside, though, the results indicated that ChatGPT did not generate significant gains in self-efficacy (surely, a crucial component of learning). It also led, surprise-surprise, to less mental effort. Since it is likely that a reduction in cognitive load may limit the depth of content engagement (Stadler, et al., 2024), this isn’t good news for learning, either.

This meta-analysis fails entirely to answer its initial question, which was the wrong question to ask in the first place. The only meaningful question concerns how the technology is used in specific settings with specific groups of learners. That is not a question that can be addressed via meta-analysis. The paper should never have been published.

Does ChatGPT make you stupid?

Researchers who are less enthusiastic about ChatGPT pick up on the question of mental effort (see, for example, Zhai et al., 2024). They draw attention to reduced cognitive engagement, over-reliance on the technology and passivity in learning. In other words, it makes us stupid, and the only way to address the issue is with more educational focus on critical thinking. Recent frameworks of AI literacy / competency make a big deal out of the need for critical thinking (see, for example, Fengchun, M. & Kelly, S., 2024). This makes intuitive good sense, but let’s pause before jumping on the bandwagon. Firstly, we struggle to define critical thinking and there is no consensus about how to teach it, or, indeed, whether it can be taught at all. Secondly, it’s not ChatGPT that leads to deleterious effects, but the way it is used. We can no more prove that ChatGPT makes you stupid than we can demonstrate that it will enhance learning.

Concerns about the cognitive dangers of ChatGPT have precedents going back millennia (Bell, 2010). Back in 2010, Nicholas Carr’s best-selling ‘The Shallows’, expanded on a 2008 article called ‘Is Google making us stupid?’ (Carr, 2008). Carr argued that the outsourcing of cognitive effort to the internet, the way the internet discouraged critical reading, the danger of trusting unreliable online sources, led to cognitive decline. As with the current hoo-ha around ChatGPT, responses varied, but it was mostly a matter of opinion rather than facts (since there was a lack of good research evidence, and since the question was not really amenable to empirical enquiry).

Go back further to another ground-breaking knowledge technology: the printing press. The new ease of access to knowledge would, said critics (like the wonderfully named Hieronimo Squarciafico), cause problems when combined with unreliable information, leading to superficial (mis)understanding. Or, back to the use of writing itself, which, according to no less a light than Socrates, leads to ‘forgetfulness in the minds of those who learn to use it, because they will not practice their memory. [It offers] your pupils the appearance of wisdom, not true wisdom, for they will read many things without instruction and will therefore seem to know many things, when they are for the most part ignorant and hard to get along with, since they are not wise, but only appear wise.’

Writing, the printing press and digital text on the internet all became normalised and valued. They did not cause cognitive decline. ChatGPT and others like it are likely to follow the same course. If we’re looking for a reason to resist the encroachment of generative AI into education, this isn’t it. There are other, much more powerful and pressing reasons. First among them, for me is the monopolistic control of knowledge production by a handful of extremely rich, profit-driven corporations – another intriguing parallel between ChatGPT and 15th / 16th century printing or late 20th century internet.

References

Bell, V. (2010) Don’t Touch That Dial! Slate, Feb 15, 2020. https://slate.com/technology/2010/02/a-history-of-media-technology-scares-from-the-printing-press-to-facebook.html

Carr N. G. (2008) Is Google making us stupid? The Atlantic, July / August 2008

Carr N. G. (2010) The Shallows. New York: W. W. Norton

Deng, R., Jiang, M., Yu, X., Lu, Y. & Liu, S. (2024) Does ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies. Computers & Education, 2024, 105224, https://doi.org/10.1016/j.compedu.2024.105224

Fengchun, N. & Kelly, S. (2024) AI competency framework for students. UNESCO, https://unesdoc.unesco.org/ark:/48223/pf0000391105

Stadler, M., Bannert, M. & Sailer, M. (2024) Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry. Computers in Human Behavior, Volume 160, 108386, https://doi.org/10.1016/j.chb.2024.108386

Zhai, C., Wibowo, S. & Li, L.D. (2024) The effects of over-reliance on AI dialogue systems on students’ cognitive abilities: a systematic review. Smart Learn. Environ. 11, 28 (2024). https://doi.org/10.1186/s40561-024-00316-7

Edtech winners and losers

Posted: November 19, 2024 in apps, Discourse, ed tech, investment, Online learning
Tags: , , , , , , , , , ,
2

If you were watching the 2024 Qatar World Cup, you may have noticed the brand Byju’s, among the list of official corporate sponsors alongside Visa, Adidas, Coca Cola and McDonald’s. You may have also noticed the pitch-side adverts for Byju’s, which flipped to read ‘The Future of Learning’. You may also have noticed photos of Byju’s Global Brand Ambassador, Lionel Messi, wearing a Byju’s strip. On messi.com, the football superstar says, ‘I chose to partner with BYJU’S because their mission to make everyone fall in love with learning perfectly aligns with my values. High-quality education changes lives, and BYJU’S has transformed the career paths of millions of students worldwide. I hope to inspire young learners to reach and remain at the top’.

Messi’s fee is estimated to be something between $5 and $7 million per year. If you were watching the Qatar World Cup, but missed the Byju’s campaign, it just goes to show that a $40 million sponsorship punt doesn’t always have the desired impact. But even if your awareness of the Byju’s brand was sharpened by the Qatar World Cup, it turns out that it was all money down the drain.

Byju’s is an Indian edtech company offering online tutoring and tutorial videos in a range of subjects, including English language. The company grew rapidly after its founding in 2011, flourishing especially as a result of the Covid pandemic. With backing from the Chan-Zuckerberg Initiative, Black Rock, Tencent, Sequoia Capital, the Qatar Investment Authority, and many others, the company was valued at $22 billion by the time of the World Cup, making it the most valued edtech company in the world. To put that into context, it is a figure significantly greater than the entire Indian national budget for school and higher education.

But, now, Messi’s contract is on hold and Byju’s is effectively ‘worth zero’, according to the founder, Byju Raveendran. Even before the World Cup, the company was on a losing streak, as financing dried up, and the bubble it had built was no longer sustainable. Key investors pulled out, and it was accused of ‘‘toxic’ targeting of low-income families, high-pressure sales, misleading advertising, and exacerbating existing educational inequalities’. It was 18 months late in its financial reporting, and cut thousands of staff. With the benefit of hindsight, it now seems likely that Byju’s World Cup advertising was not primarily targeting the likes of you and me, but potential investors. It was seeking to position the company as a financial brand, and it failed. Byju’s American affiliate is bankrupt. Sportswashing doesn’t always work.

Big edtech companies come and go (remember Knewton!). In the case of Byju’s, the immediate cause is a group of investors pulling the plug and writing off their investments. But the more interesting question is what prompted their lack of confidence. In an attempt to answer this question, a blog from Cornell University’s Business school suggests, no matter how much you wrap yourself in the discourse of social good, ‘the product matters’, and Byju’s product simply wasn’t very good. The most common criticism was that the app was buggy, but what about the educational approach?

Byju’s has a page on their website called ‘How to Learn English? Explore Tips and Techniques’. It’s worth a visit. Here are some highlights:

If you’re in the mood for more, there’s a Byju’s video on YouTube called ‘English Speaking Practice’. The lesson doesn’t actually start until 13.52, so you may want to skip the beginning. Students are asked to practise saying ‘February’, ‘library’, ‘nuclear’, ‘aluminum’ and ‘aluminium’, ‘mischievous’, ‘sherbet’, ‘asterisk’, ‘espresso’, ‘athlete’, ‘caramel’, ‘familiar’, ‘often’, ‘supposedly’, ‘height’, ‘develop’, ‘Wednesday’, ‘squirrel’, ‘realtor’, ‘cavalry’ and ‘Arctic’. After 12 minutes of this, the teacher, Asmita Ma’am (who was also Curriculum Manager of Byju’s English), moves on to practice of reading aloud (an extract of a short story by Ruskin Bond and Wordsworth’s ‘I wandered Lonely’), answering questions from the students along the way. There are about nine minutes of this, before we get to ‘the most important part of the lesson’ – ‘Introducing Yourself’. I won’t spoil things by telling you what happens.

Looking at the difficulties faced by Byju’s, it is hard to resist the temptation to indulge in a moment of schadenfreude. But it is not Byju’s who have really suffered. That is the lot of the relatively disadvantaged students whose time and whose parents’ money has been wasted in the vain hope of changing lives and transforming career paths. It is also the lot of the thousands of employees who have lost their jobs. Meanwhile, Byju’s, and its English-language teaching subsidiary, Hello English, are still out there flogging their wears, while insolvency proceedings grind on in the Indian courts. Byju Raveendran is holed up in Dubai, from where he blames the big creditors for the company’s downfall. Although I have no sympathy with him, he is probably right. He argues that investors backed him aggressively during the period of acquisitions, but pulled out at the first sign of trouble. It wasn’t, perhaps, at the first sign of trouble, but, yes, that’s what investors do. Some will shed their shares not long after purchase, making a healthy profit along the way. In the case of Byju’s. the Chan Zuckerberg Initiative falls into this category. When the bubble bursts, it’s too late and companies like Prosus can find themselves sitting on shareholdings that are fast declining in value. They can pump more money in to keep the company afloat, but they risk losing more in the process. Or they can jump ship, as Prosus did, and take a hit of $500 million. But since this is less than 10% of their annual revenue and since they know how to hedge their bets, we don’t need to worry about them too much. It’s just part of the swings and roundabouts of betting on edtech.

The initial appeal of Byju’s to investors was the size of the Indian market (over 250 million students / consumers), the potential of technology to reach a significant slice of this market, and a growth in edtech that is outstripping global growth. The ‘ed’ part of the edtech never had much to do with it. It could have been online betting, fintech or crypto. If there are enough small-time suckers who are ready to splash some cash, there is money to be made … if you have enough capital and you spread your bets.

Talking of which, at about the same time that Messi signed the contract with Byju’s, he signed another to become brand ambassador for the cryptocurrency exchange, BitGet. Two years on, a BitGet / Messi campaign , focusing on small-time investors in Latin America, Southeast Asia and Turkey, aims to ‘inspire everyone to push beyond limits and seize opportunities, transforming challenges into stepping stones toward success’. And just to bring things full circle, I learn that BitGet has entered into a partnership with the University of Zurich which will ‘elevate blockchain education, fostering innovation and expanding global learning opportunities’.

5 top tips for using ChatGPT in language teaching

Posted: October 22, 2024 in apps, ed tech, Feedback, practical ideas, vocabulary
Tags: , , , , , , , , , , , , , , ,
0

The July 2024 edition of ELT Journal (78 / 3) contains an article entitled ‘ChatGPT in ELT: disruptor? Or well-trained teaching assistant?’ (Ahn et al., 2024). It aims to ‘contribute to the growing discourse on technology integration in ELT and offers practical recommendations for creating a productive learning environment using AI-driven language models like ChatGPT.’ I’m not sure that the authors have actually tried out any of their ideas, but I have. So, to save you the trouble of reading the article, here are the authors’ top ways of using ChatGPT.

Generate content that aligns with students’ changing needs and interests

We are told that we can keep our ‘lessons engaging and relevant’, enhancing ‘student engagement and motivation, fostering a more stimulating learning environment’ by creating texts on topics of interest to our learners. If only! The topics may be of interest, but the same cannot be said of the texts. Texts generated by ChatGPT are typically ‘dull or uninspired, riddled with clichés or recycled ideas’ (Mimbs Nyce, 2023). So much so that ‘sounds like a bot’ has become shorthand for ‘boring’ (Robertson, 2024). A Guardian review of last year’s movie, ‘Ghosted’, described it as ‘so carelessly and lifelessly cobbled together that we’re inclined to believe it’s the first film created entirely by AI’.

I asked ChatGPT to create a 200-word text for low-intermediate EFL learners about Dua Lipa with all the latest news and gossip. The result was, at least, current and accurate. The sources were justjared.com, hellomagazine.com and the Bing news aggregator. But the text was as dull and uninspired as you might expect. Far better to let students find their own source of Dua Lipa news (if that’s what they want) and use a dictionary plug-in, if necessary, to read it. What’s more, they’ll get some pictures.

Adjust the difficulty of texts to align with proficiency levels such as those of the CEFR

According to Ahn et al. (2024), ChatGPT can ‘generate texts with appropriate vocabulary choices based on the target learners’ proficiency.’ The 164-word B1 text about Dua Lipa that I got included 13 items ‘above level’ (using both the Pearson GSE and the Oxford Text Checker as references) – see this sentence, for example, ‘she was also praised for her edgy, sporty-chic look on social media.’

ChatGPT has only a very vague grasp of level. This is hardly surprising. Pearson GSE, Oxford Text Checker and Cambridge Profile disagree, as often as not, on what level to assign a particular item. At best, these level tags are rough guides, since difficulty depends on a whole host of factors besides lexis and syntax – the L1 and domain knowledge of the individual learner being especially significant. Neither of which can easily be incorporated into ChatGPT prompts.

Of course, a teacher could manually edit the text, changing some words and structures. They could also, of course, just find a more interesting text and show the learners how to use a dictionary plug-in. I guess it depends on why the teacher wants the students to read the text in the first place (see below).

Generate comprehension questions for a text

Ahn et al. (2024) confuse the assessment of ‘students’ understanding of the material’ with the development of their reading skills. Perhaps someone could pass them a copy of Bill Grabe’s ‘Reading in a Second Language’ (2009). While they’re at it, they might like to throw in a copy of John Field’s ‘Listening in the Language Classroom’ (2008). Even if you want to test students’ reading comprehension with true / false or multiple choice questions, ChatGPT is not a great tool to use. As in the example given by Ahn et al. (2004: 350), it tends to either (1) pick up on individual lexical items (requiring the student only to scan in order to find the same item in the text), or (2) offer synonyms which have to be matched to corresponding items in the text (i.e. a test of lexical knowledge, rather than reading comprehension). And, as Ahn et al.’s example illustrates, it can’t do inference questions very well either. This is not very different from poorly written coursebook material, but that is hardly something to recommend it.

Provide grammatical feedback throughout the writing process

How? Well, Ahn et al. (2024: 352) suggest that ‘the process starts with students writing their own compositions and then inputting them into ChatGPT to receive corrective feedback.’ As a pedagogical approach (extremely common though it may be), this contains not a single one of the characteristics of effective feedback (see, for example, this paper). There may be a role for corrective feedback at some (later) stage of the writing process but, to be beneficial to learning, this feedback needs to be selective, tailored to individual learner needs and reliable. ChatGPT is none of these. It is certainly helpful in coming up with a reasonably accurate end-product, but its value is very limited in terms of learning to write in another language.

Facilitate speaking practice

Ahn et al. (2024: 350-351) recommend using ChatGPT as an interlocutor in ‘role-plays of real-life communicative situations.’ For some independent learners without access to real-life interlocutors, there is indisputable value here. But there are three big problems. Firstly, it doesn’t really do ‘speaking’. It generates written text which is converted to speech. It has none of the features of spoken language. Secondly, it is programmed to be polite at all times: it can’t do pragmatics. Lastly, it struggles with many accents.

So, yes, ChatGPT could be used in this way. It may provide useful listening and lexical input, but the dialogues are rarely ones that ‘could plausibly occur in real life’ (Ahn et al., 2024: 351). They’re like listening to a religious data entry worker, quick to avail themselves of any opportunity for a spot of mansplaining, the dialogic equivalent of ChatGPT written texts. It only gets fun or interesting when you deliberately set out to provoke the software.

Remember the adage ‘garbage in, garbage out’

Ahn et al. (2024) remind us that ChatGPT is ‘not designed as a language learning website – it is vulnerable to the ‘garbage in, garbage out’ problem.’ Very true. Finding good ideas for using ChatGPT in the language classroom (as opposed to self-study) is not easy – suggestions gratefully received.

References

Ahn, J., Lee, J. & Son, M. (2024) ChatGPT in ELT: disruptor? Or well-trained teaching assistant? ELT Journal, 78 (3): 345 – 355

Field, J. (2008) Listening in the Language Classroom. Cambridge: Cambridge University Press.

Grabe, W. (2009) Reading in a Second Language. Cambridge: Cambridge University Press.

Mimbs Nyce, C. (2023) AI is an insult now. The Atlantic, May 30, 2023

Robertson, A. (2024) You sound like a bot. The Verge, February 16, 2024

AI, DEI and ELT

Posted: October 14, 2024 in ed tech
Tags: , , , , , , , , , , , , , ,
0

Technology (which usually means AI these days) and inclusive practices in education are being linked with increasing frequency. Earlier this year, the IATEFL Learning Technologies SIG organised a pre-conference event at the annual IATEFL fair entitled ‘Researching and promoting inclusive language practices through technology’. The ‘big debate’ at the same event was on the same topic. Next year, they are calling their event ‘Tech-enhanced Pedagogy for Equity and Critical Thinking in the AI Era’. This month, at the ELT Malta conference, there will be presentations about ‘Harnessing AI for inclusive ELT: promoting diversity, equity, and accessibility’, ‘Using AI ethically to promote diversity and inclusion in ELT’ and ‘Enhancing bi/multilingual English language teaching with artificial intelligence’.

But will AI contribute to DEI in ELT?

It is not hard to think of instances where AI could ‘aid accessibility for some learners, for instance, allowing individuals with visual impairments to use speech to interface with computers’ (Edmett et al., 2024: 11). It is also true that AI can be used to quickly generate multimodal texts and other learning materials (including feedback) that may be more appropriate and inclusive than those typically found in internationally published coursebooks. Other examples can be found.

However, as the British Council report (Edmett et al., 2024: 11) points out, ‘the challenges of AI use in ELT are underreported’. Two issues are highlighted. The first is that AI applications reflect the models on which they are trained, and these are almost invariably corpora of standardised, white, middle-class, native-speaker English. The result is about as far as one could imagine from an anti-racist pedagogy (Ramjattan, 2024): prejudice and bias are reinforced. More generally, texts and images generated by AI are likely to perpetuate the stereotypes and prejudices. There is a well-documented history of discriminatory outcomes because of people’s race, gender, social class or disability profile (O’Neil, 2016).

The second issue that the British Council highlights is the ‘potential for AI to wider digital divides (for example, when ‘AI is widely adopted in better-resourced education systems but not in lower resourced systems’ (Edmett et al., 2024: 11).

One major point struck me from the British Council survey of teachers. There is no reporting of AI use to address DEI issues in actual classrooms. I was hoping to find out more about actual DEI use (in ELT) of AI by reading a recent article, ‘How does generative AI promote autonomy and inclusivity in language teaching?’ (Szabó & Szoke, 2024). The choice of auxiliary verb (‘does’, as opposed to ‘could’ or ‘might’) in the title suggests that generative AI actually does promote inclusivity, but the article itself is restricted to descriptions of potential use (of the kind listed at the start of this section). As the article progresses, the auxiliary ‘does’ is dropped in favour of ‘can’. There are no referenced descriptions of AI use to help English language learners with specific learning needs, how effective these might or might not be, or how widespread their use is. Although I know that there are many ways in which AI could be used to help such learners, my suspicion is, sadly, that these are not especially widespread.

On the other hand, Szabó & Szoke (2024) provide plenty of references to support their view ‘that the use of GenAI in language classrooms will continue to deepen existing inequities’ (of access).

The discoursal context

All the talk about AI and DEI in ELT is part of a broader discourse: it reflects and reinforces this broader discourse. Here are a few examples of this broader discourse.

At next year’s BETT show (the number one educational technology sales event), the top ‘global theme’ will be ‘diversity and inclusion’: ‘Prioritising inclusion means using accessible technology and design principles that accommodate different learning styles. This approach removes barriers, celebrates diversity and empowers underserved communities to fully participate in learning’. The sponsor of this global theme is HP, the manufacturer of computers and printers. According to Enrique Loras, president of HP, ‘we embed diversity and inclusion into everything we do’. This no doubt includes their provision of servers, data storage and data security for Israeli prisons.

The opening plenary at next month’s online Edtech World Forum (another sales event) is entitled ‘Leveraging Technology and Media to Design more Inclusive Learning Opportunities’. The speaker, Amin Marei, lectures at Harvard, but is also the co-founder of Edlabs, a private edtech consulting company. Main sponsors of the event include Coursera and Google. Other keynote speakers include the CEO of SchoolOnline.ai, a Google edtech account executive, the CEO of the WordUp app who is currently ‘launching a brand new UK university, focused on AI, tech and innovation’, and the CEO of a company selling personalized online instruction to kids (using their preferential learning styles). The fully-inclusive, full-price to attend the two-day (virtual) event is £529.

Google’s Chromebooks now dominate the global secondary education market. The company likes to trumpet their listing as the World’s Most Inclusive Brand on Kantar’s 2024 Brand Inclusion Index. Their Chief Diversity Officer, Melonie Parker (she / her), writes that ‘building belonging for everyone means ensuring no one is left out and each person can thrive’. Last October, a New York jury found the company guilty of sex discrimination and retaliation, and ordered it to pay $1.15 million to the woman concerned.

Yes, DEI is currently being used to sell educational technology. A year or two back, it was edtech and social-emotional learning; before that, it was edtech and twenty-first century skills. And for a while it was (and still sometimes is) edtech and learning styles. But right now, it’s DEI. As one company of edtech sales and marketing consultants puts it, ‘diversity, equity, and inclusion (DEI) are crucial to a successful business strategy’. This isn’t purely woke-washing, but woke-washing is an important part of it. This ‘sanitises toxic business practices’ although, as Guardian columnist, Arwa Mahdawi, puts it, there may be ‘something to be said for the fact that big companies feel compelled to don a progressive veneer’.

I don’t want to argue that AI cannot help DEI initiatives in English language teaching / learning. I certainly don’t want to suggest that all of those conference presentations about AI and DEI are anything other than very well-meant. Without them, AI in ELT would be even less likely to lead to any measurable gains in D, E or I. But I think it’s important to remember that, on the whole, AI is possibly more likely to reinforce prejudices and practices that we should have abandoned long ago. The association of AI with DEI is probably more beneficial to the cause of AI than it is to the cause of DEI. You don’t need to be a cynic to realise that all that stuff about DEI at BETT and the Edtech World Forum is sales-driven. My worry is that the main beneficiaries of the current focus on DEI and AI in language teaching will not be the under-represented and under-served diversity of learners, but the vendors and those others who make a living in their wake.

References

Edmett, A., Ichaporia, H. & Crichton, R. (2024) Artificial Intelligence and English Language Teaching: Preparing for the Future (Second Edition). British Council.

O’Neil, C. (2016) Weapons of Math Destruction. London: Allen Lane

Ramjattan, V. A. (2024) Imagining an anti-racist pedagogy. ELT Journal, 78 / 3: 318 – 325

Szabó, F. & Szoke, J. (2024) How does generative AI promote autonomy and inclusivity in language teaching? ELT Journal, 27 September 2024 https://academic.oup.com/eltj/advance-article/doi/10.1093/elt/ccae052/7784519

Multimodality and modals

Posted: November 23, 2023 in 21st century skills, critical thinking, learning outcomes, Online learning, research, video
Tags: , , , , , , , , , , , , ,
2

My attention was recently drawn (thanks to Grzegorz Śpiewak) to a recent free publication from OUP. It’s called ‘Multimodality in ELT: Communication skills for today’s generation’ (Donaghy et al., 2023) and it’s what OUP likes to call a ‘position paper’: it offers ‘evidence-based recommendations to support educators and learners in their future success’. Its topic is multimodal (or multimedia) literacy, a term used to describe the importance for learners of being able ‘not just to understand but to create multimedia messages, integrating text with images, sounds and video to suit a variety of communicative purposes and reach a range of target audiences’ (Dudeney et al., 2013: 13).

Grzegorz noted the author of this paper’s ‘positively charged, unhedged language to describe what is arguably a most complex problem area’. As an example, he takes the summary of the first section and circles questionable and / or unsubstantiated claims. It’s just one example from a text that reads more like a ‘manifesto’ than a balanced piece of evidence-reporting. The verb ‘need’ (in the sense of ‘must’, as in ‘teachers / learners / students need to …’) appears no less than 57 times. The modal ‘should’ (as in ‘teachers / learners / students should …’) clocks up 27 appearances.

What is it then that we all need to do? Essentially, the argument is that English language teachers need to develop their students’ multimodal literacy by incorporating more multimodal texts and tasks (videos and images) in all their lessons. The main reason for this appears to be that, in today’s digital age, communication is more often multimodal than not (i.e. monomodal written or spoken text). As an addendum, we are told that multimodal classroom practices are a ‘fundamental part of inclusive teaching’ in classes with ‘learners with learning difficulties and disabilities’. In case you thought it was ironic that such an argument would be put forward in a flat monomodal pdf, OUP also offers the same content through a multimodal ‘course’ with text, video and interactive tasks.

It might all be pretty persuasive, if it weren’t so overstated. Here are a few of the complex problem areas.

What exactly is multimodal literacy?

We are told in the paper that there are five modes of communication: linguistic, visual, aural, gestural and spatial. Multimodal literacy consists, apparently, of the ability

  • to ‘view’ multimodal texts (noticing the different modes, and, for basic literacy, responding to the text on an emotional level, and, for more advanced literacy, respond to it critically)
  • to ‘represent’ ideas and information in a multimodal way (posters, storyboards, memes, etc.)

I find this frustratingly imprecise. First: ‘viewing’. Noticing modes and reacting emotionally to a multimedia artefact do not take anyone very far on the path towards multimodal literacy, even if they are necessary first steps. It is only when we move towards a critical response (understanding the relative significance of different modes and problematizing our initial emotional response) that we can really talk about literacy (see the ‘critical literacy’ of Pegrum et al., 2018). We’re basically talking about critical thinking, a concept as vague and contested as any out there. Responding to a multimedia artefact ‘critically’ can mean more or less anything and everything.

Next: ‘representing’. What is the relative importance of ‘viewing’ and ‘representing’? What kinds of representations (artefacts) are important, and which are not? Presumably, they are not all of equal importance. And, whichever artefact is chosen as the focus, a whole range of technical skills will be needed to produce the artefact in question. So, precisely what kind of representing are we talking about?

Priorities in the ELT classroom

The Oxford authors write that ‘the main focus as English language teachers should obviously be on language’. I take this to mean that the ‘linguistic mode’ of communication should be our priority. This seems reasonable, since it’s hard to imagine any kind of digital literacy without some reading skills preceding it. But, again, the question of relative importance rears its ugly head. The time available for language leaning and teaching is always limited. Time that is devoted to the visual, aural, gestural or spatial modes of communication is time that is not devoted to the linguistic mode.

There are, too, presumably, some language teaching contexts (I’m thinking in particular about some adult, professional contexts) where the teaching of multimodal literacy would be completely inappropriate.

Multimodal literacy is a form of digital literacy. Writers about digital literacies like to say things like ‘digital literacies are as important to language learning as […] reading and writing skills’ or it is ‘crucial for language teaching to […] encompass the digital literacies which are increasingly central to learners’ […] lives’ (Pegrum et al, 2022). The question then arises: how important, in relative terms, are the various digital literacies? Where does multimodal literacy stand?

The Oxford authors summarise their view as follows:

There is a need for a greater presence of images, videos, and other multimodal texts in ELT coursebooks and a greater focus on using them as a starting point for analysis, evaluation, debate, and discussion.

My question to them is: greater than what? Typical contemporary courseware is already a whizzbang multimodal jamboree. There seem to me to be more pressing concerns with most courseware than supplementing it with visuals or clickables.

Evidence

The Oxford authors’ main interest is unquestionably in the use of video. They recommend extensive video viewing outside the classroom and digital story-telling activities inside. I’m fine with that, so long as classroom time isn’t wasted on getting to grips with a particular digital tool (e.g. a video editor, which, a year from now, will have been replaced by another video editor).

I’m fine with this because it involves learners doing meaningful things with language, and there is ample evidence to indicate that a good way to acquire language is to do meaningful things with it. However, I am less than convinced by the authors’ claim that such activities will strengthen ‘active and critical viewing, and effective and creative representing’. My scepticism derives firstly from my unease about the vagueness of the terms ‘viewing’ and ‘representing’, but I have bigger reservations.

There is much debate about the extent to which general critical thinking can be taught. General critical viewing has the same problems. I can apply critical viewing skills to some topics, because I have reasonable domain knowledge. In my case, it’s domain knowledge that activates my critical awareness of rhetorical devices, layout, choice of images and pull-out quotes, multimodal add-ons and so on. But without the domain knowledge, my critical viewing skills are likely to remain uncritical.

Perhaps most importantly of all, there is a lack of reliable research about ‘the extent to which language instructors should prioritize multimodality in the classroom’ (Kessler, 2022: 552). There are those, like the authors of this paper, who advocate for a ‘strong version’ of multimodality. Others go for a ‘weak version’ ‘in which non-linguistic modes should only minimally support or supplement linguistic instruction’ (Kessler, 2022: 552). And there are others who argue that multimodal activities may actually detract from or stifle L2 development (e.g. Manchón, 2017). In the circumstances, all the talk of ‘needs to’ and ‘should’ is more than a little premature.

Assessment

The authors of this Oxford paper rightly note that, if we are to adopt a multimodal approach, ‘it is important that assessment requirements take into account the multimodal nature of contemporary communication’. The trouble is that there are no widely used assessments (to my knowledge) that do this (including Oxford’s own tests). English language reading tests (like the Oxford Test of English) measure the comprehension of flat printed texts, as a proxy for reading skills. This is not the place to question the validity of such reading tests. Suffice to say that ‘little consensus exists as to what [the ability to read another language] entails, how it develops, and how progress in development can be monitored and fostered’ (Koda, 2021).

No doubt there are many people beavering away at trying to figure out how to assess multimodal literacy, but the challenges they face are not negligible. Twenty-first century digital (multimodal) literacy includes such things as knowing how to change the language of an online text to your own (and vice versa), how to bring up subtitles, how to convert written text to speech, how to generate audio scripts. All such skills may well be very valuable in this digital age, and all of them limit the need to learn another language.

Final thoughts

I can’t help but wonder why Oxford University Press should bring out a ‘position paper’ that is so at odds with their own publishing and assessing practices, and so at odds with the paper recently published in their flagship journal, ELT Journal. There must be some serious disconnect between the Marketing Department, which commissions papers such as these, and other departments within the company. Why did they allow such overstatement, when it is well known that many ELT practitioners (i.e. their customers) have the view that ‘linguistically based forms are (and should be) the only legitimate form of literacy’ (Choi & Yi, 2016)? Was it, perhaps, the second part of the title of this paper that appealed to the marketing people (‘Communication Skills for Today’s Generation’) and they just thought that ‘multimodality’ had a cool, contemporary ring to it? Or does the use of ‘multimodality’ help the marketing of courses like Headway and English File with additional multimedia bells and whistles? As I say, I can’t help but wonder.

If you want to find out more, I’d recommend the ELT Journal article, which you can access freely without giving your details to the marketing people.

Finally, it is perhaps time to question the logical connection between the fact that much reading these days is multimodal and the idea that multimodal literacy should be taught in a language classroom. Much reading that takes place online, especially with multimodal texts, could be called ‘hyper reading’, characterised as ‘sort of a brew of skimming and scanning on steroids’ (Baron, 2021: 12). Is this the kind of reading that should be promoted with language learners? Baron (2021) argues that the answer to this question depends on the level of reading skills of the learner. The lower the level, the less beneficial it is likely to be. But for ‘accomplished readers with high levels of prior knowledge about the topic’, hyper-reading may be a valuable approach. For many language learners, monomodal deep reading, which demands ‘slower, time-demanding cognitive and reflective functions’ (Baron, 2021: x – xi) may well be much more conducive to learning.

References

Baron, N. S. (2021) How We Read Now. Oxford: Oxford University Press

Choi, J. & Yi, Y. (2016) Teachers’ Integration of Multimodality into Classroom Practices for English Language Learners’ TESOL Journal, 7 (2): 3-4 – 327

Donaghy, K. (author), Karastathi, S. (consultant), Peachey, N. (consultant), (2023). Multimodality in ELT: Communication skills for today’s generation [PDF]. Oxford University Press. https://elt.oup.com/feature/global/expert/multimodality (registration needed)

Dudeney, G., Hockly, N. & Pegrum, M. (2013) Digital Literacies. Harlow: Pearson Education

Kessler, M. (2022) Multimodality. ELT Journal, 76 (4): 551 – 554

Koda, K. (2021) Assessment of Reading. https://doi.org/10.1002/9781405198431.wbeal0051.pub2

Manchón, R. M. (2017) The Potential Impact of Multimodal Composition on Language Learning. Journal of Second Language Writing, 38: 94 – 95

Pegrum, M., Dudeney, G. & Hockly, N. (2018) Digital Literacies Revisited. The European Journal of Applied Linguistics and TEFL, 7 (2): 3 – 24

Pegrum, M., Hockly, N. & Dudeney, G. (2022) Digital Literacies 2nd Edition. New York: Routledge

Older Entries