Generative AI technologies like ChatGPT are being treated in education with a mix of awe and fear. On one end of the spectrum, you get wild techno-optimism that AI will somehow revolutionise the education system, democratise creativity, and pretty much do everything except perhaps your laundry.
On the other end, there is abject horror at the very idea of allowing AI anywhere near students (as if they’re not going to use it without our permission) and fears about everything from learning loss to academic integrity and the mass unemployment of teachers. As with all technologies, I suspect that over time, we’ll find the middle ground. There will certainly be some horror stories, but they are much more likely to come from the real risks of deepfakes, misinformation, and erosion of trust, rather than from students using ChatGPT to write their Macbeth essays.
I’m interested as a researcher, an author, an educator, and a student in finding that middle ground and using multimodal generative AI as part of text production, in ways where I still feel confident that it’s my voice on the page (or the screen, or the AI-generated voiceover). Whilst it’s certainly possible for models like ChatGPT and Claude to write my blog posts for me, that’s not really the point of writing, as far as I’m concerned. But generative AI can certainly facilitate the publication process and support ways of writing which were only possible in the past couple of years.
Flattening the Strata
As with previous articles on multimodal texts and generative AI, I’m borrowing and adapting language from the social semioticians Gunther Kress and Theo van Leeuwen. This time, I’m focusing on the strata of text production.
Kress and Van Leeuwen have argued that multimodal text and the evolution of writing technologies over time have given extra affordances to writers and creators which have broken down previous hierarchies in text production. For example, print media no longer needs an army of authors, editors, designers, typesetters, printers, maintenance staff for the printing equipment, publishers, and distribution networks.
Even decades-old technologies such as home office laser and ink printers have condensed the multiple roles of publication into a job that can be handled by a single person. Digital technologies, and in particular the technologies which we call Web 2.0, including blogs, wikis, and social media, have made elements of that process even easier. It’s no longer necessary to spend money on ink that’s worth more than its weight in gold to get your message out into the world. And it’s entirely possible to reach millions of people from a single keyboard.
But where Kress and van Leeuwen argue that multimodal texts flatten the hierarchies of text production, they still maintain four distinct strata: discourse, design, production, and distribution. I’ve written about these distinct strata before, in a post where I tried to analyse my own production methods for blog posts like this. As with many of my posts, I’m drafting this one verbally whilst out for a slow and steady run on the roads behind the farm where I live.
As I explained in that earlier post, generative AI is a part of that entire process. But I struggled when I wrote that post, and I still struggle now, to articulate clear differences between the four layers of Kress and van Leeuwen’s strata. The authors make it clear that there is no “hierarchy” between the layers, and that at various times layers may merge and interchange. Kress and van Leeuwen also recognise digital technologies have compressed the strata and made certain aspects such as production and distribution easier through the affordances of digital technologies.
But with Generative AI, I no longer think that it’s possible to clearly delineate between the four layers of discourse, design, production, and distribution. Some of this extends from good old-fashioned Web 2.0 ideas around the creation and sharing of multimodal texts online, but some is being facilitated exclusively in novel ways by new generative artificial intelligence technologies. I think if we want to move forward with our understanding of what these technologies mean for production and distribution of text, we need to re-examine and perhaps collapse the strata.
Discourse, design, production, and distribution are good places to start to delineate different aspects of multimodal text production in my writing. Technological, educational, and societal discourses inform everything I do. I draw on discourses of technology as saviour, as much as discourses of technology as threat. Like many recent publications in higher education about artificial intelligence, I’ve sometimes leaned into and sometimes pushed against what Margaret Bearman, Juliana Ryan and Rola Ajjawi call the “Discourse of imperative response” – act now or be left behind.
But Discourse is not a one-way street. As the popularity of this blog has grown and as I have published academic articles based on my work with generative AI, these texts have started to influence and shape discourse as much as they are shaped by it. Generative artificial intelligence has had a role to play here, not only because it is the topic I predominantly write about, but also because of the technology through which I write and contribute to these discourses.
The tension between using a technology which I have described in the past as profoundly unethical and fatally flawed is palpable in some of the articles I write. Trying to balance the potential with the sometimes overwhelming seeming negatives is hard. But like educators sharing their own thoughts on social media platforms like LinkedIn, Facebook, and X, grappling with these tensions in public makes for some interesting, useful conversations.
The AIAS: Influencing and Influenced by GenAI Discourses
In early 2023, a predominant Discourse of artificial intelligence in education was the “Discourse of ChatGPT as a tool for cheating”. This narrative of threat and concern led to education institutions and sectors across the world attempting to ban and block the technology. Others turned a blind eye to the technology, hoping that it would wash over us, and not become part of the technology landscape as Wikipedia, Google search, and YouTube had before.
But the dichotomies between banning/blocking the technology versus allowing all of its use and pretending it doesn’t exist led me, through a conversation with Edith Cowan University staff, to write a blog post about the AI Assessment Scale (AIAS). If you’ve been here for a while, you’ll know much of the history of the AIAS and that it was picked up by Mike Perkins, Jasper Roe and Jason MacVaugh.
We adapted it into a traffic light colour framework, which became internationally known in K-12 and higher education, and which has just been updated to its current version.
The “Discourse of ChatGPT as cheating tool” led directly to the first blog post, which in turn started a chain of events resulting in a framework for assessment design that, within a week of publishing, had been translated into six languages and visited thousands of times. Discourses influence multimodal texts, but multimodal digital texts influence Discourses.
Design
In my earlier article on multimodal design, I talked about how these technologies have influenced my design of text pre-generative artificial intelligence. A few years back, this blog was mostly for English teachers and occasionally used as an idea testing ground for some of my thoughts on the education sector more broadly. (I also had a thoroughly enjoyable blog back in the early 2010s on sourdough baking. It’s been mothballed, but by all means, go over there and pick up a few recipes after you’ve finished here.) All of those blogs were drafted in much the same way.
Posts started as ideas in a brainstorm. For the most part, the posts were typed directly into the platform and published as is. Multimodal elements such as images, audio, and videos were either imported from stock libraries, Google image search, or embedded from websites like YouTube.
The design process has changed for me dramatically with generative AI. Because of changes in my lifestyle and career, I spend more time on foot walking and running, or in the car driving from regional Southwest Victoria to Melbourne and Adelaide. Verbally drafting blog posts is mentally and physically a different kind of process than sitting behind the keyboard. It’s a return to oracy, and my blog posts are noticeably more narrative these days. Some of this has come through experience, some through personal preference, and some through adaptations to the technology and the way that I write.
When I do sit down to write and draft, either into Word documents, directly into WordPress, with pen and paper, or sometimes into notes on my phone, the style of my written work now more closely mirrors my spoken word than the other way around. Multimodal design has changed. Images are almost exclusively generated, mostly with Adobe Firefly, unless I’m using other platforms for illustrative purposes. Soon, you’ll see increased audio and video content on these blog posts, as those technologies mature.
Production and Distribution
The layers I find truly impossible to separate are production and distribution, and this is not necessarily a generative AI phenomenon. Web 2.0 platforms such as blogs, wikis, and social media have merged production and distribution of text for almost two decades. But generative AI has a role in increasing the rate at which I publish text, as well as, of course, informing the content. Because I can write and distribute from almost anywhere, the car has become an effective space for thinking and catching ideas.
The use of generative AI platforms like Claude, with its ‘Projects’ feature, allows me to store and further work with existing blog posts. For example, I have a Project which contains almost all of the plain text from the 200-something blog posts on artificial intelligence, which can be queried to find quotes on relevant topics, quick responses to frequently asked questions, or create the basis for social media posts.
With these stored blog posts in a core project, for example, I can create a CSV of themed quotes from articles, which can then be imported into a platform like Canva to generate quote images for further distribution. This is not just a flattening of modes, but an emerging and dissolving of boundaries. As I verbally design text, the Discourses I’m absorbed in inform the way that I speak, the language, word choices, and the structure of my orally narrated articles. I write text with production and distribution in mind, Sometimes going from a series of social media posts backwards into a blog article, redistributing ideas or moving between modes of production, from short-form social media posts and back again.
Generative artificial intelligence creates a fluidity between modes and strata that I didn’t have three years ago, even with Web 2.0 technologies. Ever since I’ve had a blog (the bread one), I’ve written combinations of social media and blog posts, designing, producing, and distributing varied texts in tandem. But I did not do so with anywhere near the level of consistency and scale. Sometimes “efficiency and scale” are seen as dirty words for Real Authors, but for any writer there’s something to be said for writing a blog post which reaches thousands of people within a couple of days. You can’t argue with that kind of ego trip.
If not Strata, then What Else?
The next part of the struggle is to seek out new models which are more suitable for exploring multimodal texts with generative AI in mind. I’m not sure what that looks like yet, but it certainly feels messier and more fluid than strata. Discourse, design, production and distribution are still key elements of the process of creating multimodal texts, but there is also something distinct about Generative AI that means it is not like other digital technologies.
While word processing software, home office printers, and blogs have all offered their own affordances and flattened the strata of multimodal texts, GenAI seems to have dropped a cartoon anvil from the window of a four storey building.
Stick with me as I try to figure out what emerges from the squashed mess.
Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch:
Leave a Reply