This is part one of a long read reflecting on my PhD journey, which began just before the release of ChatGPT in 2022. The articles look at the ways in which GenAI technologies, and our attitudes towards them, have shifted and changed in the past four years.
In 2022, I made what was simultaneously one of the best and most terrifying decisions of my life. After 15 years in the classroom, I decided that my time teaching was drawing to a close, and that I wanted to pursue one of my other passions: teaching teachers.
Like many people around that time, a catalyst for that decision was the experience of remote learning/teaching. I’d spent months working from home on the farm and in doing so had discovered a lot about myself, including the fact that I much prefer working by myself and for myself. So, in many ways, the decision to leave the classroom in 2022 was a selfish decision. But I also knew that the benefits to my mental health, the time I’d have available to spend with my family, and the flexibility of working for myself would outweigh the challenges. At least, that’s what I hoped when I handed in my notice.
Because I worked in a small regional school and held a leadership position as Director of Teaching and Learning, I handed my notice in months ahead of time. The resignation went in during the term two holidays, effective around nine months later at the end of term four. Plenty of time to think and plan what was going to happen next.
On this blog, I’d already been writing for a few years for my target audience of other English teachers. I held a position on the council of the Victorian Association for the Teaching of English and I’d been running professional development for them, online and at face-to-face state conferences, for a few years since 2016. Around that time, I completed a Master’s in Education at Melbourne University focused on how quality professional learning could mitigate the risks of burnout in teachers. As I transitioned into new roles as Head of English and later Director of Teaching and Learning, I also focused on running PL for other teachers. On the runway to finishing my teaching role in December 2022, I leaned into those skills.
Not long after I’d handed in my resignation, I was chatting on the phone to Associate Professor Lucinda McKnight at Deakin University. Lucinda had also been on the VATE council and she and I had crossed paths at many conferences and professional learnings over the years. Lucinda mentioned that there was a scholarship available for a PhD in digital writing, associated with her DECRA-funded project. But, she added, it would only work if you’re available to do it full time, and I know you’re in the classroom teaching.
Funny you should mention it, I replied. I’ve just handed in my notice and I don’t really have a firm plan about what to do next.
Following that conversation, I spoke to my wife and we agreed that it was unlikely such an opportunity would come up again. Me, out of work and available to continue studies. A scholarship available in a frankly perfect field of study. And so I went back to Lucinda and asked how we might proceed.
We kicked around a few ideas and came to the conclusion that a novel and interesting area to focus on would be the emergent digital writing technologies of large language models. These conversations were happening in early 2022, long before the release of OpenAI’s ChatGPT. But products were already on the market. In my Year 12 English class I had encountered students using paragraph spinners and products like Jasper and Writesonic to paraphrase and rewrite paragraphs that they would then attempt to insert into their coursework and other essays. These applications at the time were built on top of OpenAI’s open-source GPT-1 and GPT-2 models, and they were, frankly, hopeless. But over 2022 they were noticeably improving.
I started to read around the topic and applied for the PhD. Over the next few months I wrote my first articles on what were then known as automated writing technologies, or AWTs. I tried a few myself, comparing AI-written responses to Jane Austen to human student paragraphs. In August 2022, OpenAI released DALL-E, their first public image generator. I experimented with its weirdness, producing images on the blog and writing about the implications for multimodal text.
In those early days, we supposed that these technologies would make their way in increasing volume into the writing classroom over the next three to five years. And reflecting on technologies past, we wondered about the implications of that for writing instruction and for English teachers.
Automated writing technologies were just a small fraction of the AI world, and I also turned my attention to the broader literature on digital and information technologies, reading books like The Information by James Gleick, Vincent Mosco’s The Digital Sublime, and AI-specific books which critiqued the industry, like Kate Crawford’s excellent Atlas of AI.
I’d been noticing these transformer-based technologies emerging for the past few years, particularly when they started to come into my own classroom. But I hadn’t really spent much time thinking about the system they were part of. Applying for the PhD prompted me to get much more involved in the ethics of AI.
At the beginning of November 2022, I received notification that my application had been successful and my PhD could commence. And so began the process of researching, reading in earnest, considering early research questions and avenues of inquiry. The official start date of my PhD was November 15th.
Exactly 15 days later, OpenAI launched ChatGPT.
Shortening Our Horizons
I was already immersed in these technologies and subscribed to all of the right blog posts, so I was experimenting with GPT-3.5 within hours of its release. I remember sitting on the couch, on my phone, putting prompts into the developer sandbox on my OpenAI account, rerunning prompts I’d used previously in the Jane Austen essay blog post and others. The quality had jumped enormously. But what also struck me was the user experience of OpenAI’s newest website. ChatGPT, looking like a Google search clone, took the incredible GPT-3.5 model, re-skinned it as a chatbot, and pushed it out into an unsuspecting world.
I remember speaking to Lucinda and saying that the timelines — three to five years before LLMs were in every classroom — had probably just shortened dramatically.
A week later I was farewelling my final students, many of whom had also caught wind of ChatGPT in their last weeks of school. I chatted to a couple of ex-Year 12s who would be starting university in 2023, and they joked that they’d be using ChatGPT to smash their entire first year course. At least, we all thought it was a joke.
Over drinks at the leadership team’s end-of-year dinner, I sat with the director of development and marketing and tried to explain why I thought this weird chatbot that people were using to turn shopping lists into Shakespearian sonnets were about to impact schools in a way no one had seen before.
It’s easy to look back in retrospect and believe that a lot of this was predictable. After all, the whole reason I’d applied to study “AWTs”Automated Writing Technologies” in the field of digital writing was that I believed they were going to have a profound impact on the English classroom. Others, including Lucinda McKnight, Phillip Dawson and Bradley Robinson, had been raising red flags for a few years about the implications and oncoming risks of the technology in both K-12 and higher education. But realistically, no one was prepared for ChatGPT when OpenAI dropped it on us from a great height — including OpenAI themselves.
Speed and Scale
Australian schools tend to break up in the first or second week of December and then don’t return until the end of January, taking the long summer holidays over the Christmas period. With my PhD underway and a few upcoming professional development sessions lined up in close reading and writing instruction, I spent much of those holidays writing and working.
In January of 2023, I published something like 15 blog posts in a single month. Definitely my most productive month, and definitely indicative of a new special interest coming into play. I am AuDHD, and that combination of neurological traits, for me, means that I can either concentrate on something to the exclusion of all else for hours at a time, or I can’t concentrate on anything for more than a few seconds. The first couple of months writing about AI was a little bit of both.
I was constantly experimenting with ChatGPT and the other emerging technologies of image generation and even very rudimentary audio generation. I was writing frantically about what I thought this meant for education and publishing posters about my early research regarding the ethical concerns of the technology.
At one point I recognised that my writing on the blog was becoming diffuse and frankly, even I was having trouble following my chain of thought. I wanted to put something in front of educators that was useful, knowing that in two or three weeks they’d be returning to the classroom, and whether they’d experienced ChatGPT for themselves yet or not, there was a high likelihood that their students would be using it.
Prior to my AI research I’d already written a couple of books for English teachers — Practical Reading Strategies and Practical Writing Strategies, published by Amba Press. PWS was co-authored by my former colleague and another Head of English and Director of Teaching and Learning, Benjamin White. When I mentioned to him that I felt like my writing was all over the place, he replied, “You’re the practical strategies guy, just do that.”
Great advice. So I did.
I published a blog post called Practical Strategies for ChatGPT in Education. At the time my audience was still mostly English teachers, but the practical strategies were broader than just the English classroom, covering ways that teachers could use the technology in communications, administration, differentiating lessons and curriculum design. It was the first post on my blog to ever go really viral, and to date has been read hundreds of thousands of times. It was so popular that I knew it should become the basis for a book, and it spawned 2024’s Practical AI Strategies. I’m not very imaginative with book titles (Practical AI Strategies 2 is coming soon…).
My blog had shifted almost overnight from English teaching to AI in education. There was a part of me that felt, still feels, the loss of the old blog. Even though it is at the same website address and it exists if you search hard enough, those parts of my blog — articles on the risk of TEEL paragraphs, on close reading strategies for the classroom, reflections on examinations and coursework, and an entire section of resources on Victorian Curriculum English — have faded into the archives. Every now and again, I can tell that some school somewhere is studying a text like Flames, because some combination of the Google search algorithm and sheer luck gives a sudden spike in stats from something that I wrote back in 2019. But for the most part, January 2023 marked the transition away from my blog being about and for just English teachers and more broadly being a place to talk about AI in education.
Breaking New Ground
Aside from the practical strategies post, two other early ideas gained traction quickly: April 2023’s article on a scale for AI in assessments, and a collection of articles I began that month called Teaching AI Ethics.
The first was picked up by Associate Professor Mike Perkins at British University Vietnam, and Mike, plus Jasper Roe, now at University of Durham, and Jason MacVaugh, also at BUV, contacted me to see if I’d be willing to work with them on making the assessment scale more useful for the higher education context. I had no idea at the time that the AI Assessment Scale, or AIAS, would become our most widely adopted contribution to discussions of AI in education. Now located in its forever home of aiassessmentscale.com, the AIAS has been translated into over 30 languages and adopted by hundreds of schools and universities around the world. Mike spoke about it on the stage in Paris at UNESCO’s Digital Week. It’s received awards and it continues to influence discussions of AI in assessment around the world. It started as a blog post. A happy accident which had very little to do with my actual PhD research.
Teaching AI Ethics has also taken on a life of its own and alongside Practical Strategies is the most visited aspect of my website. So much so that it also deserved its own space on the web: https://teachingaiethics.com
I’ve been blogging for almost 20 years. I’m used to throwing words out into the void, not necessarily expecting that anybody’s reading them but feeling compelled to do so anyway. By mid-2023 I found myself in the unlikely position of a PhD student whose topic suddenly had an audience of hundreds of thousands.
My blog and my PhD work often aligned. The Teaching AI Ethics series came directly from my research around the topic, where I had identified nine major concerns in the industry, from bias through to human labour and the distribution of power amongst AI companies. But frequently the blog pulled in different directions to the research, creating tension both in the amount of time I was dedicating to it and in the kinds of things I was writing about. Those early breathless months writing dozens of posts per month gave way to a more reliable cadence of two or three posts a week over the course of 2023, and the tone of my articles shifted as well.
Studying the AI industry at the level of detail required for the PhD led to a lot of cynicism and criticality about the technology. There was no specific moment when this happened, but throughout 2023 I found myself frequently recoiling from articles claiming that AI was going to revolutionise education. When I wrote an article about dead-end chatbots and why I didn’t believe that AI tutors and personalised learning were the future of education, my articles were attacked on a number of fronts, from the techno-optimists convinced by the rhetoric of Silicon Valley CEOs and pundits like Sal Khan, to the audience of exhausted educators desperate for a technological solution to their untenable workload. I suppose the fact that so many people were getting angry about my posts was an indication that they were touching a nerve.
And one of the defining things about this blog over the last few years has been treading that line between the practical strategies and advice on using AI and the often caustic critique of both the AI industry as a whole and specifically its use in education. Posts like Critic, Consumer, Creator sat at the intersection of the PhD and the blog work.
By late 2023 I was just through the confirmation stages, which in Australian PhDs essentially means I’d been given the green light to start the real research. I had decided on a community of practice with fellow secondary school English teachers — specifically, a subset of teachers who were also writers. I put out a call for participants, looking for English teachers willing to explore these emerging technologies through a shared blog and a community of practice-structured professional learning.
We were starting to see government and institutional responses to the technology that had been so disruptive in the classroom for the past 12 months. The Australian Framework for Generative AI in Schools came out in December 2023 and was top of mind for many of the participants who applied to take part in the community of practice at the start of the following year. But as we discussed during the CoP, plenty of educators still felt unsupported, stumbling around in the dark with their personal use of AI, whilst their schools offered little professional development or were still attempting to actively block and ban the technology.
Part two of this article will discuss the PhD process, and how the technology has changed in the years from 2023-2026. Stay tuned for the second part next week.
Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch:
Leave a Reply