This post is the second in a series of Q&A interviews with educators working with generative AI. These posts will explore K-12 and tertiary perspectives of teachers, academics, and professionals who are grappling with the implications of these new technologies.
Steve Brophy is Director of Digital Transformation at Ivanhoe Grammar School in Melbourne. He and I have connected a few times over the past twelve months in AI related work, including these short modules for Digital Learning and Teaching Victoria that we made way back at the start of 2023.
Most recently, I saw Steve present at the Melbourne Informa AI in Education conference, where Steve spoke about a unit of work created to explore generative AI with year 9 students.
Here’s the interview with Steve:
What inspired you to use generative AI for speculative fiction, and what were your initial objectives or learning outcomes you hoped to achieve with your students?
I teach a unit in my Year 9 Transformative Technology class called Brave New Worlds. Inspired by Aldous Huxley, students explore the world of emerging technologies and their ethical and social implications. Developed as part of Dr Sophie Fenton’s thesis research, Human-centred Learning: A transformative pedagogy for the cyber-physical world, this unit is designed to tap into the power of imagination to explore futures that have not yet occurred. One of the challenges with such conversations and explorations is what I call dystopian deflation.
Holding heavy conversations about climate change, deepfake technology and rise of AI can leave students (and adults too) feeling hopeless and helpless about the future. To combat this, I lent on the work of Dr. Jane McGonigal and the Institute for the Future and their mission to bring futures thinking to the world. Scenario planning is a methodology used in the futures thinking space. Fictional scenarios allow us to explore many futures and to decide if they are futures we want to move towards.
So inspired by speculative fiction authors such as William Gibson, Octavia Butler and Arthur C. Clarke, I had my students create Brave New Worlds to explore. Since the goal of the project was not story originality but world creation inspired by emerging technologies, students were able to use generative AI to co-create. As part of the unit, we explored with great depth the social and ethical implications of generative AI platforms. Speculative fiction also allowed students to make peace with the dystopia. Just because the future is possible does not mean it is inevitable. It is but one future and there are many available. It helped develop what the Institute for the Future calls “Urgent Optimism”, a mix of hope and agency.
Could you provide an overview of the structure of the specific task and how you integrated generative AI? Specifically, how did the students interact with the different platforms, and what role did AI play in their storytelling process?
Students were tasked with creating a speculative fiction story about the fictional world of Eohnavi (Ivanhoe backwards). Students used a futures thinking method called Signal collection to find fringe technological advancements or applications that present a new possible world. Students used websites such as Quantum Run and Futurism to collect five signals from a broad range of categories that spoke to them. These signals were then to be infused into their speculative fiction piece. In previous iterations of this task, students only used ChatGPT (once I gained parental approval to do so). I found the results bland. So, to broaden their aperture and to raise the bar of standards for AI generated content, students had to create their story using a minimum of four Large Language Models (LLMs).
ChatGPT, Bing and via Poe, Claude, Llama and PaLM were the major platforms used. We had intended to use Bard but age restrictions prevented this. How the students used these models was completely up to them. Some used the same prompts across each model and chose the best story generated. Others distributed sections such as the introduction and conclusion to the different models. Others used the different LLMs for different tasks such as planning, writing, editing, sentiment analysis, bias analysis, etc…
Once students had their story, they had to generate images across four different image generation platforms (Bing Image Creator, Playground AI, Stable Diffusion, Dreamstudio). The images needed to have an aesthetic that connected them. As well as supporting images, students needed to also use RunwayML to create supporting videos. Two four second videos were required.
All prompts and generated output were collated in their OneNote notebook and a requirement of the task submission. Once the story was completed, students put the content into a Microsoft Sway. The final part of the task was for students to engage in an ethical discussion with Pi.ai about whether their AI-generated speculative fiction piece could be considered art. Students needed to include the exported link of this interaction in their final submission.
Can you share some insights into how the students responded to using AI as a tool for creative writing? Were there any notable shifts in their engagement or perceptions about literature and technology?
There was trepidation at the beginning, which had more to do with the “AI is cheating” narrative than anything else. As a class we had plenty of robust discussions about what constitutes cheating and I really enjoyed demystifying the myths in this space. Being able to speak freely and openly certainly helped expose their questions, perceptions, and misconceptions. Watching how students engage with AI provided tremendous insight about their motives for using AI outside of this project.
Students found many limitations and potholes. Those who wanted to get to the finished product quickly found that the output of their chosen LLM wavered in quality and ability to stick to the task. Smaller prompts (aka smaller tokens) allowed the output quality to increase. Students also found the hallucinations frustrating. “It keeps making mistakes, even I gave it the right information.” Engaging with the platforms over an extended time, instead of transactionally, exposed this fallibility.
Some of the students loved their stories, others felt they were average. For the reluctant writers in the room, the ease of regeneration helped lift the standard and provided them support to deliver an engaging story. Students also noticed a role shift. Since the writing of their story was not their responsibility, the editing became a more prominent feature. From an output perspective, Bing (aka GPT4), once it was jail broken, produced the best results according to students.
Personally, the most interesting element of the whole assessment was the student discourse with Pi. While I find Pi to lack a little bite, it does facilitate a rich chat experience. Reading student responses to Pi’s questions and answers provided a fascinating insight into student ideology, understanding, bias and perception. The medium also enabled the quieter students in my class to express themselves. In the past, I would have facilitated an ethical discussion in class, drawing upon a small pool of students willing to share their views or questions. With Pi, I was able to tap into the wisdom of everyone in the room by providing them a psychologically safe environment in which to do.
The combination of multimodal content and speculative fiction is particularly interesting. How did students use AI to integrate various modes of communication (text, sound, imagery) into their storytelling, and what were some of the challenges and successes they encountered? [if you’re happy to and have student permission, a couple of images or examples would be great here]
Having to work across multiple AI models required students to be more explicit about their creative vision and aesthetic. Cyperpunk, solar punk, Black Mirror, neon, fantasy, no matter what the direction, students needed to tie the multiple creations together under one or two umbrellas. For those who were less sure about the direction, the creation of images usually sparked a theme. From here, those images were used in RunwayML or Dreamstudio as prompt inspiration. This was a little more complex for those who wanted to feature repeat characters throughout the story. Some students used a generic level of prompting to circumnavigate this, i.e. woman with long brown hair in white suit. Coupling this with their design aesthetic seemed to create the illusion of character continuity. For others, they chose to have the character looking away.
When creating the video elements of RunwayML, most students used their AI generated images as the base image to move. For those who wanted a certain movement, the video-to-video generation feature of RunwayML was used. This feature was particularly popular amongst the students but the four second duration of the free account was a limiting factor.
The image and video generation element of the assessment really brought the story to life and provided a real insight into the interests, tastes, and creative vision of the students.
Examples (edited to protect student privacy)
Looking to the future, how do you see generative AI evolving in education, especially in areas like creative and critical thinking? Are there any ethical considerations or pedagogical strategies that you think are important to address as this technology becomes more pervasive?
Human ingenuity plus the power of AI to solve wicked problems seems like a pretty good starting place. The bar can’t simply be the generation of an average 1000-word essay. My bias is towards the creation of rich learning artefacts where humans work in symbiosis with generative AI platforms. Great artefacts need vision, tireless amounts of feedback and mechanisms to test their effectiveness. Rich thinking in partnership with generative AI has plenty of potential, we just must get the balance right.
Broadening the ethical aperture of our students is the most important element in my mind. They are the future developers. We must move past generative AI for corporate greed and demand generative AI for the greater good. Syntropic AI, not entropic AI.
If you’d like to get in touch to discuss Generative AI or you have work you’d like to share with the community, please contact me via the form below:
Leave a Reply