Leon Furze

Just over 18 months on from the release of ChatGPT, many educators are still struggling with finding genuine uses for generative AI. The technology is ethically complex and has been tied up by an understandable focus in education on academic integrity and “cheating”, making many teachers reluctant to engage.

Despite those barriers, Generative AI has been used to achieve some great things in education. The best work always emerges from teachers and lecturers who are experimenting and finding ways to reduce their workload, make resources more engaging, and help students to understand the ethical and practical implications of AI.

My small role in helping educators is through this blog. In January 2023, at the peak of ChatGPT’s first wave of shock-and-awe hype, I wrote a blog post titled Practical Strategies for ChatGPT in Education which subsequently became one of the most visited pages on the site. I’m updating those original six areas now, but remain focused on helping educators to use the technology in their day-to-day work.

In the previous post I looked at the first three of six areas for more Practical Strategies:

• Designing
• Differentiating
• Engaging
• Imagining
• Editing
• Evaluating

These six new areas cover more ways in which educators can make use of LLMs and multimodal generative AI tools like ChatGPT, Copilot, Claude, and Gemini. In this post, I’ll go through imagining, editing, and evaluating.

Imagining

The creative capabilities of AI have expanded significantly. Recent research suggests that while AI might reduce group novelty, it can support individual creativity. For educators, this means LLMs can be useful tools for bringing abstract concepts to life and generating creative teaching ideas as long as they’re used in ways intended to support and not replace students’ own creativity and curiosity.

I recently had a post on Science Fiction Worldbuilding turned into a resource for the Harvard University metaLAB AI Pedagogy Project, and that series of lessons includes examples of using Generative AI to augment creative and imaginative work.

For example, students are asked to use GenAI as a research tool to ground a fictional world in reality, drawing on real websites containing research data about exoplanets.

Example prompt:

Provide information about the exoplanet XO-7 b, including its physical characteristics and potential for habitability. Use Internet search.

Students then visualise their planets, drawing on the previous information and turning it into an image generation prompt like this:

Detailed, cinematic image, extreme long shot of a planet set against a starry, galactic swirl. The planet is shrouded in purple and blue clouds. The planet has several moons of varying sizes. A gas-giant hangs in the distant background.

They then go on to create websites, character profiles, and simulations based on what they have found so far, for example:

Use code interpreter (or equivalent based on model) to simulate the gravity on my fictional habitable moon, which has 0.8 times Earth’s gravity.

Editing

AI editing tools have become increasingly sophisticated, moving beyond simple grammar checks to understanding context, tone, and even academic writing conventions. While these tools can be valuable for both educators and students, it’s important to use them as aids to develop editing skills, not as replacements for critical thinking about writing.

In our assessment work, we place “AI assisted editing” at Level 3 of our AI Assessment Scale (AIAS): not quite using the tools to their fullest, but still getting quite a lot of support.

LLMs can help identify areas for improvement in writing, suggest alternative phrasings, and even explain the rationale behind editing suggestions. This can be particularly useful for teaching writing skills and streamlining the editing process, or as an educator for editing your own work.

This can range from the very simple:

Analyse this paragraph for clarity and suggest improvements: <Insert paragraph>

To more complex structural or “macro level” edits:

Review this <type of text> at a macro level. Suggest structural improvements for cohesion, readability, logic, and so on.

Finally, you can also use style guides and other editorial documents to refine AI assisted editing even further. For example, if you’re submitting something for publication you might need to adhere to a strict guide. You can upload or copy/paste those guidelines in as part of the prompt:

Use the information here <insert link to style guide or upload> to review this document <attach document>

Multimodal generative AI, including image recognition and generation, now also gives us tools to edit visual work. For example, Adobe’s Firefly – which works as both a standalone product and part of software like Photoshop and Illustrator – has features such as generative fill and generative expand which can be used to edit images and photographs. Generation platforms like DALL-E 3 (via ChatGPT) and Midjourney have also added editing features allowing much more control and refinement over AI generated images.

Since these posts are focused mostly on text-based LLMs (like ChatGPT and Claude) I won’t go any further into image editing; however, it’s worth noting that many platforms like ChatGPT are starting to converge, with most major platforms becoming increasingly multimodal. I’ll write more about image, audio, and video generation in a separate series in the coming weeks.

Evaluating

AI-powered assessment has been a hot topic this year, with edtech companies promising the earth for AI feedback and grading. However, it’s important to note that – in my opinion at least – AI should not be used for grading or high-stakes assessments due to inherent issues with bias and inconsistency. I wrote about this at length in two posts: Don’t use GenAI to grade student work and Racist, Robotic, and Random, so I won’t labour the point here.

Instead of grades and high-stakes marks, GenAI can be a valuable tool for providing more frequent, detailed formative feedback, or for students to self-assess. It can help educators identify patterns in student responses, suggest areas for improvement, and generate tailored resources and lesson ideas based on student work.

For example, GenAI could be used to extend upon brief verbal notes, transcribed with voice-to-text apps.

Refine these transcribed verbal assessment comments for use in feedback: <add transcript>

Those notes can then be used as the basis for additional resources for the student to follow up on themselves, for example by using an internet enabled chatbot to identify extra sources of information:

Using this feedback, create a resource list of links. Use browsing to ensure your links are accurate: <add feedback>

And finally, students can self-assess or evaluate their own work in any way they choose. For me, this is preferable to the educator using the AI for a few reasons. The student can choose which parts of their work (if any) to submit to the AI, and the educator doesn’t need to provide extra feedback which, frankly, the student might not actually read.

Evaluate my response based on the attached criteria and make three suggestions for improvement. <attach criteria and work sample>

More and more practical strategies

Like the original series in 2023, narrowing the vast array of GenAI uses into six areas is somewhat arbitrary. This is a general purpose technology, and I could probably think of a dozen or more extra strategies to add to this list. But the purpose here isn’t to be exhaustive: it’s to provide a few new ideas for ways to explore Generative AI in and outside of the classroom.

If you’ve been following along since day one, you’ve now got 12 areas you could focus on as you develop your AI skills:

• Planning
• Refreshing
• Improvising
• Personalising
• Collaborating
• Communicating
• Designing
• Differentiating
• Engaging
• Imagining
• Editing
• Evaluating

That should be more than enough to get you started!

Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch: