Behind the scenes of Academy Empirica
How I use AI
I have already written about AI and will undoubtedly do so again as I believe it will fundamentally shape what education looks like in the coming years. Therefore, in this post I am going to explain how I use AI in writing this blog, which in turn will give insight into how I have been using AI in my teaching (which I will explain in more depth in later posts).
Firstly, it is worth saying that everything I write in the main body of the text is my own words, I have not used AI even for drafting any articles. This is not due to any moral aversion to this practice, I think producing a draft that forms the basis of a final product is OK in many circumstances, but part of the point of writing this blog is for me to express my ideas and think through my actions in the classroom. I don’t feel a particular pressure to create perfect content or to write masses of volume, so I prefer to write myself.
Secondly, I think it is very obvious, but also worth asserting that the images I have included in every post so far, are AI generated. As discussed in my previous review of AI developments the aim here is to generate a ‘vibe’ and an image to go with the post, and this is an easy and effective way to do that.
I have also used AI to help with names of posts, asking it to generate a set of options I can choose from in some cases (though not always, the title of this post is all my own). For example:
More substantively, I have used AI to generate some ideas, that complement my own. Referring again to my post on AI developments I used this prompt in ChatGPT4:
“I am writing a blog post on the effects of larger context windows and better video generation on use of Ai in education. Please suggest some ideas what for those effects will be”
I already knew several points I wanted to make, but the response from ChatGPT was insightful and gave me some further ideas that I included in the post, specifically the points on using AI to improve accessibility of videos, a point I hadn’t thought of at the time. Similarly the small point on connecting to augmented and virtual reality was inspired by the AI tip.
So, whilst the final words are always my own, the ideas I am expressing may have been helped by AI.
For some of my longer posts I have provided a bullet-point summary of the main points, and the initial draft of these is provided by AI with final editing done by myself. This sort of summarisation is one of the most effective uses of AI, and I am happy to use this both for speed and for quality purposes.
Most crucially, I use AI as an editor. Most professional writing has had both the author and an editor look at it, this is not the case with a blog, but AI can perform this role well. To make this process more reliable I have created a GPT in the OpenAI app store which I have called ‘Empirica Reviewer’ that enshrines the principles I want to build into the blog. Every draft I write gets put through this reviewer and I then perform final edits before release.
In my very first post I outlined what I wanted to do, so if you read this you won’t be surprised by the five principles that my GPT reviews my work with:
Focus on empirically provable results
It’s in the name after all! I don’t want effective teaching to be chosen based on feelings or anecdotes, I want to know what really works, there is a lot of evidence out there and I want to understand this well and communicate it to you.
Enjoyable teaching sessions are important
Whilst ideally, we pick teaching to be effective in the sense of students actually learning something, if that isn’t the case, either because we are unsure, or there is simply no strong evidence of an effect, then let’s pick the method that is more fun for us and the students.
Presenting answers to specific problems
I want to give practical tips that educators can implement in their own teaching. Too often discussion of pedagogy may leave actual implementation vague or be unattainable. For example, when I discussed lectures I left some tips on what to do with lectures in your course, and I’ll have some more tips at the end of this article, so read on.
Use probabilistic language and show intellectual humility
We are very rarely sure about what works, even when we have strong evidence, we should keep in mind that we might be wrong. Too often people express certainty about their solutions when such certainty is not warranted.
Steelmanning of opposing arguments
In relation to the previous point, when debating opposing ideas, we should endeavour to present the strongest version of those arguments (as opposed to the ‘strawman’). Those with different ideas will almost always have strong ideas in support of their points, and if I want to be convincing then I need to address the strongest opposing arguments (something I tried to do in disagreeing with someone here).
These five principles are embedded in the ‘Empirica Reviewer’ GPT. Having completed a draft, I pass this to the reviewer for feedback and then make additional edits based on it. So far I have found this very useful, in particular it has helped hold me to standards on including evidence, not being too certain, and representing strong versions of opposing arguments, all things which it is all easy to miss in forming your ideas about how the world works.
Let me finish with some practical tips, both for writing and for your students. The latter can be thought of as a preview for later articles I will write on use of Ai in the classroom.
Designing your own editor might be a useful addition for your own writing. Exactly what you want will depend on what sort of writing you are doing, from formal journal articles to more informal blog posts like this one. But if you have a reasonable idea what an editor of such work might look for then you can design either a GPT (or use one that is already available) or design a structured prompt (which is essentially what a GPT is) that will give you feedback on your work. This feedback is unlikely to be perfect but unless you are very lucky in being able to show your work to lots of people and get quick good quality feedback, it is almost certainly useful compared to what you would have without it.
Your students could likely benefit from the same. Do your students have a good idea what you are looking for in the work you have set them? Hopefully so (and if not, why not?) They could then get a review of their work from an LLM (I have written about this for presentations, but the general principle applies to all work). Designing the prompt they can use could be a co-creation exercise that works simultaneously to equip students with the knowledge of what is expected of them, give them insight into using AI, and allow them to do better work.
I use AI for idea generation and would encourage you to do the same, and to get your students to as well. Some tricks with prompts can help here, but basic prompts asking for ideas on a broad topic will get you some interesting results that will likely be useful. The main danger here is to become fixated only on AI generated ideas to the detriment of a wider enquiry.
How do you use AI in your writing and teaching?
Addendum: passing this article through my reviewer, it highlights the lack of empirical evidence and to be more careful in representing counterarguments, you will see more of this in a future follow-up post!
