Dr Leon Furze

This post is part of a series exploring how faculty leaders can develop generative artificial intelligence strategies specific to disciplines in K-12 and higher education. The first post gave an overview of the strategic planning process. The next post followed up by encouraging you to attack your assessments. If you haven’t already, you should read those posts before progressing on to this stage of the strategic planning.

The introduction of mainstream generative artificial intelligence at the end of 2022 has left the education sector, from kindergarten through to higher education, in a state of flux. For faculty leaders and executives in education providers, this means that we should be developing strategy which can withstand the next few years.

Of course, we’ve seen in the last 18 months how hard it is to predict the trajectories of these technologies. Nobody, including OpenAI, knew how much of an impact GPT would have on every sector, including education. Nobody could have anticipated Microsoft going all in on generative artificial intelligence or the speed at which it would be possible for developers to turn out multimodal applications like the upcoming Project Astra from Google, or OpenAI’s recently released GPT-4o.

But just because the pace of change is rapid doesn’t mean we can only plan month by month or day by day. We should still aim to strengthen parts of the education system that need to stand the test of time.

There are, of course, outdated parts of this system which generative artificial intelligence (and before that, remote learning) exposed. In the previous post, I encouraged you to sit down with your faculty and attack your assessments, using a sophisticated model such as GPT-4o or Claude 3 Opus to tackle examinations and assessment tasks in your discipline. If you completed that exercise, you’ve probably seen that these models absolutely have the capacity to address previously human-only domains such as logic, mathematical reasoning, and finding meaning from images.

These applications are still far from perfect, but the technology is very capable; it’s time to do something about it.

Spaceships, Bullets, and Cannonballs

Whenever a new technology emerges, there’s a temptation to rush right in and start building on top of it. We’ve seen this already in education with generative artificial intelligence, largely in the form of educational chatbots and tutors which use large language models to create interactive assessments that guide students through curriculum material.

Organisations like Khan Academy have partnered with OpenAI, and we know that millions, if not billions, of dollars have already been spent on fine-tuning these tutor bots. In education, we’ve seen smaller scale variants of this, from departments of education like New South Wales and South Australia developing chatbots in league with Microsoft, to individual schools, universities, and institutions creating chatbots with their internal data.

But I think that we should scale back our approach before worrying about system-wide deployments of generative AI. Of course, the technology moves quickly, and it may feel like there is an imperative to build now. In fact, I’ve written about this before, reflecting on a problem called the wait calculation.

Imagine an interstellar voyage. You can build spaceships with the technologies that we have now that might take, say, 500 years to reach the destination. Or you could wait for 50 years and, with technology advancements, build a ship that will get you there in 100, overtaking the original 2024 spaceship while it’s still chugging through the first part of its journey. Sometimes it is genuinely better to wait and see. Collins’s “fire bullets then cannonballs” analogy speaks to a similar logic.

Imagine ships approaching each other through a dense fog. The captains give the order to fire without being able to see their opponents; they could be shooting at nothing at all.

One captain loads up the cannons, filling them with gunpowder and launching cannonball after cannonball into the ocean. The other captain instructs his crew to load their rifles and fire a barrage of shots in all directions until they hear the telltale clink of bullets on timber. The moment they’ve identified that the shots are landing, he switches to the ship’s cannons.

You can put all of your time and energy into building big, flashy systems that may or may not work. Or you could spend the next few months trialling smaller experiments, and when you have evidence that your approaches are landing, and if and when you have evidence that your approaches have a positive educational outcome, then you can start to fire cannonballs.

I’m making no excuses for mixing up my interstellar travel and naval battle analogies, although I will say that strategic planning is overladen with analogies from war. But this is serious stuff; lives are genuinely at stake. If billions of dollars worldwide are invested into technologies which have been rushed into, deployed too early, and ultimately fall flat, then students will suffer.

If chatbots are rolled out into schools with assurances of safety and guardrails that remove bias, and then these systems are proven to be insecure or faulty, students will suffer.

If schools and universities run headlong into making decisions based on FOMO rather than evidence, then students will suffer.

Which bullets will you fire first?

Having familiarised yourself with local and national policies around generative AI in step one, then taking on your assessments in step two, you’ll now need to decide which small experiments you’re going to carry out on a faculty level that you can measure and get rapid feedback on. If you teach in design and technology, STEM, engineering, or entrepreneurial education, then you’ve got an advantage here, because I’m going to suggest that you run through a design thinking cycle in your next faculty meeting to determine which bullets you’ll fire first.

You might only have an hour in your faculty meeting to go through this process, so I’m also going to suggest that you use generative AI extensively in all stages, but that you always fall back on the expertise of the people in the room. At a very high level, you are going to decide whether your first experiments will be for teachers (the educators), the students, or both.

To an extent, this will be determined by the level of familiarity with the technology in your faculty. If everyone has little to no experience with generative AI, it will be much more productive to try some small experiments where teachers or educators are using the tools for themselves in their day-to-day work before worrying about using it in the classroom. If, on the other hand, you’ve been experimenting with these technologies for a while, it may be time to introduce it to students. Whichever approach you adopt, you will need to create a process leading to measurable outcomes.

Here are the steps you’re following:

1. Quick recap on everything that you’ve learned so far, from the initial documentation review, draft vision, and attacking your assessments.
2. Clarify your audience
3. Design Thinking workshop

You can access a free ebook on the AIAS with over 50 activities for the 5 levels by signing up for the mailing list here:

The Design Thinking Cycle

Empathise: Begin by identifying the audience for this experiment. Is it yourselves, the educators? Is it the students? Or is it both?

What can you do right now to gather evidence of what this audience needs? If you’re focusing on yourselves, you should know most of this already from the previous stages. If you’re a bit further along and have decided to work with students, this would be a good opportunity to invite some of them into the discussion to hear their opinions on generative AI and what they think might be the advantages and disadvantages of using generative AI in your discipline. If you don’t have time to speak directly to students at this stage, don’t worry, because you will do that later. You might choose instead to have the discussion amongst yourselves or to use a chatbot with a prompt like the following:

You are a [e.g., secondary school, university] student studying [discipline]. We are the faculty team and have discussed generative artificial intelligence and its implications for our subject. We want you to identify some of the advantages or disadvantages of using generative artificial intelligence in this discipline from your perspective.

If you’re in a small faculty or if people are insecure about generative AI, you could, of course, tweak this prompt to have the chatbot join in as a faculty member rather than as a student. Neither of these replaces human feedback.

Define: Next, define the exact problem you’re trying to solve. For example, is it increasing staff understanding? Understanding how students interact with chatbots?

Since you’re a group of educators, consider framing this with the language you might use for an objective, such as “to understand” or “to know”.

If you really must, this is an opportunity to get out post-it notes. I personally have a “no post-it” policy when it comes to strategy (and I think I’m allergic to butcher’s paper), but you do whatever works. Try to get as close to the real problem as possible, and remember, small experiments not huge commitments.

Ideate and validate: You’re going to brainstorm as a group for small experiments.

At this stage, it might be good to capture the discussion as an audio recording or take minutes of all of the ideas. If you do use the forbidden butcher’s paper and post-it notes, encourage people to write legibly. Take photographs of those materials to use with an AI model with image recognition like Claude, Copilot, Gemini, or GPT-4o.

Brainstorm for as long as you have the time to generate as many ideas as possible. Consider using a structured approach to the brainstorming, such as a visible thinking routine. This can help to remove some of the fear of the blank page and encourage more meaningful ideas. Use the same processes that you would use with students in your classroom. There’s nothing worse than when I work with strategic planning in education and I see people using brainstorming activities that we know wouldn’t work with students (list as many words as you can that are associated with this school…). You can’t just ask people to pull ideas out of thin air; idea generation often benefits from structure.

Now, I want you to refine those ideas using your generative AI platform of choice. I would recommend Claude or GPT-4o for this stage because they handle a lot of data and do an impressive job of interpreting language and images. If you recorded the discussion, use an application like Otter.ai to transcribe it and upload the transcript to Claude via copy-paste or by uploading the Word document or PDF. Provide the following prompt:

We are a faculty team trying to create some small experiments with generative artificial intelligence that would work in our discipline. These experiments will be carried out with teachers/students [in the subject area]. To help us clarify and extend these ideas, we would like you to carry out the Generate-Sort-Connect-Extend visible thinking routine. <copy/paste initial ideas>

Prototype: Once you have a refined set of ideas from the AI-assisted brainstorming session, it’s time to select a few of the most promising ones to prototype.

These should be small, manageable experiments that you can implement relatively quickly. For example, if you’re in a math department, you might prototype using ChatGPT to generate practice problems for students. Or if you teach writing, you could experiment with using AI to provide initial, low stakes feedback on student drafts. The key is to keep the prototypes small and focused.

Test: With your prototypes ready, the next step is to test them out. This is where you gather data on how well they work in practice. Set clear metrics for success upfront. Are you looking at student engagement, quality of work, time saved for teachers, or something else? Collect both quantitative and qualitative data – survey students and teachers, look at student work, track time spent. Document what works and what doesn’t.

Here’s a suggestion for a feedback survey to use with your faculty:

1. Which AI experiment(s) were you involved with? [Short answer]
2. On a scale of 1-5, how effective do you feel the AI was in enhancing the educational experience? (1 = not at all effective, 5 = extremely effective)
3. What worked well about the use of AI in this experiment? [Long answer]
4. What challenges or concerns did you encounter with the AI? [Long answer]
5. Did you notice any positive or negative impacts on student learning? Please explain. [Long answer]
6. How do you think the use of AI in this experiment could be improved? [Long answer]
7. On a scale of 1-5, how comfortable are you with the idea of integrating this AI application into your regular teaching practice? (1 = very uncomfortable, 5 = very comfortable)
8. What support or resources would you need to effectively use this AI tool in your teaching? [Long answer]
9. Do you have any concerns about the ethical implications of using this AI in education? [Long answer]
10. Any other comments or feedback you’d like to share? [Long answer]

Importantly, start to get feedback from all stakeholders. What do students think of the AI-powered tools? How do teachers feel it impacts their work? Do parents or administrators have concerns? Testing isn’t just about validating your ideas, it’s about learning and iterating. We’ll do more of this in the next stages of the strategic planning when we start to communicate the strategy, evaluate and review.

I regularly work with schools, universities, and faculty teams on developing guidelines and approaches for Generative AI. If you’re interested in talking about consulting and PD, get in touch via the form below: