Is GenAI in education more of a Blackberry or iPhone?

There's been a rush to incorporate generative AI into every aspect of education, from K-12 to university courses. But is the technology mature enough to support the tools that rely on it?

Image: Midjourney, Dall-E and Photoshop

I try and stay clear of analogies to describe the emergence and impact of artificial intelligence. The technology is so fast-moving and disruptive at the moment that few comparisons come close to doing it justice. But I’m going to go out on a limb this week and explore two possible analogies for AI’s adoption in education. And then I’m going to explain why I don’t particularly like either of them.

The two analogies are the rise and demise of the Blackberry, and the emergence and transformational impact of the iPhone. Both of these are somewhat cliche’d as stories about how different innovation trajectories can lead to very different outcomes. That said, they’re useful — to a degree — in helping make sense of the challenges being faced by educational establishments as they’re gripped by generative AI fever.

The GenAI EdTech Gold Rush

Before November 2022 few schools and universities were investing significantly in generative AI. Then OpenAI made GPT-3 widely available through a very simple and intuitive natural language interface, and almost overnight generative-AI became the next must-have educational tech.

I exaggerate a little of course. We first had to get through the challenge of students cottoning on faster than their instructors as to what ChatGPT could do, leaving universities and schools scrambling to navigate a potential threat to how they’d taught for decades.

Over the past year though we’ve seen generative AI flip from a possible threat to a potential educational game changer. Mirroring this, the rate with which AI-enabled educational products have arisen has been truly astounding.

The result is a growing number of apps, tools, and resources, that claim to revolutionize what we teach, how we teach, and how we maximize our time as teachers.

For anyone in education technology, this is is a pivotal moment. It’s also one that I’m behind — as long as we proceed with eyes wide open and a good dose of critical thinking. And as part of this, we need to be deeply aware of the risks of early and naive adoption, as well as potentially limiting tech lock-in.

As anyone who regularly uses generative AI will tell you, these technologies are still evolving. And because of this, it is very hard at this point to tell which provide a solid foundation for long term apps; and which do not.

And this is where the Blackberry versus iPhone analogy comes in.

Blackberry vs iPhone

The Blackberry was a great product. But it was niche. And it was inflexible.

The Blackberry was vulnerable to new technologies overtaking it and killing the market for a phone that did email — and not a lot more. In other words, it had a limited shelf-life that placed enterprises that relied too much on it in a tough position.

In contrast, the iPhone heralded the emergence of a transformative concept that was far less dependent on the underlying technology.

While the original iPhone is long gone, it ushered in an era of smart phones that are as versatile as the imagination of those creating new uses for them. And while specific devices may have built in obsolescence, the concept of the smart phone continues to evolve and thrive.

Applying this to the current wave of generative AI, are educational establishments investing in a time-limited “Blackberry,” or an “iPhone” of a concept that they can continue to build on with confidence?

The analogy is a flawed one — I’ll get to that in a minute. But it’s useful in highlighting the challenges of potentially backing the wrong horse in educational technology (to gratuitously mix up metaphors) when the horse may turn out to be a one trick wonder with a limited shelf life.

The problem is that the current crop of natural language interface-based generative AI systems — including apps like OpenAI’s ChatGPT, Anthropic’s Clause, Google’s Gemini, and Meta’s Llama — represent capabilities that are changing rapidly, that are largely untested in critical applications, that have limitations that are only just becoming apparent, and that are likely to be far superseded in the future.

This does not mean that these don’t represent transformative technologies, nor that that they should not be used in education — far from it. But it does raise concerns where they are being used as the basis for education-based applications that are expected to be capable, trustworthy, and long lived.

Here, the Blackberry scenario is one where there is massive time and resource investment in educational technologies that rapidly become outmoded, and that are so constrained in what they are reliably able to achieve that they end up limiting rather than opening up educational opportunities.

It’s scenario that worries me — a lot. There are growing indications that large language models may not be able to achieve as much as they initially promised — and that making them larger won’t necessarily make them better. There are also fears that capabilities which have superficially impressed many over the past 18 months may not end up leading to reliable tools in some cases.

To make matters more complex, these models are constantly being tweaked — including the fine tuning and guardrails that determine how they will respond to prompts.

This is where things begin to diverge from a neat analogy with the Blackberry, but technological constancy is important if an educational tool is to continue to do what it’s supposed to. Having a personal tutor for instance that provides different advice from one week to the next, or a syllabus builder or AI grader that is constantly changing its behavior, isn’t ideal.

The risk is that, in the rush to embrace generative AI, educational establishments will make substantial investments just to be left holding an obsolete and unreliable set of AI tools that they are, nevertheless, committed to.

But there is another possibility — and this is where the iPhone comes in.

Concepts vs Products

Despite my concerns around Blackberry-like scenarios, it may be the case that emerging AI-based educational tools end up being resilient to changes in the underlying technology.

This iPhone-like scenario will require educational apps that are resilient to constant tweaking of the underlying generative AI technologies they’re built on, and sufficiently agile to adapt to substantial changes in these technologies — all while providing educators with consistent and trustworthy support.

It is, I must confess, hard for me to imagine how this will work. If, for instance, you’re sold on the idea of an AI autograder that you can deploy with confidence now, you’re going to feel a little foolish when the next iteration claims to be even better (grading not being something where sometimes being right is OK). Or if you’ve been using an AI-based personal learning tool with your students, discovering that the next iteration is twice as good is likely to make you wonder just how bad the previous version was — and whether you should have trusted it at all.

These and other scenarios become a problem if tool developers and educators adopt a Blackberry mindset that assumes we have hit peak generative AI, and the AI landscape isn’t likely to change a lot over the coming years.

But of course it is.

So what does an effective iPhone scenario look like? My sense is that will depend on a much more cautious and experimental adoption of AI in education — one where educators are able to leverage current capabilities, but rapidly pivot as new possibilities emerge.

This is likely to mean relying less on enterprise-level deployment of AI tools which are inflexible and, as a result, fragile, and more on creative uses of general purpose AI technologies.

It’s also going to mean continuing to rely on established ways of teaching while augmenting these with AI in ways that, if everything goes pear shaped, you still have a solid foundation to rely in.

And I suspect it will mean allowing students more flexibility to explore and experiment with AI platforms that they find helpful, rather than hard wiring AI into the educational system — just to discover that the system needs to be rewired every couple of months.

In other words, it’s going to mean investing in concepts, not products.

This, to me, is at the heart of an “iPhone mindset” as opposed to a “Blackberry mindset” when it comes to AI in education — an approach that avoids hard wiring in constantly changing technologies, and that builds experimentation and innovation into the very DNA of learning.

Dubious Analogies

As I said at the outset of this article, I don’t particularly like either of these analogies — the Blackberry or the iPhone. They are certainly useful in beginning to tease apart the complexities and challenges of rapidly embracing AI in education. But like all analogies, they come with a lot of caveats.

The emergence of AI as a general purpose technology — generative AI in particular, but AI foundation models more generally — has attracted many comparisons. The invention of of the calculator, the emergence of the internet, even the transformations brought about by the printing press or the industrial revolution.

All are attempts to understand the advanced technology transition we’re experiencing in the context of what we’ve previously encountered. And all fail to capture the sheer uniqueness and profundity of how AI is changing our world.

And in the grand scheme of analogies, the Blackberry and iPhone are pretty low down the list of good comparisons.

One aspect of generative AI that neither captures well is just how untrustworthy current platforms are — not in the sense that they are deceitful or misleading, but we’re still getting the measure of what they can and cannot do well as they evolve.

For instance, they ways that ChatGPT responds to prompts today is different to how it will have responded few months ago as OpenAI continue to develop and refine the technology.

By all accounts we’ll see substantial differences in responses when GPT-5 is released. And a year from now we’re likely to see emerging generative AI architectures and platforms that dramatically change how the technology behaves and what it is capable of.

We’re already seeing this with the emergence of on-device AI and advanced AI assistants.

This constantly changing AI landscape is certainly more aligned with the evolution of smart phones than the Blackberry — but even the iPhone analogy fails to capture the speed at which it’s shifting and morphing. And to think that we can build robust learning environments around something that is likely to look like something else in a few months seems precarious to me — especially when what’s at stake is the long term success of our students.

An iPhone Mindset

For all my concerns here though, maybe there is something to being inspired by the Blackberry/iPhone analogy — not as a playbook for developing and using AI in education, but as a mindset that embraces innovation while avoiding becoming locked in to apps that are detrimentally unreliable and that ultimately lead to dead ends.

And as a mindset, I’m more comfortable with the iPhone as an analogy of a transformational concept, than the Blackberry as a model of a brittle tech implementation.

But that’s just me — I guess time will tell!