Over the last few days, I’ve been asked how ChatGPT (particularly allied to Wolfram|Alpha) will affect education, how it relates to “computational literacy for all” and the computer-based mathematics education that my book The Math(s) Fix provides a blueprint for.
Wolfram is involved in Edtech in many other ways too; it will be great to see how the full range of powerful integrations emerge that can deliver better education.
For now, I’d like to put this new technology in context, hopefully, to help to suggest how we might think about changes it should and shouldn’t cause.
Pedagogy v. subject matter
Let’s start by not confusing two potential effects of ChatGPT in education: how the pedagogical process can be improved and what changes in subject matter or curriculum there now ought to be.
Take the latter first. Concentrate for a moment on how ChatGPT might change the real world we’re trying to educate people for before thinking about how it affects the learning of tomorrow’s or today’s subject.
We need to try to understand the point of education for areas impacted by a technological revolution, and that will shift as the real-world consequences occur. Just as high-powered computers have fundamentally changed the real-world subject of maths, ChatGPT promises to change how writing and communications occur, though we don’t yet know how fundamentally.
With maths and computation, we already have decades of real-world change that’s reconfigured our society: with data science, modeling, the start of AI, now encompassing ChatGPT(!). An increasing fraction of decisions are taken, and answers are produced, using the maths or computational process. That’s because computers liberated maths from hand computation, enabling it to be widely and deeply deployed—far beyond what was possible without it. The change is deep and fundamental, arguably the most dramatic mechanization of any field.
What’s the real-world change?
What will ChatGPT’s real-world consequences be? Very hard to tell. One of a raft of AI age innovations or standing out as a central cause of dramatically changing human communications? We’re only a few weeks into mass exposure of ChatGPT’s abilities. But it’s pretty clear that it will affect how written discourse is constructed, whether creating answers to questions, marketing pieces, essays, or even asking data science queries and writing computer code. Those are the first-order effects: evolving what’s currently done by people but using ChatGPT to scale up or assist, or automate. This changes who can perform the activities, and what skills are needed: editing a piece or critiquing one requires different skills to writing from scratch.
“There’s no question that modern computinghas pushed back the boundaries of whatmost considered quintessentiallyhuman thinking skills”— THE MATH(S) FIX
It won’t end there, like it doesn’t for all significant technologies. Because the available format of pure human discourse has been coloured by human restrictions of writing (and reading), with automation, new workflows and new formats will emerge that are only efficient or possible at all with the new technology. In the computational world, for example, machine learning is not a process practically amenable to slow human calculating (the clue’s in its name); it wasn’t deployed through human calculating; it was new because of (powerful) computers, because the bottleneck of human calculating disappeared.
New formats of discourse?
Of written formats, which will live on unchanged, which will die and which will be metamorphosed: books, essays (and computational essays), marketing flyers, scientific papers, contracts (and smart or computational contracts), Q&As, blogs and so on? Next, we need to ask what the human quotient will be in producing today’s or tomorrow’s formats. Even if, say, the essay persists, will it typically be produced automatically from a human starting point or a mixed AI and human endeavour?
Answers to these questions are the starting point for working out how the subject matter of education—the curriculum—should change. We need mass education to prepare our students for the reality of what they as humans will be doing, not (poorly) replicating machinery that works well. That’s optimising results by working with the machinery in tandem. It’s moving up to the next level, not being newly subservient to our latest industrialisation.
In which part of the curriculum are foundations of essay-writing skills supposed to be learnt, for example? In most countries, your language labels that core school subject—English for me. But in fact, it’s much broader than that because reading, writing, critiquing fans out across almost all other subjects, e.g., history, science and so on. (That’s unlike maths, in which only primary school level stuff is used most of the time in any other school subjects: you write essays in history, you don’t do mathematical analysis, though in my view you ought to as it gives you an extra dimension of insight; and you could be doing so, if only you had been learning applicable computer-based maths).
Transmitting subject change to education
Education’s record in echoing real-world subject transformation is diabolical. The ecosystem is stuck, held in place by the lynchpin that is assessment. With maths and computation decades on (as I laid out in The Math(s) Fix in detail), it’s pretty obvious what’s happened in the real world and how divergent maths in education is from this—and yet the subject remains tethered in the 19th century. Much of what I discuss in my book’s section 3 “The Change” and some of section 2 “The Fix” may apply in the future, for ChatGPT and “English” and other affected school subjects, when the real world has reconfigured to take it into account. The archaic way in which curricula are produced, how qualifications are approved and so on.
“The ecosystem is stuck, held in place by the lynchpin of assessment.”
Right now for ChatGPT, we’re ahead of most of these issues because we don’t yet know how the real-world requirements will change. But we should be experimenting in the classroom with the new tech to supplement the curriculum, see where it helps in pedagogy, just like we are outside the classroom.
Is it cheating?
While the curriculum may not transform quickly when needed, the question “Is it cheating to use ChatGPT to do my essay/homework/assessment?” is with us immediately. And my response is “Depends on your homework”. If it tries to use the tech as you would use it in the real world, sets challenges that you, the human, need to learn how to handle, then why would it be cheating? If it ignores that the technology exists, then no doubt it will count as the student cheating, but I’d argue that really it’s cheating our students if we don’t allow them to use the tech and build an understanding and the capabilities they actually need, including some they may learn from the tech itself!
The question sounds very familiar to me because back in 2009, I gave a TEDx talk entitled “Is it cheating to use Wolfram|Alpha for math homework?”. But just because the question is familiar, we shouldn’t jump to all the same conclusions. Chiefly, most people’s real world hasn’t changed yet because of ChatGPT, unlike the decades of change from computation.
Clearly we should be looking to play with and integrate the tech, not ignore it. There are slightly different buckets of capabilities to separate: how to employ today’s tech to achieve what you want with transient details that you must be ready to relearn as the tech changes, and more enduring core skills that get to the essence, the objective, of the subject. (Here’s the start of a section from The Math(s) Fix with a longer discussion).
In either case, there are human capabilities we need to hone and introduce and experience for working with the automation. This is how we step up to the next level.
A central way to get these human capabilities is experience. Experience of as close to real-world actual utilisations of what is imagined as possible. Push the tech just like you would in the real world; find out where it works and where it fails, getting a feel for the best workflows. Definitely do not lock it away and pretend it isn’t there.
Can we use it now as a teaching tool?
Let’s see how ChatGPT can help students write an essay, construct a poem, draft a speech. Let’s set tasks that we consider important but doubt it can do, like critiquing an essay, “being original”, “reading between the lines”, fact-checking (though Wolfram|Alpha can often automate that…). Expose students to what it can’t do, also, not just what it can. But at least some of the time, do expose them to its capabilities. Maybe if the homework is just an essay, after the initial amazement, what we value will shift and ChatGPT “cheating” essays won’t score as well as the best human ones. My guess is that hybrid human-ChatGPT ones will do best; and that itself will be an education on how to produce the best results.
We need to ask, what exactly are we trying to teach with writing essays? Structuring an argument, getting facts, taking a position, making a judgement, project management? Some ChatGPT can help with in the real world (or even educate the human on—like structuring), some it can’t. Some it can uptick, some it may genericise so it initially looks good but then in fact has diminishing utility and no useful originality, and so on.
It’s hard enough to map this out with maths when we have decades of experience in the real world. We know the change is massive, and fundamental. The comparison with ChatGPT is nigh on impossible, so we need to keep humans performing key parts of the process that they may indeed need to continue to do. Let’s accentuate different human skills that we know will need to be sharpened and transformed. For example, I expect even more urgent need for computational literacy to cut through AI write-ups.
AI age educational outcomes
For computational thinking, we found it very helpful to map out modern outcomes we want to achieve from education and see how different they are from those listed for traditional maths. The answer was: almost entirely. What’s surprised me today is how the main headings with “writing” substituting “computation” seem helpful to post-ChatGPT “English”.
This backs up the many comments I have had from readers of The Math(s) Fix who say “it’s not just maths; you’re writing a lot about education in general!” Just as the new ease and power of Wolfram|Alpha’s 2009 release forced a look at maths—though I’d argue not enough—perhaps ChatGPT will do the same for “English” in schools. It’s a step change in itself, but it’s also forcing a fresh look, even in absentio, and exposing whether there’s been a cumulative divergence in subject-matter already.
Here’s the funny thing: I’ve been looking at a few English language curricula and actually, some don’t seem so off base with skills one can imagine still needing in a post-ChatGPT age. Very different to the position with maths curricula and computer-based maths! But as ChatGPT’s consequences emerge, this may change. The key is to keep on comparing the real-world and educational curricula and be ready to change when the world changes. Also key is to keep clear what the essence of the human subject is, its purpose, and separate that from the “mechanics of the moment”, today’s machinery that will evolve, in some cases rapidly, and certainly during a lifetime.
Pedagogy and personalised learning
Let’s turn to pedagogical process. Can ChatGPT transform the “how” we teach as well as “what” we should teach, whether for the unreformed curriculum of today or the needed subject-matter change of tomorrow?
I think it’s an important tool in the toolbox.
“Taking the messy and making sense of it.”
A holy grail in education is “personalised learning”. Rather than a one-size-fits-all of industrialised classrooms, we need personal tutoring for each. Much of this hinges on “understanding” the student so that what’s offered up next optimises their learning path. So far, and bizarrely, doing hand-calculating is a typical place this has been employed. Harder and harder examples of a technique…except of course in many cases that’s exactly the procedural hand-calculating that the computer ought to be doing itself…and it‘s right there, ready to compute! Easy for the computer to understand, easy to codify, but also exactly what the computer does best. Instead, it’s exactly what the computer hasn’t done well that we need to help in guiding the student to understand. Building confidence. Being able to explain. Understanding unstructured student questions. Taking the messy and making sense of it.
Accelerating experience
What’s great about having the computer ask and answer questions, or help, is the volume of experience you can get and instantaneous feedback. Think flight simulators. A simpler case of this for me was spelling correction. I’m naturally a lousy speller, but when good spellcheckers became standard my unaided spelling improved rapidly. Why? Because I confidently used more complex words I’d avoided; when I got them wrong, I instantly got the right spelling in front of me, not a week later; it was in the same font, the picture better imprinted in my head for next time. But this is contrary to much of the educationalist prophecies of the time and wishes to ban spellcheckers in schools.
Get your facts straight
Of course, to parallel this, the personalised learning environment has not only got to seem to understand and react, but get its facts straight too. That’s why the interface with Wolfram|Alpha and other Wolfram tech is so promising in this context. (My brother described and exemplified this power in general, not just for education, in his recent blogpost). Add in the need-to-know maths, science, having data ready for computation and putting ChatGPT with the Wolfram Edtech solution is way beyond anything on offer today for personalised learning.
Summary: Friend or Foe?
Do I have fears about ChatGPT being unleashed in education? I do: that fear of negative effects will entrench the failure to have computers available at all times. Those trying to adhere to a pre-computer age will use ChatGPT as yet another reason to ban computers in assessments and other learning situations. It will drive more of a wedge between reformers and traditionalists when there is no such clear-cut dichotomy.
Overall, I’m optimistic. When we talk of mass literacy, average reading ability is higher than of writing. High quality writing is much more specialist. Could ChatGPT and the like change this? Hopefully so. If everyone can generate as well as consume communication more effectively, they’re better empowered. Ally this to achieving mass computational literacy that’s vital for good decision making at all levels in society and I believe we can enter a much brighter AI age.
Comments