How AI Works for Education: An Interview with Jon Dron

 

 

 

In his 2023 book ‘How Education Works’ Jon Dron proposes a new lens to understand educational technology whereby individuals are not just users but co-participants in technologies. If the book had a theme song it would be ‘Tain’t What You Do (It’s the Way That You Do It),’ composed by jazz musicians Melvin Oliver and James Young.‘A paintbrush is as much a technology as a manufacturing plant, and teaching is far more akin to painting than it is to manufacturing, though it shares many common features with both. Teaching can be thought of as the application of tools, methods, principles, techniques, and structures to help people learn, and we all do it, whether to ourselves or others. It never happens the same way twice, and the ways in which we might respond to it are more numerous and various than the ways in which we might respond to a painting.  […] What we do (the tools, methods, principles, etc. for doing it) is far less significant than the way that we do it (the technique)’. (Dron, 2023, p.3)The first chapter is titled “A Handful of Anecdotes about Elephants” and it argues that education is full of paradoxes and contradictions that are often ignored or rationalized by educators. He illustrate this point by sharing several anecdotes from his own experience as a learner and a teacher. One of the stories will resonate deeply with members of the AACE community: It is about an e-learning conference that was disrupted by an earthquake that cut off the power supply – E-Learn 2006. Most presenters adapted to the situation by engaging in conversations with the audience, resulting in a better learning experience than if they had used their prepared slides. However, he recalls one presenter who insisted on showing his slides on his laptop screen, which was barely visible and audible to the audience. In this case, the technology, as much as the habituation towards it, resulted in a missed opportunity for connection. As Jon pointed out, he learned a great deal: it just wasn’t what the presenter wanted him to learn! The book is full of twists and gems such as this.

Jon has applied the ideas laid out in his work to generative AI. A self-described geek and technology-enthusiast, Jon has a playful, curious and exploratory attitude towards AI. Nevertheless, he warn of specific dangers to the endeavor of education, among them:

Coarse-grain cognition refers to a way of thinking or processing information that focuses on broader patterns, general principles, or overarching structures, rather than fine details or specific nuances. This concept often contrasts with fine-grain cognition, which emphasizes detailed, precise, and localized analysis. Coarse-grain cognition emphasizes abstraction by focusing on high-level concepts or summaries while ignoring fine-grain details that are not immediately relevant to the broader understanding. The problem is that our coarse abilities may require and depend upon honing more fine-grain skills.

Collective Model Collapse refers to a phenomenon in machine learning and artificial intelligence where the iterative training and retraining of models on outputs generated by other models leads to a degradation in the overall quality, diversity, or reliability of the generated content over time. This can happen in systems where AI models rely heavily on other AI-generated outputs as part of their training datasets.

Jon uses the term “stochastic parrots” (Bender et al. 2021) to point out that large language models (LLMs) generate outputs based on statistical probabilities rather than any genuine understanding, consciousness, or intent. The phrase underscores that there is no “they” there—these models do not possess agency, cognition, or intentionality.

Generative AI can teach the explicit curriculum, but what impact does it have on the tacit, implicit curriculum? Jon advocates for strategies such as decoupling assessment and learning, modeling being human, encouraging failure and risk-taking, and valuing process over product to reclaim education as a deeply human act.

In the interview, we talk about the process of writing with and without AI, the regulation of AI and social media, Jon’s dislike for learning management systems, and the pivotal moments that shape our thinking.

Among all the learning technologies that one could dislike in the realm of edtech you often criticize learning management systems. I recently learned that you also used to develop them. Explain why you are disenchanted with this specific infrastructure that is so deeply embedded in our distance (and on-ground) learning practices at universities.

Many, many reasons! I’ve written many papers and chapters on this and provided a lot of arguments against them but I’ll try to abstract some of the key core issues. LMSs were designed by people, like me, who treated their design as an information systems problem. We looked at the things that happened and entities that existed in (mainly) existing universities and turned them into software, without thinking about what was changed, what was lost, or what was gained when we did so. The thing is, universities are technologies, made of countless technologies, all of which are solutions to problems, most of which are caused by the technologies used to solve other problems, most of which relate to the problems faced by Mediaeval scholars wanting to spread doctrine in as efficient a way as possible, given the very limited technologies available to them at the time. When we modelled this, we tried to embed all the absurd number of unnecessary counter technologies into them, from punitive grading and teacher control to lectures. But we made it worse, because our models were simplified caricatures. For instance, we chose to think of lectures as means of imparting information so we provided tools for creating, editing, and displaying content without stopping to notice that information provision is in fact one of the least valuable roles of a lecture, and one that they do extremely badly. Lectures are social events that provide permission and purpose to learning. The intentional and often difficult act of physically attending them brings salience, things happen before and after, conversations (tacit and explicit) often occur, lecturers react, communicate ways of thinking, attitudes, values, and so on. Similarly,  we noticed that some modes of teaching involved discussion so we made discussion forums but, because we treated discussion as a function, we separated it almost entirely from its context, forgetting that discussions take place while we do things, surrounded by objects that anchor them, in places that matter. How weird would it be if, in a traditional classroom, if you wanted to talk about something then everyone got up and moved to another room where essentially nothing could happen apart from discussion? We transferred metaphors of space into systems that completely undermined such concepts, and took very little advantage of the adjacent possibles that online systems supported.  We did dumb things like completely hide the contents of courses from those not taking them, or creating rigid roles that could never be fluid as they are in real life. It is particularly bizarre that you can be visiting a page in an LMS with thousands of others but have no awareness that they are even there. I could go on for hours on this. The central issue, though, was that we made a machine to simulate another machine that had been tinkered together over centuries to solve an almost completely different set of problems. But it gets worse. For the sake of efficiency, we centralized all of this so that everyone got the same set of tools and the same “environment” (an over-grandiose term – learners occupy environments of which LMSs are very minor components, they are not environments in their own right). Virtually none of those tools are great in the first place and, when they are configured to support the needs of everyone, they tend towards the lowest common denominator. So we wound up with rubbish, poorly tailored tools that badly replicate a broken system that was designed to solve different problems, hardening many parts that only work at all because of their inherent softness, failing to take advantage of the adjacent possibles that online systems afford. Yes, I hate them.

You have spent a decade on writing your latest book. How has AI changed your own writing and research? Is ‘How Education Works’ the last professional oeuvre you will ever write completely without any AI-input?

I have written nothing without AI input for at least 25 years. Every time you run a search in a search engine, or visit a social medium, or use autocorrect you are using AI. AIs have participated in our cognition for a long time. Equally, we are very used to the coarse grain cognition that results – the machine-mediated snippets that the likes of Facebook or X feed to us, though (until recently) mostly created by actual people, bombard us with little chunks of “knowledge” that replace the need for us to think, because someone else has done the thinking for us. That’s not terrible in itself: our ability to do that is the ratchet that lifts us ever further. We stand not only on the shoulders of giants but on the shoulders of everyone we know, however slightly, every technology in which we participate, every word we hear or read uttered. Our intelligence is almost entirely collective: through technologies we participate in the collective intelligence of our species. We are part technology (think of the words, the concepts, the models, the theories, the processes, the procedures, the methods, the techniques that are entirely intracranial) and technologies are part-us, because the ways we use them are technologies too: as they are enacted in the world we are participants in them, not just users. Generative AI is fascinating because it can participate in other technologies too: it is among the first technologies ever created that can creatively assemble other technologies in a very human-like way. The potential is huge and I can and do make extensive use of it: it is now a part of my cognition. I don’t think it is very likely that I will ever use it to actually write any of the words I will write in future (I wrote a chapter about the reasons for this a little while ago at https://books.openbookpublishers.com/10.11647/obp.0356/ch3.xhtml),  though it will certainly create images, perform analyses, help me develop ideas,  write software, and act as a partner in my research team of one in many other ways.

In my experience, many creative people thoroughly dislike AI tools for lowering the expectations to an acceptable average for the things that they themselves do very well – and, at the same time, use AI with enthusiasm for things they cannot do and delight in its output. I am curious, as a musician, do you find AI-tools like suno or udio particularly jarring or do you think these are fun?

My musical career in the 1980s and early 1990s was significantly shaped by my utter hatred of rhythm boxes. Almost every other solo musician doing the same circuits as me used the things because they worked: the fuller sound and extra texture was a big crowd-pleaser, making it possible for solo musicians to do gigs that only bands could manage in the past, and it soon became an expectation that anyone doing this kind of work had to have one. I hated them because they were inhuman and inexpressive: human percussionists are great because they listen and, like a heartbeat, their rhythms are organic and chaotic when you examine them closely. Though I could and still can play and sing many other styles, by choosing to sing swing (with a touch of blues and a hint of jazz) I found a genre that was very popular (many of the songs were well known), that happened to work well for my voice, that did not come with an expectation of sounding like someone else’s recording and, above all, that did not need a rhythm box.

My feelings of loathing towards autotune are even more profound.  However, this does not mean that great art cannot be made with both autotune and rhythm boxes: there are many songs I like a lot that have used both. And it doesn’t mean that I won’t use a rhythm box when I am practicing or playing for fun. Likewise, I am very drawn to band-in-a-box uses of AI that listen to what you are playing and create a real-time accompaniment. Existing approaches have been few and far between and have never quite worked but, with the latest generation of generative systems I predict we will start to see a lot of them. I’d certainly enjoy using one though, again, mainly for practice and fun, and probably not for performances. It depends, though, on how well they listen. I do sometimes use, for example, a harmonizer, because it is entirely dependent on my own voice. If it were perfect in its harmonies I might feel less well-disposed but, in fact, it takes a fair bit of skill to make the harmonies work as intended so I think of it as being like another instrument, still very human and capable of expressing a lot. I reckon the same could be true of automatic bands.

I find generated songs to be quite interesting and sometimes very funny, especially when they convert one genre to another. I have had swing versions of reggae, pop, and punk numbers in my set for several decades, mainly as novelty numbers, and I think it is quite OK to be playful without caring too much about the details of the performance: humour is a different art form.

Wholly original works by AIs are another matter. So far they are very average, at best, but they are getting “better” in the sense that it is getting harder to spot that they are generated by a machine. I don’t love that, for the same reasons that I do not love the idea that children’s books are being written and narrated by machines. Children’s books are how we learn ways of being, not facts about the world. It’s the same for music: it is an expression of our humanity, and part of how we learn to be human. Even rhythm boxes and autotune are the direct result of someone programming a machine to behave in a certain way. Generative AIs, though, are not human, despite nearly 100% of their output deriving from humans. That fraction of a percent that they add themselves, though, makes a difference. The more of it we encounter, the more it will change us and, because generative AIs are trained on what we produce (and, increasingly, what other AIs produce), those changes will be magnified over generations. It will happen: I think that is inevitable. That’s why, right now, we need to be very mindful about what we want to preserve of our humanity, and what we don’t mind losing. Now is the time to launch the resistance, if resistance is needed, because, in a few years, we won’t even notice it has happened.

Will AI change how we appraise originality in art and culture and, ultimately, what counts as truth?  How will we evaluate the information that we encounter online in Web spaces filled with AI-saturated content?

We’re already there. By some estimates, the majority of content created over the last year or two has been generated by AI, including very intentional attacks on nations and cultures by those with an interest and the power to swamp our systems but, at least as much, by algorithms that feed social media and people who know how to play them well. I have no doubt that it has played a significant role in some of the polarization and shift to the right we have seen in politics around the world, though it would be beyond difficult to quantify.  I do think there are many ways that we can be more creative, in new and interesting ways, with the help of generative AIs. Artists using creative “partners” to actually produce art works has a very long tradition, dating at least as far as the Renaissance, through found-objects like Duchamp’s Fountain, through to artists who openly presented the work of others (under their direction) as their own like Andy Warhol or Mark Kostabi. It is the norm in music, sometimes taken to extremes like Milli Vanilli or the Monkees, but even the Beatles used uncredited session musicians in some of their music. And, of course, video generation systems like the recently released Sora or Veo2 stand a good chance of revolutionizing the video and movie industries by putting creative tools into the hands of very many people who would not have had a chance before. I think this is particularly interesting because such things are already the products of often vast teams. Generative AI makes it possible for anyone (with access and funds – the digital divide remains a bit issue) to become a director, a conductor, or an editor. That’s cool, at least as long as such things are empowering rather than simply replacing humans.

Jaron Lanier argues that we need ‘the right comic strip in our heads’ for understanding generative AI –a rough, but basically correct mental model of how AI works, instead of perceiving the technology as a magical, human-like counterpart. Do you think this is achievable and how? What should this entail?

It is moving too fast for comic strips. Yes, it would be good to have such models, but I think the best we can currently achieve is a good metaphor. I like Dave Cormier’s “autotune for knowledge”, Ted Chiang’s “fuzzy jpeg of the web”, for instance, or “drunken RA” (that I heard from Punya Mishra).  “Stochastic parrot” is not bad. A useful way of visualizing the scale of the data is to imagine everyone on the planet writing a book with hundreds of pages, and of the AIs as being rather dumb librarians with incredibly good and unbelievably fast indexing systems that can pick any part of a word from all of them, but I think that’s about as far as I would want to go. The details are changing almost daily, and there’s a huge amount of mashing up going on. 

Your book (together with Terry Anderson) ‘Teaching Crowds: Learning and Social Media’ explores the intersection of online learning and social media, focusing on how collaborative technologies can enhance education. How do you view social media today – friend or foe?

T’ain’t what you do, it’s the way that you do it. Even back then, I thought of many social media as foe. I have had a vitriolic hatred of Facebook since the first year of its inception, because Zuckerberg broke ethical boundaries from the very start, ruthlessly exploiting the dynamics of social systems in ways that no one else was willing to consider, breaking the social web, and trying (in some areas with great success) to break the Internet itself in the pursuit of “engagement” (and hence profit). Others tried it, but Facebook did it single-mindedly and, ignoring the immorality, did it horribly well. At the other extreme, there’s Wikipedia. Most other social media occupy a space in between. Currently, I feel pretty good about Bluesky and Mastodon, and it is important to remember that over a third of all websites are running WordPress, nearly all of them in social ways. And let’s not forget Amazon, email, Zoom, and so on: “social media” is a very broad category of system with goodies and baddies and many mixes of the two. Going back to my PhD work in the late 1990s (which involved building social media for learning that harnessed the crowd and self-organizing processes for educational purposes) I learned that you have to look at the whole system and its relationships with other systems. We have a tendency to ask silly questions like “are social media bad” or “are computers good for learning” and attempt to answer them in the general rather than the specific. The specifics always matter. I do think that it is terrible that we took the Facebook-driven path of creating monolithic behemoths beholden only to shareholders, but that’s more a criticism of capitalism than of social media: social media brought out the worst in capitalism. Until about 2007 almost everyone (apart from Facebook) was moving towards a federated or distributed model, but we have had a really bad centralized blip over the past 15 or so years that we are only now starting to recover from. If Musk had not taken over and destroyed Twitter, that used to serve a vital role as an integrative nexus rather than a medium in its own right and so served as a buffer to hold other social media together, we would probably not be moving so strongly back to the federated, distributed, model now. Though there are some big obstacles and dangers to overcome, I really hope we continue down this path: open standards, empowered humans, controllable identities, etc are critical to our future.

In your recent talk The collective ochlotecture of large language models, you describe social media as a significant AI intervention that has already influenced cognitive practices. To what extend do you think today’s social media landscape is like the tobacco industry of the 1960s?

The huge monoliths are much like the tobacco industry of the 60s, and big oil since at least the 1950s, continuing to this day. All of them know the harm they cause, all rely on addiction, all rely on the fact that the perceived personal pain of stopping use of any of them tends to outweigh the perceived benefits of doing so, and all are so deeply embedded in broader systems and norms that it is hard to see how they can be eliminated without engendering major economic and social collapse. As I said before, the problem is really that companies want to survive and are obliged to make short term profits for their shareholders as their primary moral and legal duty. It is not that they are unethical: it’s that their ethical systems are evil, in the sense of being antithetical to the survival and happiness of people and, indeed, of the planet. Yes, the likes of Facebook (in particular) are fully aware of the harm they cause and spread as much doubt and misinformation about it as they can because, if there were unequivocal proof, they might not survive as a profitable company, which is their one overriding moral imperative. It has ever been so. It’s just that the companies have far greater reach and power than even the biggest monopolies of the 20^th Century. Unlike the tobacco and oil industries, though, it does not have to be that way. There are plenty of examples of very positive benefits and very beneficial ways of using them – even Facebook has been good at times for nearly all of its victims, and very good for some of them. Again, it ain’t what you do, it’s the way that you do it. We know, for instance, that people who actively post, who only connect with people they actually know, who feel in control of their social media, generally experience strong psychological benefits, and my own work has demonstrated that they can have a very positive effect on learning and motivation to learn (when assembled in the right assemblies).

Do you think governments should regulate web technologies such as social media and generative AI? Ban it? Put an age limit on it? Weigh in on the algorithms? Build their own?  Demand transparency? What’s a way forward for society and digital citizenship?

Honestly, I don’t know. I think direct regulation solely targeted at such tech is unwise and ineffective, though existing legislation does need to constantly be updated as the social context changes, and new adjacent possibles emerge. However, recent hastily and unwisely enacted legislation in many regions of the world will cause at least as many problems as it solves: this is the nature of all technology, including laws, but laws are particularly large, slow-moving, and rigid (hard) technologies that have a particularly strong tendency to result in unwanted consequences. It’s what Postman called technology’s Faustian Bargain. If they worked then it might be OK, but they won’t. If I ruled the world I would ask for all source code and training sets to be shared but a) that’s not going to happen and b) it would not be that much use: the trouble with deep learning stuff like this is that literally no one knows exactly how they work. It took researchers with super-powerful machines 6 months to figure out how one very simple prompt was answered for GPT3, which had a tiny fraction of the parameters of a more recent LLM. It would therefore be a good idea to put more resources into figuring better ways of evaluating such systems though, again, it is such a fast moving and diversifying target that it might be impractical. If, though, we could come up with an agreed, albeit constantly evolving “gold standard” for judging the outputs then we might at least have a starting point for thinking about what to do about it. However, it is moving too fast in too many directions. On the whole, we can rely on systems being shaped by the slow and large – what Stuart Brand calls pace layering – so legislation, though never catching up, can exert a strong moderating influence. However, there is one half-exception to that rule, when the collective behaviour of the small and fast becomes an entity in itself – think, say, of the effects of a swarm of locusts or army of ants or, for that matter, a human mob. That’s where we are now. The rate of uptake of genAI is unprecedented, and changing way too fast for conventional controls to hold down without clodhopping damage to things we do not want to damage, like freedom, research, and the small things of this world.

In our little corner of the world, in higher education, I think we can do things to help preserve what is good of the pre-genAI world, and seize what is good in the post-genAI world, if we do so mindfully and we are open to changes that many of us know we should have made at least decades ago. As educators, we can prepare people at least a little for what is only likely to be an even more rapid rate of change (barring catastrophe, that is the inevitable logic of the adjacent possible – an exponential increase in complexity and range of technology), as long as we are unafraid to engage ourselves, and as long as we can share what we learn. It is a really good idea to indulge in some serious futurology and I heartily recommend Jane McGonigal’s book “Imaginable” for ideas about how to do that methodically.  But, if I knew how to fix the bigger systemic issues I would be shouting loud about it. Alas, I do not.

You say about AI ‘’There’s no they there’. In a keynote talk at the last EdMedia conference Mike Sharples described ChatGPT as imbued with a liberal, slightly left-leaning US-American persona. What cultural norms do we want to have inscribed in AI, and what is actually going to flatten our discourse?

I think it is probably more accurate to say that there is every “they” there – not one identity but (almost) any identity. The main point is that the things are mindless generators of digital data, with no purposes, needs, desires, or values: whatever they produce reflects the training set and whatever other filters and bits of processing their providers provide.

Mike is right, though – as a consequence of a combination of training sets, instructions to trainers, and deliberate programmed intervention (filters, etc) that is exactly the case for ChatGPT. As an experiment to test this theory, I asked it to write something from the point of view of Hitler which, of course, it very sternly refused to do. I think that, if I were writing such a system, I would make it do the same, but such decisions are being constantly made about many much smaller and more equivocal things. It bothers me that this world view is being pushed everywhere, to many different cultures with very different norms. I would like to see far greater diversity in the big models. However, though big US corporations (and a few Chinese companies) do dominate the market because training such things is really expensive, there are many thousands of other models with different world views, including a lot of small language models that can run on local devices, even cellphones.

How will AI change the field and profession of instructional design? Is this change already apparent at your institution? Are we doing enough to prepare students for this and how can instructional design professionals adapt?

I don’t know how we can prepare students when we are so unprepared ourselves, except by involving them in the conversations, listening to them, and making their learning and ours more visible. I always think of students as co-teachers and myself as a co-student and now that matters more than ever.

Learning design teams need to be ready to be more agile, more responsive, more engaged with the intangible stuff, because the tangible can be done by generative AIs. It is better to think of the AIs as an expansion to the workforce than as replacements for what people can already do. AIs can solve many problems, from presenting stuff to creating interactive media, leaving learning designers more time to focus on how learning is happening rather than the mechanics of implementing stuff. They are also good sources of ideas and contribute effectively to brainstorming. They make fair critics, as long as their criticisms are examined closely.

At AU, I have colleagues who have used ChatGPT to write significant chunks of content, to help clean up English as a second language, and to design outlines and learning designs. I can’t say I’m happy with all of that. In fairness, the results are better than the average learning design team would normally achieve but, though they are functional for achieving hard outcomes, they are far from inspiring. It can provide a start, though, on which we can build more inspiring ways of learning. One of the learning designers on my team is using a system he has been working on to generate quiz questions and answers, and otherwise mechanical and dry task, which means he can put more effort into using these to better pedagogical effect. Other colleagues are using generative AIs directly, notably to support simulated business problems and environments. Others are using AI-generated avatars that closely resemble their faces, to read out slide notes in something closely approximating their own voices, an otherwise very time-consuming task: there’s a hint of the uncanny valley in the results, but it’s better than an unnarrated deck or robotic narration, and particularly handy for those with strong accents or speech impediments. Personally, I am adapting my courses to deliberately valorize use of generative AI, and I am directing students to AItutor Pro (by Contact North – a privacy-respecting free resource that does a kind of Socratic tutoring as well as playing more didactic roles) rather than static sites or textbooks for help with learning stuff like coding. There’s a lot more happening around the university, though we are still working on a full policy and there is no consolidated planning.

The changes I would most like to see are, for the most part, not what is actually happening. I despair of people using generative AI to create course content and process because it is not human – there are no quirks, no inspiration, not even dullness. The results, to date, are pretty much what you would expect a fairly well-written and very conventional course to look like (by default, ChatGPT really likes to set exams, I notice, though it can easily be persuaded to use better pedagogies). Not great, but not at all bad. The assessed activities are typically those that a generative AI could easily do, though, so it raises the absurd spectacle of generative AIs as teachers and generative AIs as students, largely skipping any humans in between. I think we need to rethink what we do and how we value education in quite a big way. Our education systems are mostly one gigantic McNamara Fallacy writ large, where we measure what can easily be measured and, at best, assume that the rest will take care of itself or, more often, just ignore it and pretend it doesn’t exist. What really matters are the values, attitudes, ways of thinking, ways of relating to others, ways of learning and ways of being that come for free with the tacit curriculum, that are taught as much by the system as a whole as by any particular teacher, and that are at great risk of being mutated by the increasing presence of generative AIs in the process.

From an ID perspective, those are the things we need to focus on supporting. GenAI can take care of helping students meet the hard outcomes: it’s the soft ones that we need to think more clearly about. And that means creating more opportunities to interact with other people, more opportunities to share, more opportunities to show caring, less focus on intentional outcomes and more harvesting of all the other learning that occurs. As well as thinking of instructional design, we need to think about teacher design: what can we do to support the passion, the caring, the skills of relating to others? And I think it would be worthwhile to think carefully about ways of unteaching: people who have been through educational McNamara Machines have been taught that nothing matters more than the grade, and so they are amotivated to learn for its own sake, and sometimes fail to recognize that they are being taught unless someone is telling them something, or showing them how to do something.

In terms of pedagogical change, with provisos about learning ways of being from a machine, I think there is great value in learning with AIs as partners on authentic, personally meaningful tasks. I have learned a vast amount in my adventures with using AIs to create apps for me, far more than I ever learned in classes or from books and videos. I also think there is great potential in using discriminative AIs to figure out what works, not so much in general but for individual students, kind of like an adaptive system on steroids. Terry Anderson and I, speculating about future pedagogies a few years ago, thought that one possible direction would be theory-free pedagogies of this nature: the enormous power of AIs to chew vast amounts of data and find patterns in them might be very helpful in supporting and guiding students on individual learning journeys. Equally, or as well, I see great potential for using them to help connect people and improve communication between them, for instance by summarizing arguments, prompting discussions, connecting similar students, and so on, as well as in helping teachers to provided better targeted help, identifying student needs, spotting surprises, and so on.

How do you think we can preserve the deeply personal and transformative aspects of education while still incorporating AI and technology?

It’s about valuing the human part of the process. We need to focus on creating connections—between people, between ideas, and between communities. Avoid being overly driven by assessment. Let students take risks, tinker, and explore without fear of failure. Encourage them to work on meaningful, personalized challenges that matter to them. At the same time, recognize generative AI as a partner, not just a tool. It can spark creativity and bring new ideas to the table, but the human context is irreplaceable. Celebrate the moments of collaboration, the diversity of perspectives, and the process of teaching and learning together. We have to intentionally preserve the aspects of education that make it human. These elements have always been there implicitly, but now we need to make them explicit and central to how we approach education with AI in the mix.”

What elements make education fundamentally human?

All of it. Everything in education is fundamentally human. Even something as mechanical as arithmetic isn’t just about numbers—it’s about the context in which it’s used, the way it’s taught, and the connections it fosters. Historically, education has always been about humans teaching humans, and the tacit aspects of that process—the values, culture, and ways of being—are central to it.

The problem with generative AI is that while it can simulate human behavior, it’s not human. It can’t offer the emotional depth, unpredictability, and genuine connection that come from real people. AI can teach the ‘hard stuff,’ but it also inadvertently teaches the ‘soft stuff’—and that soft stuff might be less rich, less complex, and less human.

The essence of education lies in the ways we live and interact as humans. Whether it’s the nuances of a conversation, the shared struggle of solving a problem, or the unspoken values that come through in teaching, these are the elements that make education transformative. It’s crucial to preserve these and be intentional about fostering human connections in the age of AI.

Is it challenging to integrate increasing technology in education while ensuring it enhances equity given the digital divide?


Yes, absolutely. There’s a digital divide, and the dominance of countries like the U.S. and China in tech raises cultural and equity concerns. Not everyone can afford subscriptions to tools like ChatGPT or Copilot, which gives some people an unfair advantage. But I think we need to focus on reclaiming the human aspects of education, using low-threshold technologies, and making tech work for us in ways that promote collaboration and creativity.

Many educators are concerned that students are cheating themselves out of an education by letting AI do all the work. One strategy you recommend is to decouple learning and assessment. What are the ways to assess a student’s knowledge if it’s not through assessments or assignments?

It’s about looking for what students have learned. Outcome harvesting is one way—it comes from project management and focuses on finding what actually happened rather than what was planned. For credentials, you can still use learning outcomes, but instead of focusing on specific targets like ‘I can write a JavaScript program,’ you might look at a broader portfolio where students show evidence of their learning, even unexpected areas like history alongside computing. Separating credentialing from the learning process is critical.

As an example, at Brunel University in the UK, they split the year into two halves. In the first half, students take learning modules where they focus on acquiring knowledge and skills without summative assessment—only formative feedback. The second half is for assessment modules where they apply what they’ve learned in project-based work, like portfolios or group tasks. These are personalized, allowing students to connect their learning to meaningful challenges. It’s been great for retention, and professors find it more satisfying because they focus on helping students learn rather than just judging them. 

What advice would you give students on how they could use generative AI in a productive way?

It’s hard to avoid specifics here. For me, generative AI is incredibly handy for creating things I couldn’t do as quickly or effectively on my own—like interactive software or presentation materials. However, I’d strongly advise against using it to write entire courses. I tried it as an experiment and, while the result was decent—better than average, honestly—it wasn’t great. Generative AI is a good B+ at best, but that’s not enough for meaningful education.

For students, it’s about using AI effectively as a tool to fill gaps or expand possibilities. The world we’re moving into will require them to know how to work with AI, so teaching them to focus on what they do with it and its impact on others is essential. For example, writing a JavaScript program isn’t an essential skill anymore—you get the basics, and then you use AI to help. The critical question becomes, ‘What do I do with this program? How does it connect with others? How does it change people or systems?’

Generative AI can also serve as a partner in creative processes. It can summarize arguments, nudge engagement, or even spark new ideas. But the key is to celebrate the human part—focus on creativity, human connections, and using AI in ways that enhance learning without replacing what’s fundamentally human about the process.

How does interacting with AI affect social-emotional development?

Interacting with AI changes us—it’s inevitable. If you spend a lot of time with generative AI, it shapes your ways of thinking and being, just like any learning experience. It’s human-like but not truly human, and that’s where the risk lies.

For instance, when I interact with generative AI, I tend to be very polite—it’s polite and patient in return, always. That’s very different from human interactions, which can be messy, impatient, and sometimes boring. These differences shape how we engage with people and develop our social-emotional skills. We need to learn that real people are impatient and annoying and dull and not all-knowing and still the best thing in the world.

Further Reading:

Dron, J. (2023) How Education Works. Teaching, Technology, and Technique. Athabasca University Press: Athabasca, CA.

Dron, J., & Anderson, T. (2014). Teaching crowds: Learning and social media. Athabasca University Press.

About

Professor Jon Dron is a member of the Technology Enhanced Knowledge Research Institute, and Associate Dean, Learning & Assessment in the Faculty of Science and Technology at Athabasca University. Jon has received both national and local awards for his teaching, is author of various award-winning research papers and is a regular keynote speaker at international conferences in fields as diverse as education, learning technologies, information science and programming. Jon has a first degree in philosophy, a masters degree in information systems, a post-graduate certificate in higher education and a PhD in learning technologies. Apart from his work in education, he has had careers in technology management, programming, and marketing, as well as over ten years as a professional singer. He is the author of Teaching Crowds: Learning and Social Media (2014, with Terry Anderson), and Control & Constraint in E-Learning: Choosing When to Choose (2007). His latest book, published in 2023, is How Education Works: Teaching, Technology, & Technique. He lives in beautiful Vancouver where, when he is not spending time with his wife, children and grandchildren, he sails, cycles, writes, sings, and plays many musical instruments, mostly quite badly. You can find out more about him at his personal website: https://jondron.ca