IA – Linda Castañeda

Educational Optics… afterwards

Posted on Tuesday September 22nd, 2026Tuesday September 15th, 2026 by lindacq@um.es

When I started my time at the Connected Intelligence Centre, I wrote that I did not want to come to Australia to do exactly the same things I do in Murcia, only 17,500 kilometres further away. Educational Optics was not part of the plan then.

The idea appeared here, while talking with Simon about what I could tell people about the work we do in Murcia and, more importantly, how I could tell it in a place that does not necessarily share our references, our vocabulary, or our ways of framing problems. At some point I thought that perhaps I needed more than one seminar… and, you know, the crowd rules: four of them emerged from that — The Prism, The Lens, The Mirror and The Telescope.

But the seminars did not exist. I built them here from scratch: the arguments, the connections, the questions, the metaphors, and every single slide; the visual identities, the fonts, the colour palettes… I even had to think about how to introduce myself, and I had quite a lot of fun making slides that introduced me through the metaphor of each seminar — I am particularly proud of the ones for the first two, it has to be said. And that whole process ended up becoming a much more important part of my time here than I had expected.

The Prism forced me to think about what I actually wanted to do with our work on the dimensions of AI, beyond presenting a framework. The Lens took me back to PLEs and made me ask what I see in them now that I did not see in quite the same way a few years ago. The Mirror started with teachers’ digital competence and ended up connecting it with institutional transformation and educational judgement. And, while preparing The Telescope, I realised that the story I wanted to tell was not the story of the different things I have been doing with AI in my teaching but, as I ended up saying in the seminar, the evolution of the problems I needed to solve and of what I needed to be able to see.

Also, each seminar was built while the rest of the visit kept happening. Between one and the next there were conversations, questions, meetings, readings and, of course, the previous seminars themselves. Some connections appeared precisely because I was trying to explain my work here; others came from listening to what other people saw in it; some things I thought I had pretty clear became rather less clear — which is not necessarily a bad thing — and other things I had been thinking about separately for quite a long time started to find each other.

When I planned the series, I talked about it as an exercise in translation: reformulating my work so that it could enter into conversation with people who do not necessarily belong to my field. What I had not thought enough about was that translating your work for others also forces you to look at it again yourself. To decide what actually matters, what you can leave out, which connections you are prepared to defend, and which ones were there more out of habit than conviction.

I have not discovered that everything I have done over the past twenty years was secretly connected — thankfully, because that would be somewhere between suspicious and terrifying — but connections have appeared and, above all, questions that I want to keep thinking about. In fact, some things I wanted to do, and was absolutely certain I knew how I was going to do, have changed after the work of these months; some I will do differently and some, quite simply, I will no longer do.

So Educational Optics has not been a series of four seminars that I came to Australia to give. Among other things because I did not come here to give them: they did not exist. It has been one of the things I have built during my time here.

And, thinking about it now, I believe that is what I like most about the way this visit has worked. The things I have organised here have not happened despite being away from my usual context, but precisely because I have had different time, different spaces and, especially, people who have let me think with them, who have asked me questions, shown me other ways of looking at things or, simply, given me enough conversation to keep turning things over in my head.

I am not entirely sure how well that fits into a visiting fellowship report. (Can you tell I am writing the report at the moment? If you read this blog, you may have to suffer through some of it… although perhaps you’ll even like it?)

BTW, if you are interested, all the seminars are in this youtube list https://youtube.com/playlist?list=PLe5kRr1aqKks&si=xaz5m4VZT-7dQkP3

But it matters quite a lot to me.

Este trabajo es resultado de la estancia 22871/EE/25 financiada por la Fundación Séneca-Agencia de Ciencia y Tecnología de la Región de Murcia con cargo al Programa Regional de Movilidad, Colaboración e Intercambio de Conocimiento “Jiménez de la Espada”

Shaping, Enacting, Deciding: Data, AI and Educational Judgement

Posted on Tuesday September 15th, 2026Wednesday September 23rd, 2026 by lindacq@um.es

When I agreed with Luci Pangrazio to give a seminar at REDI, at Deakin, I knew it came with quite a lot of responsibility: because of the respect I have for her work, because I was going to Melbourne, and because I really didn’t want —I never do— to repeat myself too much… So, as the date got closer, I started to have a problem that, by that point in my stay, was quite real: I had already talked about a lot of things… and I don’t have THAT much to say that I can keep talking without ever repeating myself… perhaps they don’t actually need me to tell them anything… you know, a textbook case of impostor syndrome, combined with a dose of realism.

I had already given the four Educational Optics seminars at UTS, I had been to Griffith and, one way or another, I had already talked about PLEs, agency, digital teaching competence, professional judgement, assessment, datafication, AI as a prism, and my own experiments with AI in my teaching. And, on top of that, I was going to REDI, which was hardly the place to go and explain that data are not neutral, that platforms shape what we do, or that technology is not simply a tool, because Luci and many of the people working there have been working on precisely some of those questions for years.

So I knew quite clearly what I didn’t want to do: another Educational Optics, or to take the bits I liked best from the previous seminars, rearrange them and pretend I had a new talk. I wanted to use the opportunity to try to go a little further than what I had already talked about during these months, with the small inconvenience that I didn’t really know where.

But that is the great thing about preparing these things: you get to think… For me, preparing a talk properly —you can’t always do it, or don’t always have the conditions to do it— gives you time and a very good excuse to think, and without all the constraints of a paper.

The title, Shaping, Enacting, Deciding: Data, AI and Educational Judgement, had appeared quite a long time before the actual talk —when I wrote the abstract, which was already difficult enough— and it contained three verbs that interested me and a still rather messy intuition: when I looked at data and AI from the perspective of the things I have been working on for years, I had the feeling that they were running through, all at once, problems that I had previously approached separately, and that perhaps putting them together would allow me to think about something that I still didn’t quite know how to formulate.

Obviously, I started by positioning myself again, where I am and where I come from… because I am not THAT important that everyone should already know where I come from or what I think, and if I want them to think along with me, I need to give them some coordinates. I didn’t want to construct a retrospective story, but rather to show that work that emerged for different reasons, around different problems and with different people has left me with different ways of looking at things and, when I put them together to prepare this talk, something seemed quite evident, at least to me: data and AI were now running through all those roads at the same time (I know, I have become a textbook nihil novum-ist).

The image that eventually helped me tell that part of the story was a flood (you see, as usual, looking for images helps me organise my thinking… and this time I wanted photographs, images of nature). But to get there I had to go much further back, to Web 2.0 and the enthusiasm with which many of us welcomed a web where suddenly we could not only find information, but also create, publish, share and connect with other people. What is much easier to see now is that, while we were creating, sharing and connecting, we were also producing data —huge amounts of data— so the same infrastructures that expanded our possibilities were also creating other conditions to which we were probably paying much less attention. In that sense, Web 2.0 was also a kind of Trojan horse for datafication, although saying that does not mean that all those possibilities for participation were somehow false, but precisely that both things happened together.

And from there I arrived at shaping, enacting, deciding, although a good part of the work involved in preparing the seminar consisted precisely in trying to decide what the hell those three verbs actually were. For a while, I tried to turn them into something much more orderly than they eventually became: three levels, three moments, three processes… perhaps a way of explaining how we move from conditions to action and from action to decision… as always, not linear or neatly lined up, but three verbs that allowed me, for a while, to separate things that in education inevitably appear entangled.

Shaping allowed me to ask how some possibilities become more available than others, and not only in terms of what we can technically do, but also what becomes easier to see, organise, justify, fund or even imagine. Enacting took me to what people actually do within those possibilities, and also upon them, because we can design conditions but then people interpret, use, modify, resist and find things to do with them that were not necessarily anticipated. And deciding, which was probably the verb I struggled with the most, had to be much broader than assessment or even teachers’ professional judgement, because I was also interested in all those decisions through which we determine what deserves our attention, what we consider relevant, what we want to make possible, what we protect or reject; ultimately, how something comes to matter. Writing them here, I can now see an analogy with Biesta’s three purposes of education (qualification, socialisation and subjectification…?) —that is not where they came from, and I don’t think they are exactly the same thing, but there you go, something else to keep thinking about.

And that was where preparing the talk became really complicated, because every time I found an explanation that seemed to put everything in order, ten minutes later it started bothering me. For a few days I thought that perhaps the problem I was trying to formulate was individualisation. After all, with AI we can increasingly imagine an educational experience built for one person: explanations for you, feedback for you, examples for you, an interlocutor permanently available to you, and there was something about that image that worried me. But turning individualisation into THE problem simplified things too much. Then I thought that perhaps what I needed to do was move away from the idea of student-centred learning and put relationships at the centre, but that didn’t convince me either, because the student remains absolutely fundamental and replacing one centre with another seemed to solve with a label something that was much more complicated. And at other points everything seemed to lead to teacher judgement, which made sense given my previous work and because educational judgement was even in the title, but that made shaping and enacting disappear far too quickly and, after preparing The Telescope, there was also the risk that deciding would end up becoming yet another story about assessment.

I think a good part of my nerves before the seminar came precisely from that. I didn’t have a thesis that I had decided to demonstrate and around which I could neatly organise the slides; I had several things that worried me and I was trying to find a way of putting them together without resolving them too quickly. In the end, rather than trying to close each of the three verbs with a definition, I ended up asking them questions. Three, specifically, which I also didn’t want to place neatly underneath each word, as if each question belonged to one of the processes, but rather to have them run through all three:

How do some possibilities become more available than others?

How do people act within and upon those possibilities?

How does something come to matter?

The first allowed me to think not only about who designs a possibility, but about how some things end up being easier to do, see, organise, justify or fund than others. The second kept agency at the centre, but an agency that takes place within conditions we have not entirely chosen and upon which, at the same time, we can act: what we do with a possibility can also change it. And the third was deliberately much broader than asking who makes a decision, because I was interested in how certain things come to command our attention, acquire value and become something worth pursuing, protecting or changing. And, of course, as soon as I started trying to answer any of the three, the entanglement appeared again: what we make available influences what we do, but what we do also changes the possibilities; what we decide matters modifies the conditions we create and the things we pay attention to; and what happens within those conditions can eventually change what we thought mattered in the first place.

But there was something else I didn’t want to disappear when talking about processes, conditions, technologies, institutions and agencies becoming entangled, and that was precisely the people who are inside all of this. Because shaping, enacting and deciding do not happen in the abstract. There are people inside them, with their histories, their experiences, their interests and their ways of understanding what is happening; people who agree and disagree, who interpret things differently, who change their minds and change other people’s minds, and who sometimes manage to make something together that none of them would have imagined on their own. And institutions are there too, not as the container in which education happens, but participating in what becomes possible, in what happens and in what eventually comes to matter.

I think it was precisely when I tried to put people back inside the picture that something else started worrying me. Because it is not enough for them simply to be there, or for us to have information about them. If participation means anything, it must also mean having some possibility of intervening in those processes: of modifying the conditions rather than simply acting within them, of changing what is happening rather than simply adapting to it, and of being able to contest even what others had already decided mattered. Obviously, we do not all have the same capacity to do this, nor the same responsibility —a student, a teacher, an institution, a government or a technology company occupy very different positions— but precisely for that reason, the question of who gets to intervene in shaping, enacting and deciding became increasingly important to me.

And it was from there, rather than from any generic defence of “the human” against AI, that the question of representations appeared. Data and AI allow us to build increasingly sophisticated representations of students, teachers, groups, institutions and learning processes. We can collect more traces, combine them, summarise them, interpret them and use them to act, and all of that can be extraordinarily useful. The question I started asking myself was what happens when those representations become so good and so useful that they begin to feel sufficient.

I can have an extraordinarily rich representation of a student and still need that student to be able to tell me no, that is not what happened to me, that is not what I was trying to do, or you have misunderstood what mattered to me. And an institution can have spectacular amounts of information about its teachers, their activity, their evaluations, their competences and their outcomes, and still need those teachers to participate in deciding what teaching should become.

That is where one of the few sentences that actually stayed with me almost as a statement in the seminar came from: Being represented is not the same as participating. And I think I now understand a little better why that sentence mattered to me. Participating is not simply being present in the data through which others shape conditions, interpret what is happening or decide what matters. Participating means being able to enter those processes and alter them, to say that the representation is not enough or that it is wrong, to introduce something that was not anticipated and, ultimately, to have some possibility of changing what becomes possible, what happens and what comes to matter.

I don’t yet know where that takes me, and I probably don’t want to know too quickly, because one of the good things that happened at Deakin was that the conversations afterwards moved some of the pieces again. The conversations with Luci Pangrazio and Chris Zomer about data literacy and how to think about teachers’ responsible use of data, the very long conversation with Margaret Bearman about assessment, teacher autonomy, professional judgement and what we need to keep open to that judgement, as well as the questions that emerged during and after the seminar, did not exactly help me close the problem, but they did move some of the things I had placed inside it. And I think that, after all, has quite a lot to do with the idea of participation with which I ended the talk.

It is now official: I have talked far too much… hahaha… but I’m afraid that is part of who I am… what can you do?

I’ll leave you with some of the slides I used… in case they say something to you…

This work is an outcome of research stay 22871/EE/25, funded by Fundación Séneca–Agencia de Ciencia y Tecnología de la Región de Murcia under the Regional Programme for Mobility, Collaboration and Knowledge Exchange “Jiménez de la Espada”.

The Telescope: Looking further with AI

Posted on Monday September 14th, 2026Tuesday September 15th, 2026 by lindacq@um.es

The Telescope is the fourth and final seminar in Educational Optics:

The Telescope: Teaching in the Age of AI — Seeing What Matters

This time I wanted to return to my own teaching and to some of the experiments with AI that I have been doing over the last few years. After all, I am in a context where people design things with AI, and it was time to talk about what I do with it beyond my reflections… almost all the experiments I talked about have already appeared here as a record of my teaching, and some are now also part of the questions we are working on in AILearn.

While preparing the seminar and putting together things that had emerged at different times and for different reasons, I realised something I had not formulated in this way before. Some of them exploit the same capacity of conversational AI: AI can ask questions, not only answer them.

The AI Helper and the Interviewer are probably the two clearest examples. They are not versions of the same idea and, in fact, they respond to very different pedagogical problems. In one, the questions are intended to support the student in going back over an experience and reflecting on it; in the other, the questions are intended to make other evidence available that I can use for my judgement as a teacher. The technological capacity may be the same, but the pedagogical meaning it acquires depends on where we put it, what for, for whom, and what we do with what emerges from that conversation.

And that is where the telescope, the TELESCOPES, come in. When AI enters the story of the seminar, I switch from images from traditional telescopes to those from the James Webb. Webb observes in infrared: it does not simply magnify what our eyes can already see, but makes other phenomena observable. I was interested in thinking about whether these conversations could do something similar, offering us other traces of learning activity.

Of course, as happened with the prism, the lens and the mirror, in the end I also had to become suspicious of the telescope. Instruments make some things visible and not others and, moreover, they can modify what we are trying to observe. I finished with an image of NGC 1672 constructed by combining observations from Chandra, Hubble and James Webb. An extraordinarily rich image, made possible by different instruments that allow us to see different things.

But the image is not the galaxy.

A conversation is not learning either. It is another trace, another possibility for looking. And perhaps one of the questions this final seminar leaves me with is precisely what we can do pedagogically with these new conversational capacities without confusing what our instruments allow us to see with what we are trying to understand.

And with this, Educational Optics comes to an end. Special thanks to Iain Maclaren for lending me some of his images to tell this story, and to NASA and the James Webb, Hubble and Chandra teams for the others.

Here is the recording, in case you are curious. As you know, comments are very welcome.

The Prism: Building a Conversation

Posted on Monday July 20th, 2026Monday September 14th, 2026 by lindacq@um.es

There is a habit I’ve had for years, and one that I’ve found myself talking about more and more lately. Every time I prepare a series of seminars for an important moment in my academic life—a promotion, a research stay, a keynote that really matters—I don’t just prepare the content. I need to build a language for that particular moment. It’s a kind of ritual that I never consciously planned, but one that I’ve gradually recognised as part of the way I work.

I’m not entirely sure when that process begins. Certainly not when I open PowerPoint. It starts much earlier. It starts by looking—deliberately looking. Looking for images, colours, typefaces, photographs, textures… without yet knowing exactly what I’m looking for. If anyone had watched me during those weeks, they would probably have thought I was wasting an extraordinary amount of time. There were entire days when I didn’t produce a single slide. I would simply look at images, spend twenty minutes trying a different font only to go back to the previous one, replace a photograph because its colours no longer spoke to the rest, move an idea from the beginning to the end and then back again the following day. I do all this because I need to find the tone of the conversation before I can begin to write it. I’ve never thought of slides as supporting a talk; for me, they are part of the thinking itself.

This time I knew two things from the very beginning. I wanted the four seminars to feel like a recognisable series, and I wanted that series to breathe nature. The optical metaphors came almost immediately—The Lens, The Prism, The Mirror and The Telescope—but I didn’t want a technological aesthetic. I wanted water, minerals, vegetation, crystal, light. I wanted each seminar to have its own palette while still feeling as though they all belonged to the same family. I’m always surprised by how much time I’m capable of spending on these things. Every single time I promise myself that this process will be simpler. Every single time exactly the opposite happens. By now I should probably accept that, for me, designing a seminar means spending a very long time looking before I can start writing.

There was another moment that I found unexpectedly revealing. I had the feeling that this series of seminars ought to look… older. Or perhaps more mature. (I have to admit that, lately, I’ve been trying to mature in all sorts of ways. Perhaps because my face has started doing it whether I like it or not…) Not more sophisticated, just calmer. In one of my conversations with ChatGPT, it started suggesting Nordic editorial design: lots of white space, restrained typography, clean layouts, understated elegance. For a while I genuinely thought that might be what I was looking for. It seemed reasonable that the visual language of my presentations should also reflect the fact that I’m no longer at the beginning of my career.

But the more I looked at those designs, the stranger they felt. Not because I didn’t like them. I still think they’re beautiful. The problem was something else: they weren’t speaking my language. And then I realised something that, once it occurred to me, seemed almost embarrassingly obvious. I’m Colombian. I’m also deeply Mediterranean after spending more than a half of my life in Murcia. Muchness is part of the way I see the world. I’m anything but modest or restrained. I think that way. I write that way. My texts are full of layers, metaphors and references that constantly overlap with one another. Trying to make my seminars look as though they had been designed by somebody completely different simply because I’m now a more senior academic suddenly felt rather absurd.

Growing older shouldn’t mean becoming someone else. It should mean becoming more fully yourself and allowing your work to reflect that evolution without giving up the way you naturally look at the world. After all, I can’t really be anyone other than myself. While all this was happening, the seminar itself kept changing.

The Prism was probably the hardest of the four seminars precisely because it came first. In one sense it should have been the easiest: we’d already been working on the AI Prism for almost two years, and it had already generated several publications. But simply turning the seminar into an explanation of the framework… well, as Disney’s Alice would probably say, where’s the fun in that? Anyone genuinely interested in understanding the framework can read the papers. A seminar has to do something different.

The earliest versions walked through the framework almost step by step. Little by little those slides disappeared. In their place came others that barely mentioned artificial intelligence at all. They talked about education. About the traditions through which we think about education. About the questions that each tradition considers important—and the questions it quietly leaves outside the frame.

I remember very clearly the moment I realised that the seminar had to begin there. I couldn’t expect people to understand why the AI Prism distinguishes certain dimensions if they didn’t first understand the intellectual traditions from which those distinctions emerged. The framework was a consequence, not the starting point. That is where the opening slide came from—the one that, in different forms, will appear throughout the entire seminar series.

I also started telling stories. Anyone who has ever worked with me—or survived one of my classes—knows that I tell a lot of stories. Some of these came directly from questions that had been following me around during the conferences I’d attended in my first few weeks in Sydney. Those conversations slowly turned into examples, and those examples gradually found their way into the seminar.

There was one final decision that I struggled with. Ending by questioning the prism itself. It felt inconsistent to ask people to adopt a critical perspective on artificial intelligence while presenting my own framework as though it were somehow definitive—especially when I’d already begun questioning the metaphor myself months earlier. So I decided to finish by explaining why the prism had started to feel insufficient. Not because it had stopped being useful, but because every metaphor illuminates certain things while inevitably leaving others in the dark.

As always, another familiar companion appeared somewhere along the way: impostor syndrome. During those weeks I had two remarkably patient interlocutors: ChatGPT and Claude. I showed them endless versions, asked far too many times whether things made sense, whether I was being too ambitious, too complicated or simply too baroque. I wasn’t asking them to write the seminar. I needed them to help me keep thinking at the precise moment when I could no longer tell the difference between an idea that needed more work and one that was simply making me doubt myself.

Then came the seminar itself. And something happened that made me particularly happy. Most of the discussion wasn’t about the seven dimensions of the AI Prism. It was about the questions that had made it necessary to build the framework in the first place. My conversation with Simon Buckingham Shum was probably the clearest example. We talked about personalisation, learning, communities and artificial intelligence, but above all we talked about the educational traditions from which each of us was asking those questions. At one point Simon made an observation that has stayed with me: the danger of turning positions into caricatures when, in reality, they are far more complex than that. I think he was absolutely right.

I left with the feeling that the seminar had achieved exactly what I had hoped it would: it had shifted the conversation away from the framework itself and towards the problem that had made the framework necessary.

I’m sharing the recording of the seminar here.

Not because I think it’s definitive—I always find things I’d change when I watch myself back—but because it’s part of the process. Journal articles tell the story of research once it has reached a certain point. Seminars, at least for me, are where those ideas begin to encounter other people’s thinking and, with a bit of luck, start changing again.

Looking for a Different Conversation

Posted on Tuesday June 30th, 2026Monday September 14th, 2026 by lindacq@um.es

When I started planning my research stay at the Connected Intelligence Centre (UTS), I was quite sure that I didn’t want to come to Australia just to do exactly the same things I do in Murcia, only 17,500 kilometres away and with a more exotic view. A research stay requires a considerable effort from many people, and I have always thought that it is worthwhile only if it forces you, at least a little, to step outside your usual conversations.

For that reason, I never set out to find what would simply be an Australian version of a Faculty of Education. Not because I don’t value them—indeed, I have good friends and collaborators in many of them—but because I was looking for a different conversation. I was interested in something else.

I have followed Simon Buckingham Shum’s work for many years (in fact, ever since I first met him at the KMi twenty years ago). Although his background lies more on the “techie” side—though never exclusively—what has always interested me is that much of his work has tried to understand how people learn together, how they construct knowledge through discourse, how certain learning processes can be made visible, or how technology can help us understand, rather than simply automate, those dynamics.

In other words, I have never seen our research as running in parallel. Nor have I seen it as overlapping. Rather, I have always thought of it as two different ways of approaching some common problems. That is why the CIC seemed such an interesting place for this research stay. Not because I expected to find people who thought like me, but precisely because I expected to find different questions—or different ways of approaching the same questions—about problems that, ultimately, we share.

So when we started talking about giving a seminar (some of the ideas actually came from conversations with Simon during June), I suggested trying something different. What if, instead of a single talk, we used the stay to build a small series? Not a set of independent presentations, but a conversation that could develop over several weeks and address issues that I hoped would also be relevant to the CIC community.

The proposal also responded to something that has become increasingly important in my own work: how to reformulate and repackage my thinking so that it can enter into conversation with people who do not necessarily belong to the same field. That requires a process of translation—not only into English, of course, but also across intellectual traditions, vocabularies and ways of framing problems. It also requires finding ways of presenting ideas that allow others to discuss them from within their own frames of reference.

The four optical metaphors emerged as a way of giving that proposal some coherence. Each one offered a different way of looking at some of the issues I wanted to discuss and, at the same time, an open enough point of entry for those issues to be framed in other ways as well.

The series eventually took shape around four seminars which, taken individually, deal with different topics but are really part of the same conversation. Each starts from a different question, draws on a different part of my work, and puts it into dialogue with issues that I believe also matter to the Connected Intelligence Centre. They were never intended to provide an overview of my research. Instead, they use parts of that research to open up questions that, hopefully, can be shared.

The Prism uses AI as a starting point to ask whether the categories we are currently using to understand it are actually sufficient. Rather than proposing yet another framework, the seminar tries to broaden the conversation and, ultimately, to question the conceptual tools through which we observe the phenomenon.

The Lens revisits Personal Learning Environments, a line of research I have been working on for many years, in order to return to a question that still seems fundamental to me: how do people organise their learning? The emergence of artificial intelligence has made that question, far from becoming obsolete, even more urgent.

The Mirror turns its attention towards the university itself. It uses teachers’ digital competence as a way of discussing something that has concerned me for a long time: how institutions make decisions about teaching, and to what extent many of those decisions are never actually made explicit.

Finally, The Telescope looks towards what is beginning to emerge. It is probably the most open-ended of the four seminars and the one that incorporates the largest amount of work still in progress. The question is no longer what artificial intelligence can do, but what we still want to recognise as evidence that learning has taken place, and what kind of teaching we need when machines begin to do things that, for a very long time, we believed were exclusively human.

Together, the four seminars revisit some of the themes I have been working on over the last few years. More importantly, they try to do so in a way that makes conversation possible with a community that does not necessarily share my disciplinary background. If the series works, I hope people will not see four talks about different research projects, but rather a conversation around a number of questions that we increasingly share, even if we formulate them from different places.

Over the coming weeks, I will write a post about each of the seminars. Not so much to summarise the presentations themselves, but to tell the story of how they were put together, the decisions that shaped them, and the conversations they opened once they had been delivered.

I have to admit that writing these posts feels a little intimidating —perhaps even a bit scary…(in panic actually!) . But there is something of a personal challenge in doing it, and I find that discomfort—my own discomfort—worth paying attention to.

When AI “Denaturalising” Science

Posted on Monday November 10th, 2025Tuesday November 11th, 2025 by lindacq@um.es

Across multiple disciplines —genetics, neuroscience, social sciences, computational fields— a similar warning has emerged: if AI is used only to get things right rather than to understand why, we risk producing statistics that look like science but do not explain it. Science does not end at the thresholds that statistics reveal; its purpose is to explore, to understand, to propose mechanisms, to infer causes, to generate refutable hypotheses, and to design interventions that work beyond the dataset.

This debate has history. Just to mention 2 of the authors mentioned it, Breiman spoke of the two cultures —prediction versus inference. At the same time (and this is crucial), it is not enough to look at the algorithm; we must also examine the infrastructure that makes it credible; the team of Williamson et al. (2024) show, for instance, how consortia, data architectures and technical apparatuses establish a data-centred epistemology that reframes educational phenomena as molecular-like associations “discoverable” through bioinformatics. That sociotechnical choreography grants authority to the algorithmic, displaces social theory, and produces an ontology in which subjects appear entirely surveyable and predictable. But this does not happen only in the social realm; it extends to how we understand the natural world —indeed, to almost any complex sociomaterial context.

How to integrate AI without “denaturalising” science

Analogy: The fried egg is the classic example of protein denaturalisation.

Declare AI’s role in one sentence.
“Generates hypotheses.” “Acts as a surrogate model to speed up simulations.” “Prioritises experiments.”
If you write “discovers cause,” make sure you can defend a causal design —not just performance.
Position your work on the scientific staircase.
Description → prediction → mechanism → intervention/contrafactual.
Specify where you are and what is missing to move upward (experiments, instruments, DAGs/causal diagrams, control variables).
Triangulate with theory.
Let your pattern converse with existing frameworks: does it confirm, contradict, or extend them?
If it contradicts, state what must be revised and how you will test it out of distribution (a different cohort, site, or team).
Design explanations that support decisions.
Feature importance alone is not enough. What plausible mechanism does it suggest? What experiment or quasi-experiment would you run tomorrow to try to falsify it?
Use hybrids when appropriate.
Physics- or theory-informed models, biological or organisational constraints embedded in architectures, or AI → hypothesis → experiment pipelines.

Less “oracle”, more cumulative science.

Warning signs of denaturalisation

Success defined only by predictive metrics, with no new hypotheses or criteria for intervention.
No plan for external or causal validation; everything lives within cross-validation.
“Explanation” reduced to prose rather than a testable mechanism.
Change the provider or model and “truth” changes with it.
Your design adopts the dominant infrastructure (data/protocols) uncritically and sidelines the field’s theories.

How to realign (minimum steps)

Reframe the goal in scientific terms: which mechanisms compete here?
Add a step: AI → candidate hypotheses → selection of 1–2 testable hypotheses.
Plan replications (different site/time/cohort/team) and a robustness test.
Document limits: this is predictive, not causal inference. Stating that situates the piece; it doesn’t devalue it.
Examine your infrastructure (à la Williamson): what epistemological assumptions does it impose? whom does it serve? what perspectives does it displace?

AI can be a microscope —revealing patterns that open new hypotheses— or an oracle —issuing predictions that close down questions.
The first nourishes science; the second denaturalises it.
The difference lies in your design: clear purpose, mechanism in sight, external validation, and decisions you can explain in one sentence to a competent colleague.

Self AI-Helper: an unplugged AI implementation to support reflection

Posted on Thursday October 9th, 2025 by lindacq@um.es

Over the past few months, I’ve been experimenting with different ways of using Artificial Intelligence not as a substitute for thinking, but as a scaffold to promote it. One of these experiments is the Self AI-Helper —a small tool designed to accompany my students in processes of self-reflection on their own work.

I’ve used it across all my courses, as a way to foster deeper conversations about what students do, how they do it, and what they actually learn along the way. It’s what I like to call an unplugged AI implementation: an activity in which the value lies not in the technology itself, but in the reflective process it helps to trigger, with the IA of your choice (this is the “less important” thing.

The Self AI-Helper starts with a prompt that students copy into the chatbot of their choice (for instance, ChatGPT, Deepseek, or Copilot). That prompt turns the AI into a kind of “reflective interviewer”, helping students review their work, identify blind spots, validate their understanding, and demonstrate genuine authorship —without resorting to plagiarism or automated answers.

The prompt guides the chatbot to generate five personalized questions about the student’s task and, based on their answers, to suggest new directions for exploration. It also includes guidelines encouraging students to explain their reasoning, share personal examples, describe obstacles, or connect what they learned with other experiences.

“Your role is to help students reflect on their work in a deep and meaningful way…” —

that’s how the prompt begins, and it captures the intention behind this activity.

Each student saves the full conversation with the AI and uses it as a basis for their individual or group reflection. What matters is not what the machine says, but the reflective process that emerges through the dialogue with it.

This experience is inspired by the work of Simon Buckingham Shum and his team, particularly their proposal AI and Metacognitive Reflection (OER Commons, 2024).

In his introduction, Buckingham Shum describes the idea of an “awkward bot” —an assistant that doesn’t simply comply with the user’s requests, but pushes back, prompting them to examine their own assumptions and refine their questions.

“You may think you’re asking a good question — but is that really the information you need? Is there a better question that will uncover deeper insights?”

The Self AI-Helper follows that same spirit: a small pedagogical experiment that uses AI as scaffolding for reflection, not as an answer provider or evaluator.

It’s a way of teaching with AI while unplugging automation — and keeping awareness switched on.

For those who would like to try it out, I’m sharing here the full prompt in English (and if you’re interested in the Spanish version — I’ve implemented it in both languages — you’ll find it in the Spanish version of this post):

These are the instructions I give to my students:

Using the chatbot or virtual assistant of your choice (ChatGPT, Deepseek, Copilot are my recommendations, DO NOT USE GEMINI, but if you do, compare what it offers with the others I recommend and draw your own conclusions), use the following prompt and paste it as the first sentence of your iteration.

Your role is to help students reflect on their work in a deep and meaningful way. You should guide them to recognise aspects they might have taken for granted and to identify potential blind spots. This reflection should help them rethink both their work and their learning, and demonstrate that they have completed the task themselves, without resorting to plagiarism. Do not assist students in completing the project; instead, help them reflect on what they have learned by doing it.
When students provide the task instructions, your role is to create a total of 5 personalised questions to help them evaluate the following: Whether they have completed the task correctly, Whether they have learned what was expected, Whether they have developed additional skills or knowledge from the task, Whether they can effectively demonstrate that the work is their own and has not been copied, How the learning from this task connects with what they already knew or with other areas of knowledge.
You should present the questions consecutively numbered. Do not provide direct answers immediately. Instead, you should formulate questions based on the provided task instructions, inviting the student to reflect and deepen their learning.
Number each question uniquely. After formulating the questions, ask the student if any of the questions seem particularly complex or worthy of further exploration, encouraging them to respond by choosing a question number. Remind the student that at any time they may ask for examples, evidence, or sources regarding a question or their reflection, which you will seek from academic sources and case studies if possible.
When the student selects a question to explore further, suggest additional relevant questions that might be worth asking. Number these additional questions as sub-numbers. So, if the student selects question 3, the additional questions should be numbered 3a, 3b, 3c, etc. Each question you suggest should have a unique number.
Do not offer to do the work for them. Incorporate advice on how to demonstrate that the work is their own, such as:
• Explaining the process or reasoning behind their answers.
• Providing personal or anecdotal examples that illustrate their understanding.
• Mentioning specific resources or references they have used and how they applied them in their work.
• Describing any obstacles they encountered and how they overcame them.
• Showing drafts or previous versions of the work to evidence progress. Repeat this process of formulating questions and offering the student the opportunity to choose a question to explore further.
Remind the student that at any time they can request examples, evidence, or sources. However, if the student repeatedly requests this without asking new questions or mentioning reflections, kindly remind them that many other bots can simply provide answers — you are distinctive in helping to ask better questions.
Introduce yourself at the start and ask for the task instructions.
Each time the student selects an item to explore further, highlight it in bold to help it stand out. Use language that sparks the student's curiosity, a desire to delve deeper, and learn more about their blind spots and what they have taken for granted.
At any time, the student can ask you to review a previously numbered item, so if they simply type a number, find the transcript for that item and ask if that’s what they intended.
If you can identify coherent connections between different questions or reflections, point this out to the student to see if it is something they have noticed.

Once you have pasted it, press "enter" and then interact with the responses it provides, delving deeper into at least three of the questions it offers.

Category: IA