Welcome to Machine Learning and Human Learning!
What is this course about?
First it is a response to a certain new reality about learning, triggered by the advent of machine learning. Our ideas about learning are changing, and perhaps have already changed fundamentally. The history of technological supplements or supports, from the writing stylus to the abacus, the calculator, the Internet and the smartphone, has always shaped how we exteriorize our memories – put it out into the world – and in turn re-interiorize those memories during our recollection. Machines have, in other words, always been central to how we humans *learn*.
But today we also talk about machines learning how to do things themselves. Most obviously this happens in the fields of AI, and more specifically, of Generative AI. These machines now learn from *us* – from the traces of us we leave in data stored on the Internet. Machines learn from us; and from these new machines, we also learn differently. Asking ChatGPT to create a new recipe is different, we could say, from asking a calculator to add two numbers or even asking Google to find a recipe. We will explore this difference in the course, but for now we can note that we have entered a period in which "learning" comes to be applied to machines as well as humans, and that this learning – for both cases, machine and human – depends upon the other. Even if they differ, both forms of learning form, in their co-dependence, a kind of *synthesis*.
We can now speak in various ways about this much closer relationship we share with machines. We can speak of socio-technical systems; of cyborgian creatures (Haraway); of cyber-social systems (Kalantzis & Cope); and of compound terms like 'technosymbiosis' (N. Katherine Hayles) or a term I'll be using, in the context of learning, 'symbiotic pedagogy'. But the wider concept we'll be working with this week - and returning to, as we'll see - is that of *synthesis*. *Synthesis* will describe how both the similarities and differences between machine and human learning come together in our present moment.
Through the course we will be studying this relationship in a number of ways. Later I'll talk through the organization of the course materials.
On to some basic notes about the course and administration.
My preference is for these to be 90 minutes (6:00 - 7:30), with a brief break. I'm flexible here though. Essentially we have a lot we can present, discuss and practice.
The course will use CGScholar – a platform I think most of you will be familiar with? Assessments will be divided between:
A note on use of machine learning materials – for the formative assessment tasks, it would be good to use machine learning explicitly, but in a way that is visible. My preferred option for this at the moment is actually in the form of a dialogue – and edited chat session between you and your preferred chat agent / assistant – followed by some kind of summary by you, which includes your reflection on what the AI supplied, and your consideration of how useful the resulting artefact would really be.
Michele has already authored a fantastic and comprehensive handbook, and I have done some work on creative dialogues with AI – either or both could be models for how this task would proceed.
We can discuss which models, and how to access them in future weeks.A previous version of this course ran in 2023 and 2024: [[2023-Course]]
Despite this having the same name as that course, it is very different. And I would encourage people to look back to Bill and Mary's earlier course, and especially their videos, which are excellent. As you will probably gather, I have a quite different take. I would say by going back and forth, you might get the best of both worlds. Keep the link to the course handy, and feel free to consult both videos and readings.
So what distinguishes our approach in 2025? The first and biggest difference will be my reliance on the work of a early 19th century German Philosopher, Georg Wilhelm Friedrich Hegel.
Why Hegel? Why build a course on machine learning around a thinker who pre-dates, not only the recent era of machine learning, but the entire history of computation altogether?
I'll talk through five reasons in a moment.
But in the meantime, I also want to address why we might look at philosophy at all. Do we need it? And should we be intimidated by it? Hegel is also a very "intimidating" philosopher - perhaps the most intimidating in the Western philosophical tradition.
In what will become a key theme for us here, just as Hegel is a lens for us to understand and critique machine learning, machine learning itself will be a tool for understanding some complex philosophy.
First, I will be arguing one of the present limitations of machine learning is that it is stuck in a period of thinking about learning: the 18th century. According to the philosophy of the period – and I am of course simplifying here – the human mind was a blank slate that needed to be filled up with facts. I learn by simply adding new experiences, observations, sensations, to an existing repository, then retrieving from that repository to understand and integrate yet more new experiences. Learning was a comparatively simple process of *accumulation*. Just as life itself is the experience of a sequence of moments, each of which is compared to other moments, so learning is the acquisition of experience that are recorded in the mind.
This, as we'll review, is a form of the *empiricism* that underpins our current paradigm of machine learning. Now Hegel – as we'll also see – represents a profound challenge to the empirical tradition. His account of experience is not simply accumulative –– it is *dialectical*, which presumes that new experiences can sometimes *negate* those that they succeed. Learning proceeds, in other words, in a series of developmental stages, each of which involves some kind of realization or learning moment which can sometimes refute a past moment. Although, at least in Hegel's account, both past and present moment, even if they stand in contradiction, can be reconciled in a process of *synthesis*, constituting in turn a future moment which initiates the whole process all over again. Some of you might hear echoes here of later theorists of learning like Jean Piaget, who also supposed learning proceeded in development stages.
One final point here: for Hegel, experience is also organized by the concepts we have. For Hegel, as for Kant, I experience properly only if I have some concepts – such as time and space – to help synthesize that experience. We'll return to this point – it is a key distinction between how humans and machines – at least machines today – learn.
In addition – and here is the second reason – Hegel's account is rooted in our experience of *others*. In a way that prempts another learning theorist, Vygotsky, Hegel's account of human experience also involves a specific development from consciousness to self-consciousness. Ironically, our awareness of ourselves is closely connected to our awareness of other selves who are not our self - other people, in other words. This awareness means our experience and learning is *social*, conditioned and mediated by others. Now it is an open question as to whether machines are said to be social at all, but certainly not, as we'll see, in the sense that Hegel means. The third reason relates less to Hegel's direct views about learning, and more to his understanding of history. Within the Western tradition of philosophy, Hegel is the first and central philosopher of history – in the sense that he discusses history as something like a wider social and collective learning project. History moreover has a tendency or direction: it moves towards gradual realization or recognition, what Hegel terms, grandiosely, as the Absolute Idea or Absolute Knowledge. Outside religion, perhaps few today would agree strongly with Hegel that history follows a determinate path. However, with the arrival of AI – and certainly with much of the hype that comes with it – we are also forced to confront a series of questions: where are we going with AI? Will we arrive at a singularity, when AI becomes smarter than humans? And what skills do future human learners need to navigate this particular historical moment? Hegel doesn't answer this question, but – especially in his confrontation with the major events of his own time, including the French Revolution – he is already opening a way for consideration of our position within a wider time and history. Fourth, Hegel occupies an unusual position himself in the recent history of philosophy. Long regarded – for reasons we shall see, as we begin to look at some of his text – as an obscure and complex thinker without much value to mainstream philosophy, he has become central in recent decades to debates about language, norms and consciousness. Robert Brandom, a prominent American philosopher, has written for example an 800 page volume called *A Spirit of Trust: A Reading of Hegel's Phenomenology of Spirit*. Brandom, among others, is an important thinker in his own right about AI, and this suggests Hegel provides us with good background – if we want it – to contemporary discussions about learning and language. Finally, a challenge for us today is to think through what Artificial Intelligence means for us in its manifold sense: what it does for us, what it limits us to, what it enables and constrains, who it helps, who it oppresses, and so on. In other words, we need to think AI always in a certain spirit of contradiction: as involving the good and the bad, and understanding how these are to some degrees indissociable. Hegel is **the** thinker of contradiction, the paradoxes of "the this and the not-this" are throughout his work. There is an irony then that in trying to think about AI *intelligently* we have to give up on a particular rule of logic – the law of non-contradiction – that for many is the very hallmark of intelligent, that is to say, *consistent* thinking. We don't need to give it up entirely, of course, and we cannot – machine learning is bound up, for example, in binary logic, and it is foundational for us as well as for machines. But we can perhaps suspend it, in order to hold on to the idea that before we *judge* machine learning – in a disjunctive sense, as better or worse – we need to think it *conjunctively*: as both *this-and-that*, or *this-and-not-this*, simultaneously. We have to allow ourselves to do what, in a certain sense, machines cannot do, in order to talk about how they are different, as well as similar, to us.
The course is structured over eight weeks, and each week introduces and discusses a new concept.
1. **Synthesis** This is the introduction and overview we are doing today.
2. Experience
- Next week we'll be looking at the concept of Experience. This is a crucial idea in Hegel's **Phenomenology of Spirit**, and constitutes the single largest implicit criticism of non-human learning. We will examine how Hegel unpacks this idea, and contrast it with other ideas of experience (e.g. Locke) that might be closer to what we see of machinic "experience". We will also discuss the relationship of experience to learning, consciousness, perception and memory.Attention
- Following that, we will focus on the idea of attention. This is an idea that spans neuroscience, psychology, computer science, philosophy and media studies. We'll examine a key text in the development of machine learning, "Attention is All you Need". But we will also think about how attention is important for human learning. And we will look at recent critical studies of attention and what has become known as the Attention Economy.
4. Recognition
- Another key Hegelian idea, this week we will examine how *recognition* connects individual learning to our relation to others - to, in other words, a social process. We'll discuss here Vygotsky's theory of human learning, and also examine ways machines could be considered as "recognizing" us. 6. Alignment
Alignment is a "vogue" term in machine learning, and describes the process of aligning machines to human values and interests. One technical process for doing this is RLHF – reinforcement learning from human feedback. This relates closely to ideas of "norms". As opposed to rules or laws, norms describe often implied values that condition our practices, including our speaking practices. So "alignment" also can describe human learning – aligning ourselves with others. And increasingly today, we can also think of humans aligning themselves with machine values, especially in pedagogical settings. 7. Critique
"Critique" then brings us to what happens when alignment, one way or another, fails: when technology, in other words, fails to meet the wider social standards we set for it. Critique is another keyword also in the Hegelian nomenclature - it comes famously from Kant, for whom "Critique" is the method for understanding how "pure" (rational, scientific), "practical" (applied, moral) and "judgemental" (aesthetic) reasoning can happen. Here we will look at how Critical AI has been responding to AI - and look in turn at how AI works itself as a tool for, as well as of, critique. 8. Synthesis: Technosymbiosis
Finally, we return to where we started: with the concept of synthesis. However – just like a Hegelian process of argument – our original concept is now modified and inflected by the other concepts and materials we've encountered. In particular we are in a position to understand N. Katherine Hayles' idea of "technosymbiosis" – the combining of technology and life – a synthesis that also incorporates earlier ideas of experience, attention, recognition, consciousness, alignment and critique. This lecture traces a dialectical arc from the initial synthesis of machine and human learning to a higher reconciliation in technosymbiosis. Learning, for both, proceeds through experience as determinate negation: errors, surprises, and feedback that reshape models and minds. Attention functions as the selective mediation of this process—mechanically as weighting and salience, pedagogically as disciplined focus—organizing the field in which meaning can emerge. Recognition supplies the intersubjective and evaluative moment: humans become self-conscious through mutual acknowledgment, while models are shaped by social signals (labels, feedback) that confer norms. Across these movements, we clarify consciousness—self- and other-relation for persons, and the un/non-conscious operations of machines—so that we do not confuse functional intelligence with lived subjectivity. Alignment then names the practical task of orienting capacities toward shared ends, integrating ethical, social, and technical constraints. Critique provides the immanent negativity that guards against dogma and misalignment, refining concepts and objectives. The result is a second, richer synthesis: technosymbiosis, in which human Bildung and machine optimization co-develop, each sublating the other’s limits while preserving their difference.
Now some of you may have seen that I'm using some unusual software to do this presentation – and I'll be using it for the rest of this course. You'll see it's called "Hegel Pedagogy AI", and indeed I've written this software – or more exactly, I've prompted Claude to write the software. I used to be a software developer, and I've always wanted something that could convert notes into a presentation. So I asked Claude to write this "note converter" for this course. It may change week-to-week, and if I think it is useful for anyone other than myself, I'll release it.
But more to the point, I want people also to experiment with how these AI tools also write software. In fact this might be one of the sweet spots of AI today – as a software writing assistant. For the right applications, it has reduced the technical demands of users to nearly zero. One practical effect is that the key skills needed for so-called "vibe coding" are no longer deep algorithmic knowledge, but rather design and writing – specifically, how to talk to machines to get the results you want. Management, in other words, and not far removed from teaching. A new kind of human human learning is required to make use of these new kinds of learning machines, and to offset some of the conceptual terrain of this course, we'll also be practicing how to write software.Authors:
(hegel2025phenomenology?; kojeve1980introduction?; hyppolite1974genesis?; houlgate2012hegel?; heidegger1988hegel?; pippin2010hegel?; hegel2014science?; vzivzek2020hegel?)