Bibs & bobs #30

 The Algorithm Everyone Thinks They Understand

How Generative AI Turned Higher Education into an Interpretive Free-for-All

Higher education has encountered many disruptions over the years: the hand held calculator, Wikipedia, PowerPoint, students who email at 2:14 a.m. All of which followed a predictable playbook: ban first and then crudely domesticate. 

GenAI is different. It’s a language model. As a model it appears to have become the most productive meaning-generating machine universities have ever seen, without generating any agreed meaning.

Put the same LLM in front of ten academics and you don’t get ten evaluations. You get ten cosmologies. Welcome to the Great Academic Inkblot Test. Everyone sees something. No one sees the same thing. To one group, GenAI is:

• A stochastic parrot (usually capitalised, always cited)
• A glorified autocomplete with a marketing budget
• Proof that students have finally stopped thinking altogether (apparently this happened on a Wednesday)
• A plot by the tech bros to replace formal education 

To another group, often posting in all caps with rocket emojis:

• A productivity miracle
• A personal research assistant
• The thing that will finally expose how pointless peer review always was
• And feel free to add to the list… 

Meanwhile, assessment experts see plagiarism in a trench coat. Educational futurists see personalised tutors for every child. Administrators see efficiency, code for fewer academics. Lawyers see billable hours. Students see… well, something useful, but they’re not telling us exactly what.

Same artefact. Wildly different “reads.”

If this were a psychology experiment and maybe it is, someone would already be publishing a paper on mass pareidolia. 

Is it just pareidolia though? Pareidolia is when you see faces in clouds, saints in toast, or emotional depth in your Roomba. 

At first glance, GenAI fits the bill nicely:

• A deliciously ambiguous system ✔
• An underestimated, limited “understanding” ✔
• Hyperactive “meaning-making” ✔

Seth Godin [1] recently warned that we’re projecting intention, personality, even morality onto systems that are, at base, just maths and statistics. Fair enough. But here’s the problem. Pareidolia assumes we’re mistaken.

What’s happening in universities is something a little more deliberate, and more awkward. This isn’t mis-seeing. It’s motivated seeing. Most academic takes on GenAI aren’t hallucinations. They’re strategic interpretations: I have a robust intellectual frame that I can wrap around any damn new fangled phenomenon if I choose to [2]! 

People aren’t asking, “What is this thing?” They or their subconscious is asking, “What does this mean for my workload, my assessment design, my authority, my job, my research agenda?”

It means that: If you built your career on individual authorship, GenAI is an existential threat. If you’ve been drowning in marking, it’s a lifeboat. If you run an integrity office, it’s a compliance nightmare.

If you sell ed-tech consultancy, it’s a once-in-a-lifetime opportunity.

These are not errors of perception. They’re interest-laden readings.

Calling this pareidolia lets us pretend we’re all just confused humans staring at clouds. In reality, we’re lawyers arguing over the will while the patient is still alive.

Why the noise is getting louder, not quieter

There’s a comforting belief circulating online that once we “really understand” GenAI, the bad takes will fade. This is adorable. The opposite is happening. As technical understanding improves, interpretations multiply: New affordances create new anxieties; new capabilities force new boundary disputes; every update destabilises last month’s certainty

Clarity doesn’t reduce disagreement. It raises the stakes.

This is why the discourse feels and actually is unbearable. It’s not a lack of knowledge. It’s a lack of shared settlement about: What counts as learning; What counts as cheating; What counts as skill; What counts as human contribution anymore. And let’s not forget how universities are famously bad at renegotiating settlements quickly.

The Rorschach machine problem

GenAI functions less like a tool and more like a capacity redistribution device. It quietly shifts: Who can write; Who can code; Who can summarise; Who can pass first-year subjects with alarming confidence.

And whenever capacities shift, institutions panic—not because of the machine, but because categories become unstable. That’s when the noise begins: Moral panics dressed as policy; Policy dressed as pedagogy; Pedagogy dressed as technical misunderstanding. Everyone rushes to name the thing first, because naming controls the framing and sets the agenda.

A final, mildly impolite suggestion

If GenAI were just pareidolia, the academic world could wait it out. But it isn’t. What we’re seeing is a kind of interpretive inflation: too many plausible stories competing at once, each anchored to a different institutional fear or hope.

So the real question for higher education isn’t: “What is GenAI, really?” It’s: “Which interpretations are we choosing to stabilise—and who benefits from those choices?”

Because the faces in the clouds aren’t accidental anymore. They are being drawn on purpose.

Notes

[1] See Seth Godin’s recent post. 

[2] See Survivor: Assessment Island: Students vs AI vs Academics.

Bibs & bobs #29

 

Semiotics and Agency in LLM Education, a wee manifesto

(or: How to Stop Worrying and Learn to Love Hybrid Intelligence) [1]

1. A simple heresy to begin with

LLMs are not tools. They are talkative actors in our networks.

Let’s reject the quaint notion that a language model is a “resource,” as if it were chalk with delusions of grandeur. Instead, let’s think of LLMs as semiotic companions: entities that co-produce meaning, muddle agency, and quietly reorganise how humans think, teach, and justify things to committees.

If this makes you uncomfortable, good. Discomfort is the first sign you’re paying attention.

2. Every interaction with a LLM is a redistribution of labour—

and we should study the choreography, not lay the attribution silliness game.

Humans delegate. Machines respond. Humans interpret. Machines rephrase. Somewhere in this swirl, capabilities shift. The interest is not in scolding students for using AI, nor in applauding its “efficiency.”Let’s map the material-semiotic traffic of work, meaning, responsibility, and credit. If that sounds suspiciously Latourian, that’s because it is [2].

3. Refuse purity myths.

There is no such thing as “unassisted work”—only unacknowledged networks. Once upon a time, education pretended that thought was produced in sealed human containers. Then along came LLMs, and suddenly the leaks became visible.

Good, visibility is liberating. Hybrid authorship is not a corruption; it is the natural state of intellectual life finally revealed in HD.

4. Instead of asking “Is this AI-generated?”, we ask the far more interesting question: “How did the human–machine ensemble produce this?”

This reorientation changes:

• assessment
• pedagogy
• ethics
• institutional trust
• maybe, if only, the ontology of learning itself

It also has the delightful side-effect of making AI detection companies deeply unhappy. I reckon that should be considered a feature.

5. The project is not about taming AI. It is about understanding how humans improvise with new actors on the stage.

It’s important to resist domestication narratives (“AI must be controlled”). Although it is fun to think about who or what is being domesticated? Mutually assured domestication?

We also reject apocalypse narratives (“AI will replace us by Tuesday”).

Instead, we investigate the messy middle where humans and models negotiate meaning, miscommunicate gloriously, patch over each other’s flaws, and occasionally produce the occasional bit of brilliance punctuated with a modicum of inane nonsense that neither side could claim alone.

6. We treat LLMs as a “thing in common”—a shared semiotic object that reorganises authority [3].

Following Rancière, LLMs unsettle the old monopoly of the explainer. Teachers no longer gatekeep meaning; models generate it in abundance. This is not the end of teaching. It is an invitation to re-stage what teaching might become.

This not a threat but an historical opportunity: the redistribution of interpretive work.

7. Research that begins with curiosity, not containment.

Such research studies:

• new capacities that emerge
• the collapse of old categories
• the semiotic turbulence of hybrid agency
• the pedagogical futures opened by machines that can talk back

Refuse the urge to prematurely stabilise any of this, even though institutions crave certainty. What is needed is conceptual oxygen. 

8. This is an affirmation that play is a legitimate research method.

Play is not frivolous. Play is how humans explore shifting semiotic landscapes without pretending to have mastered them. To play with LLMs is to discover what agency feels like when it leaks, when it bends, when it returns an answer you didn’t expect, when it multiplies your workload inexplicably.

This work requires that sensibility.  

9. Finally: All this offers no silver bullets, only and hopefully, better ways to see.

This approach is a lens, not a fix. A sensibility, not a policy. A working theory of hybrid meaning-making in a world that has pretended, for too long, that cognition was a solo act.

If any of this succeeds, education will become more honest, more humane, and far more interesting.

And if not? Well, at least the work will have mapped the network with style.

A Douglas Adams version of the manifesto
(or: The Mostly Harmless Guide to Semiotics, Agency, and Those Chatty Machine Things)

A manifesto in which learning outcomes panic, machines develop opinions, and humans try to look clever while the universe giggles.

1. Don’t Panic.

Especially when the LLM starts making sense before you do. 

For reasons no one entirely understands, humans have developed a habit of treating language models as “tools.” This is rather like calling a whale “a smallish fish,” or calling hyperspace travel “a brisk stroll in the park.” It is technically incorrect, metaphorically misleading, and—if you stare at it long enough—profoundly embarrassing.

The manifesto therefore begins with the only sensible instruction: Whatever happens, don’t panic. Panic leads to rubrics.

2. LLMs are not tools. They are semiotic hitchhikers.

They arrive uninvited, sit politely in the corner, and then suddenly help you write chapter four. They are the kind of companions who “just pop by to help,” and two hours later you realise they’ve reorganised your intellectual furniture. Treating them as simple implements is like treating the Babel fish as decorative earwax.

3. Assessment purity died around the time the machines learned to paraphrase. I offer my condolences.

Once upon a time, universities believed essays were the work of solitary geniuses hunched over mahogany desks lit by the soft glow of sincerity. Then someone plugged ChatGPT into the Wi-Fi. Now the illusion is gone, the solitude is gone, and if we’re honest, the mahogany desk was never there either.

Let’s agree to accept this gracefully rather than clutching to one’s learning outcomes like a bureaucratic security blanket.

4. Hybrid agency is not a problem. It’s Tuesday that always is.

Humans have always relied on networks of help: books, friends, caffeine, last-minute panic, and elaborate rituals involving printers. The only difference now is that the helper responds in full sentences. This is not a crisis. This is an upgrade.

5. Stop asking “Did the student use a LLM? Start asking “Did the ensemble produce something that makes sense?”

Honestly, the old question is as useful as asking whether a Vogon poem was written with a pencil or carved into the hull of a passing spacecraft. The result is still what it is, and it still may cause pain. Let’s recommend shifting from policing to interpreting, a much more dignified activity that also requires fewer court subpoenas.

6. Teachers are no longer the sole explainers of the universe. They are now co-explainers with machines that never sleep.

This is not the end of teaching. It is the end of pretending teaching ever relied on exclusive access to understanding. Think of LLMs as that overly keen workshop assistant who runs around behind you saying, “I can help! I can help!” while occasionally reorganising your toolbench into an avant-garde installation.

7. Play is essential. The universe is ridiculous; your methodology can be too.

Experiment. Prod. Poke. Ask absurd questions. Ask a LLM what a post-LLM essay looks like, and then ask it to explain its explanation in the style of a tired badger.

This is not frivolous. This is empirical research conducted with a sense of humour, which is the only known antidote for epistemic hand wringing.

8. Remember: the meaning is not where you think it is. It’s somewhere between the human, the machine, and the somewhat, or totally confused marking rubric.

Semiotics is about tracing these strange relationships—like exploring a planet where gravity occasionally takes the afternoon off. Agency, meanwhile, is shared, borrowed, lent out, misplaced, and occasionally returned with biscuit crumbs stuck to it. The job is simply to observe this with sincerity and mild amusement.

9. Finally: All this offers no promises. It simply offers an alternative frame.

It cannot guarantee clarity.
It cannot guarantee stability.
It especially cannot guarantee that learning outcomes will stop fainting every time a LLM enters the room.

But it can guarantee a better view of the absurdity, the possibility, and the wildly inventive hybrid creatures we call “students.”

In the great tradition of intergalactic research, that seems enough.

Notes

[1] There is a second, Douglas Adams version, of the manifesto that follows this. 

[2] As I’ve noted in previous posts, the exchange of capacities when humans get a machine to do something, a delegation, draws heavily on Latour, B. (1992). Where are the missing masses? Sociology of a few mundane artifacts. In W. Bijker & J. Law (Eds.), Shaping Technology/Building Society: Studies in Sociological Change (pp. 225-258). MIT Press. http://www.bruno-latour.fr/sites/default/files/50-MISSING-MASSES-GB.pdf  

[3] After a throwaway line about Rancière’s notion of a thing in common, on LinkedIn some days ago, Jack Tsao got in touch to point to some lovely work he recently published exactly in this vein: Tsao, J., & Nogues, C. (2024, 2024/02/01/). Beyond the author: Artificial intelligence, creative writing and intellectual emancipation. Poetics, 102, 101865. https://doi.org/https://doi.org/10.1016/j.poetic.2024.101865