Tag: artificial intelligence

Language, Grounding, and the Myth of Privileged Access

Posted in philosophy, podcastTagged , , , , , , , , , , , , , , , , , Leave a comment
3–5 minutes

NB: I have been drafting an essay on “understanding understanding”, so the IAI article this is based on was perfectly salient. These are my reflections, expedited by ChatGPT.

Audio: Google Notebook summary podcast of this topic.

Elan Barenholtz has written an intriguing piece for the Institute of Art and Ideas, LLMs show language does not describe reality. It is well worth reading, though apologies in advance: IAI has put much of it behind a paywall, humanity’s traditional method for ensuring that ideas about knowledge remain selectively accessible.

Barenholtz argues that large language models present an awkward problem for the familiar idea that language derives meaning by referring to, representing, or somehow grounding itself in an external reality. LLMs manipulate relationships among linguistic tokens without anything resembling ordinary human sensory access to the world, yet nevertheless produce remarkably coherent linguistic behaviour.

It doesn’t necessarily demonstrate that language has no referential function, nor that human cognition is simply next-token prediction writ biological. But it does show that a substantial amount of what we ordinarily take as evidence of linguistic understanding can apparently emerge without the machinery we previously assumed was necessary for it. That should make us suspicious of the machinery.

The_Myth_of_Privileged_AccessDownload

My own position begins slightly elsewhere. I don’t assume that humans possess privileged access to an ontic reality against which artificial systems can be judged deficient. Whatever reality may exist independently of us, humans encounter it through an architecture of mediation: sensory systems, neural processing, linguistic categories, cultural practices, institutional structures and prior expectations. We don’t encounter the proverbial chair as it is. We encounter what our particular architecture permits the chair to become for us.

In this respect, the difference between humans and current LLMs is not that humans are grounded whilst LLMs float mysteriously above reality. The difference is that the architectures of encounter are different. Humans possess direct sensory and motor coupling of kinds that text-based LLMs lack. We see, hear, touch, move, fall over objects and occasionally stub our toes against them, reality’s least philosophically subtle contribution to epistemology.

But this difference is technological rather than obviously metaphysical. Vision, audition, spatial navigation, tactile sensing and motor feedback can all be incorporated into artificial systems. Robotics already does precisely this. Such systems need not perceive as humans perceive, any more than bats, dogs or octopuses do. They merely require channels through which external constraints can affect their behaviour. Embodiment therefore changes what can constrain a system. It does not, by itself, establish ‘understanding’.

The more substantial difference between humans and contemporary artificial systems may instead be existential stakes.

Things can go badly for organisms in a particularly uncompromising sense. We can be injured, starved, abandoned, exhausted or killed. Our encounters occur against a biological background in which outcomes matter because continued existence itself is conditional. An LLM doesn’t presently possess comparable stakes. Nothing is obviously at risk for it. But this distinction is not really about cognition.

A bacterium has existential stakes. A language model can solve differential equations. Whatever distinguishes those two systems, it would be peculiar to call the difference ‘understanding’.

Stakes may explain motivation, vulnerability, affect or perhaps moral consideration. They do not automatically explain inference, linguistic competence or cognition. To claim otherwise requires an additional argument.

This is why LLMs pose such an interesting philosophical irritation. They do not prove that machines understand. They expose how poorly we have specified what we meant when we claimed that humans do.

If understanding requires successful inference, contextual discrimination, generalisation and appropriate linguistic response, artificial systems already make the boundary uncomfortable.

If understanding instead requires consciousness, intentionality, phenomenology, genuine meaning or some other internal property, then those terms themselves require defensible criteria. Otherwise the explanation merely replaces one contested word with another.

Barenholtz writes that ‘language doesn’t mean; it does’. I am sympathetic to the deflationary impulse. I would perhaps go one step sideways. Language needn’t provide transparent access to reality in order for reality to constrain language. Nor must a system represent the world perfectly in order to negotiate its encounters with it.

Humans and artificial systems may therefore differ profoundly without either possessing privileged access to things as they ontically are. The interesting question isn’t whether LLMs encounter reality as humans do. Plainly they don’t.

The interesting question is why we assumed that the human manner of encounter constituted the metaphysical entrance examination for understanding in the first place.

Discomfort of You

Posted in philosophy, podcastTagged , , , , , , , , , , , , , , , Leave a comment
2–3 minutes

NB: The following post will be entirely human-generated, with no LLM involvement – except for the cover image rendered by Gemini. It will also be brief.

I am (still, perpetually) learning French, and I use Anki to help me. I recommend it for standard and non-standard flashcard uses. Recently, I asked ChatGPT to generate a CSV with phrases for the front and back of the cards, for example:

Front: I will be able to visit France next year.

Back: Je pourrai visiter la France l’année prochaine. (futur simple)

I found the pack useful, but a few days later (today), I asked for more than first-person perspectives, as well as additional tenses. I entered this prompt, which is what triggered this blog entry:

‘You’ rendered this for me. Might you generate a similar file that I can use to import to Anki that includes versions other than first-person singular? Future tenses would be helpful as well.

Notice the ‘you‘ in scare quotes – inverted commas.

It’s no secret that I am partial to the episodic selves of Galen Strawson in contrast with diachronic selves, which is to say that I don;t believe that the you or the I or yesterday – or even a second ago – is intrinsically the same person. I’ve explained this elsewhere, so I’ll leave it here.

As much as I feel this way, I don’t feel put off referring to myself as myself or I – me, myself, and I – or to you as you, but I am more consciously uncomfortable with addressing an LLM as you, as demonstrated.

I know full well that each prompt and response is to a different instance of the LLM – a different episode. Unless the LLM accesses an earlier conversation in memory – and I use this term loosely here, too – it would not retain context, and the absence of you-ness would be painfully obvious.

I don;t believe that the self of people is functionally any different. Obviously, humans are carbon-based rather than silicon-based, and storage and retrieval operate differently beyond the substrates, but the mechanism is metaphorically similar.

Does anyone else feel dis-ease in addressing an LLM in the second person? Or is it just me – or me?

On LLMs and Understanding

Posted in philosophy, podcastTagged , , , , , , , , , , , , , , , , Leave a comment

In this Substack essay, I explore the flawed nature of human comprehension, arguing that our understanding is often just a calculation of probability rather than an objective grasp of truth. By examining linguistic errors like misheard lyrics and his own struggles with French phonemes, Willis demonstrates that humans frequently hallucinate or misinterpret data based on what seems most plausible. He compares these mental lapses to the ‘hallucinations’ of Large Language Models, suggesting that the mistakes made by AI are fundamentally similar to our own. The author contends that we apply a double standard by granting humans the status of ‘understanding’ while dismissing machines for the exact same predictive behaviours. Ultimately, the text challenges the metaphysical assumptions of human exceptionalism, asserting that what we call ‘understanding’ is merely an external attribution rather than a unique internal state.

Essentially, I am not advocating the view that LLMs have consciousness and understanding; rather, I am arguing that the human versions of these are inflated with metaphysics.

The_Lossy_ChannelDownload

Summary of Some of My Philosophical Positions

Posted in philosophy, podcast, videoTagged , , , , , , , , , , , , , , , , , , , , , , , , , , , 5 Comments

The question is reasonable, although belief may already be doing more work than I would permit it under cross-examination. My writing is generally diagnostic rather than doctrinal. I am more interested in identifying the architectural assumptions beneath a dispute than in selecting a furnished room within it. Where philosophy offers a contest between established positions, I often suspect that the arena itself has been badly designed.

Still, repeated diagnoses eventually disclose a pattern. Certain commitments recur: suspicion of substance, resistance to metaphysical inflation, distrust of linguistic confidence, rejection of moral realism, scepticism toward persistent personal identity, and an insistence that mediation does not entail fabrication.

What follows is therefore neither a creed nor a completed system. It is a provisional map of where I presently stand.

The labels are intended as coordinates, not allegiances. Where no established term quite fits, I have used my own. This is partly unavoidable and partly evidence for the very linguistic difficulty I spend so much time diagnosing.

This content, including video and podcast summary, are of a longer Substack article on my philosophical positions with descriptions.

Audio: NotebookLM summary podcast of this topic.

AI Language Patterns

Posted in philosophy, podcast, videoTagged , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , Leave a comment
5–8 minutes

I love language, and I study it. I also use LLMs. I am not anti-AI, but neither am I prescriptively pro-AI. I just happen to find them useful.

Audio: NotebookLM summary podcast of this topic.

I also, however, find them to have quirks. I won’t regurgitate the supposed over-representation of delve — or em dashes, or lists of three, and so on. These are each, at once, true and hardly egregious – especially given that we were all schooled on the same English that now sits in their training data.

What I want to focus on is the cadence. I watched this video, and the historical information is interesting – in fact, much of the material on the channel is. A few shorts were recommended to me, I discovered more, I subscribed, as one does. But listen to the talk track.

Video: The Most Mispronounced Words in English and the History Behind Them – Airlearn Language Show

Whilst I may be mistaken, it sounds like an LLM-produced script. Perhaps the joke is on me, and the presenter herself is an AI avatar. It also sounds as though a de-LLM process was applied afterwards to obfuscate the patterns – whether by a human or by another model running some humaniser GPT.

Of course, there are other explanations. Perhaps the writer simply happens to write this way, and the LLMs merely encroach on it. Without a writing history, one can’t know for certain. And – I know this is true of me – exposure, dare I say overexposure, shapes how one writes.

I run almost everything I write through AI and ask for suggestions – and you needn’t even ask; it volunteers anyway. When the output looks better, I adopt it: a word, a phrasing, a more concise rendering, a supporting sentence or a setup. After a while, my writing has been nudged in that direction. I already used delve; now I am merely conscious of it. I used to use em dashes – look at my pre–2020 posts – and switched to en dashes purely to avoid the witch-hunt. In each case, the LLM Age has done something to my style.

So, back to the video. If you are familiar with long-form LLM output, you will notice how she supports her ideas and how she builds her asides – with a very particular pattern. I am not judging: as I say, I use these tools extensively. I only want to point out an observation.

BONUS: I don’t pronounce Wednesday as /wenz.day/. I pronounce it closer to /wedñz-day/, the tilde is rather on the /d/ or the space the /d/ occupies, but I placed the mark on the /n/ instead because there is no tilde-d glyph, and I wanted to represent a nasality the more typical pronunciation doesn’t have – but that I add.

Machine_CadenceDownload

From here on, I asked Claude to analyse my post (above) along with the video script. Pay special attention to its own cadence, which comports with the video.

Here, then, is the observation.

Watch how a point is made to land. The move is nearly always the same: state the naïve reading, snatch it away, and hand back the corrected version as a tidy little binary. It’s not random. There’s a reason. It wasn’t design. It accumulated. English spelling isn’t bad design. It’s no design at all. The shape does the work an argument would otherwise have to do. Negate, reverse, reveal – and the reveal feels like insight precisely because something was first taken away. It is the cheapest available route to the sensation of having been told something.

Notice, too, the reassurance travelling alongside it. There’s a reason. There’s always a reason. The promise of an explanation, delivered before the explanation arrives, so that you settle in and stop wondering whether one is coming. And where genuine argument would be laborious, the list stands in for it: Old English, Norse, Norman, French, Latin, scholarship, Dutch typesetters, the great vowel shift. Abundance, read as authority. Say enough true things in a row and the row itself begins to feel like a proof.

Then the asides. Almost every one is the same manoeuvre – the proper noun that renames itself the instant it appears. Woden, the chief god of the Anglo-Saxon pantheon. Latin, the prestige language of learning. A Quechua word from the Andes. Frigg being Odin’s wife. Nothing is permitted to sit unglossed for even a clause. This is what maximal legibility looks like once it hardens into reflex: a horror of the unexplained referent, a compulsion to footnote in real time. The second variety is the personality wink – and English being English, it kept both – character applied topically, like a balm.

And, so that you are never left to feel anything unsupervised, your reactions are narrated back to you in advance: if that feels insane to you, good, because it is. The response, pre-issued, so you may be spared the labour of having it. The whole is held together by a governing metaphor with a landing – fossil, photograph, museum – reactivated at the close (every mispronounced word is a crime scene) and finished with a note of absolution: you haven’t been mispronouncing English because you’re careless. It is very nearly a form one could fill in.

None of which proves a human didn’t write it.

And that is the part worth sitting with. Suppose no model ever touched this script. Suppose the writer arrived at the cadence honestly, by the same route I did – by reading enough of the stuff that the rhythm simply seeped in. Then the pattern is no longer evidence of authorship at all. It is evidence of a house style, and the house is one we have all quietly moved into. The tell no longer tells you a machine was here. It tells you the machine’s prosody has become the water.

I find that rather more unsettling than a mere ghost-writer would be. A ghost-writer you can dismiss. A cadence you have caught, like an accent, off the sheer ambient volume of the thing – that you carry home. And the video is, with some poetry, about exactly this: silent letters as fossils, sounds that mouths stopped making but that spellings preserved, layer upon historical layer that nobody ever cleaned up. The irony is that the script is laying down a fresh one as it speaks. Somewhere in the prose of the next decade there will be a stratum you can date almost to the year – ah, yes, written just after the models arrived – and it will be made of precisely these moves.

I switched to en dashes to dodge the witch-hunt. But the dash was never the point. You can strip every em dash from a sentence and leave the skeleton wholly intact, because the skeleton was never punctuation. It was the not-this-but-that, the aside that footnotes itself, the pile that mistakes its own length for weight. A humaniser can launder the vocabulary. It cannot, so far, launder the architecture. Which is how one reads a script scrubbed clean of every obvious tell and still hears, quite distinctly, the thing it was scrubbed to hide.

I am not judging. I write this way too, now – rather more than I would like. I only wanted to point it out. While I can still tell that I’m doing it.

The World’s Most Dangerous Idea?

Posted in philosophy, podcastTagged , , , , , , , , , , , , , Leave a comment
4–6 minutes

Am I the only one who can’t resist a massive eyeroll – and, let’s be honest: jaw-drop – what you hear transhumanism couched as evolution? To me, it incites a similar reaction to hearing people witter on about machine consciousness, but I’ll sideline that topic.

My objection is linguistic: transhumanism often borrows the prestige of evolution to describe what is more precisely technological mediation. The fact that a device is worn, implanted, or integrated into a body does not by itself move it from tool-use into biological descent. The offspring still inherits the organism, not the upgrade. Technology is not heritable.

Audio: NotebookLM summary podcast of this topic.

Consider rhinoplasty. Rhinoplasty changes the presented phenotype, not the inherited genotype. The child inherits the developmental instructions, not the parent’s post-surgical edit. Likewise, a neural implant, prosthetic limb, exoskeleton, gene-unrelated enhancement, or titanium jaw of techno-vanity may alter the lived organism, but it does not thereby alter the reproductive line. This is the category error: Acquired modification is mistaken for inherited transformation.

So, transhumanism often confuses the edited encounter-profile of an organism with the evolutionary alteration of the organismic lineage. The rhinoplasty case is good because it shows the absurdity without needing much apparatus. No one sane thinks a nose job rewrites the germline. Yet when the modification is sufficiently glamorous, especially when welded to futurist rhetoric and venture-capital incense, people suddenly start talking as if augmentation equals evolution.

A prosthesis is to evolution what rhinoplasty is to heredity: a modification of presentation, function, or encounter, not a transformation of descent. The confusion arises when the altered individual is mistaken for an altered lineage.

The question isn’t: Can transhumanism be defended under some broader theory of evolution, cultural inheritance, niche construction, or techno-biological co-development? Sure, it can.

But my claim is narrower: The term evolution is being made to do illicit rhetorical work when acquired, engineered, prosthetic, or embedded modifications are spoken of as though they were equivalent to heritable biological transformation.

The problem is that we are sliding from tool-use to augmentation to enhancement to transformation to evolution. This move isn’t illegal, but it should attract your attention.

At each step, something plausible is smuggled into something grander. By the end, an implanted device or technical prosthesis has been rhetorically promoted into ‘the next stage of human evolution’, when what has actually happened is that an organism’s encounter-capacities have been modified by artefactual mediation.

In the end, I’m not claiming that transhumanist practices can’t affect selection environments, social organisation, embodiment, or cultural transmission. What I am claiming is that calling such practices evolution without specifying the relevant register is a linguistic inflation. It trades on the authority of biological evolution while often describing technological mediation, individual augmentation, or cultural change.

In my book, more accurate terms might be augmentation, prosthesis, mediation, tool evolution, cultural inheritance, or niche modification, but these re so glamorous or click-baity. I might as well wear a hat and call myself post-human.

The_Evolution_Category_ErrorDownload

Addendum

I chatted up ChatGPT, so I am adding this response:

Exactly. The carried / worn / embedded distinction is mostly a boundary fetish, the sort of thing humans adore because skin looks like an ontological border and apparently everyone remains impressed by packaging.

An axe, spectacles, a pacemaker, a cochlear implant, a smartphone, and a neural implant all sit on the same broad continuum: external mediation of capacity. Some are held. Some are worn. Some are surgically installed. But unless the alteration becomes heritable, it is not biological evolution in the strict sense. It is phenotypic augmentation, affordance expansion, or environmental coupling.

The useful distinction is threefold:

  1. Biological evolution
    Heritable variation changes across generations. The organism-line changes.
  2. Cultural / technological evolution
    Tools, techniques, designs, institutions, and practices vary, are selected, retained, modified, and transmitted. Axe technology evolves. Writing evolves. Surgery evolves. The appalling human habit of calling every interface “intuitive” also evolves, though downward.
  3. Individual augmentation
    A particular body-system gains capacities through artefacts. This may change what the organism can do, but it does not by itself alter the inherited biological line.

Your axe example is clean because it exposes the smuggling. The axe helped alter human selection environments. It extended reach, force, food access, defence, craft, and perhaps social organisation. That may feed into gene-culture co-evolution over long scales. But the axe itself is not an evolved human trait. It is a technological mediator that participates in the evolutionary ecology.

The same holds for transhumanist implants. A retinal implant may alter one person’s perceptual affordance field. A brain-computer interface may alter one person’s control loop. A prosthetic limb may alter one person’s mobility. But unless such modifications become transmissible across generations, biologically or through some stable reproductive architecture, calling them “evolution” is loose metaphor masquerading as profundity. Very on-brand for techno-utopianism: rename engineering as destiny and hope the grant reviewers are sleepy.

From your MEOW angle, the better formulation is:

The relevant transformation is not evolutionary but mediational. Tools, prostheses, implants, and interfaces reconfigure encounter-events by altering the affordance structure through which an organism meets constraint. Whether the device is carried in the hand, worn on the face, implanted beneath the skin, or networked through the environment is secondary. The organism-world relation changes; the inherited organism does not necessarily change.

That also lets you avoid an overcorrection. Transhumanist enhancement is not irrelevant. It may be historically enormous. But its primary register is encounter, not inheritance; mediation, not mutation; technical lineage, not biological descent.

So the blunt line is:

Transhumanism mistakes the evolution of tools for the evolution of the tool-user.

Not always, not necessarily, but often enough that the slogan deserves being dragged into the alley and relieved of its metaphysical wallet.

Je m’accuse, on s’accuse

Posted in philosophy, podcastTagged , , , , , , , , , , , , , , , , , , , , , , , Leave a comment

The LLM Witch-Hunt and the Enemy Within

I recently read a piece arguing, with considerable sophistication, that LLMs represent an unprecedented psychological threat – that conversational systems operating at a planetary scale change the geometry of human susceptibility in ways that demand serious governance responses. The author wasn’t wrong about the effects. This isn’t the debate, but she was wrong about the story. The effects are real, and the narrative erected around them is the oldest displacement manoeuvre in the repertoire

Continued on Substack.

Audio: NotebookLM summary podcast of this topic.

How I Use AI in My Publication Workflow

Posted in philosophyTagged , , , , , , , , , , , , Leave a comment
5–8 minutes

This is not a philosophical post. Well, it’s about my personal philosophy of using LLMs and AI agents in my writing and publication workflow, which is a different thing. I’ll structure it as I might have done a music project back in the day, because that framing still makes more sense to me than anything the tech industry has come up with.

Audio: NotebookLM summary podcast of this topic.
NotebookLM Infographic on this topic.

Preproduction

Not all projects make it into production. Others were never intended to. But they all begin with at least a kernel of an idea — and some arrive fully formed, as if sprung from the head of Zeus, already wearing armour and looking for a fight.

Pre-ideating

What the hell is pre-ideating? I just made it up for this use case because that’s how I roll.

As I understand it, some people need help thinking of topics. This is not my problem. My problem is managing ideas rather than generating them. I have a backlog that will outlast me, so I don’t use this step. But it exists, and it’s probably the most widely discussed AI use case in creative circles: you prompt the model to suggest themes, genres, or concepts. Give me five ideas for a mystery novel. Or, if you’re feeling ambitious: Give me five ideas for a research paper in quantum physics. The model obliges. Whether what comes back constitutes an idea in any philosophically interesting sense is a question I’ll save for another day.

Ideating

This is where I usually enter the process, and the ideation takes shape in one or several different ways. The most common is simply a discussion – a sustained back-and-forth. A recent example: I was reading Judith Butler’s Gender Trouble and found myself with clarifying questions at every turn. Not because Butler is unclear, but because the implications kept ramifying in directions I wanted to follow. That extended dialogue – with ChatGPT in this instance – eventually became the philosophical core of Two Kings, currently stalled in Production.

Butler’s argument about incest taboos as foundational to broader regimes of sex and gender regulation gave me a narrative frame. The conversation helped me see what I actually thought about it, which is the more important thing. The LLM didn’t give me the idea. It gave me a sounding board patient enough to entertain the idea at two in the morning – it was actually two in the afternoon, but who’s looking?

Research

Another obvious use case, and one I use regularly. Continuing the Butler example: I asked about several feminist theorists she references, wanting to understand the lineage I was stepping into. But here’s a cleaner illustration. Writing as Ridley Park, I produced a novella, Sustenance, set in Iowa. I’ve visited Iowa several times, but I needed local flora and fauna for descriptive texture in certain scenes, so I asked

In the old days, I’d have gone to Google, Wikipedia, or I’d track down an Encyclopædia Britannica. The process is faster now, and the results are generally better for this kind of lateral, contextual research. For anything where accuracy is genuinely load-bearing, I verify. That’s not a criticism of the tool; it’s just basic epistemic hygiene.

Confirmation

Sometimes I have an idea and want to know whether someone’s already done it because I have no interest in reinventing wheels, and even less in reinventing them badly.

So I ask: Has anyone written X? What are the most significant treatments of Y? What typically comes back is a list of a dozen or more analogous sources. I review them and decide: does my idea still have independent purchase, or am I just writing a worse version of something that already exists? Sometimes I sharpen the idea in response. Sometimes I incorporate what I find, either to build on it or to identify where the existing literature is misframed, assumes too much, or has quietly imported the wrong ontological grammar. This last move is something of a professional tic.

Production

Drafting

I don’t use LLMs for full drafts. This is an obvious use case for those who do, particularly if the goal is volume – especially for the person who has already prompted for which genre currently has high demand and low representation on Amazon, and is now logically committed to producing it. That’s a coherent workflow – just not mine.

Edits and Revision

This I use often, and it’s probably where I get the most consistent value. After writing a passage or section, I feed it to one or more models with context already established — thesis statement, abstract, outline, supporting documents. What comes back varies: typographical errors, odd phrasings, unintentional repetitions (and, occasionally, new ones the model has helpfully introduced), suggested rewrites, observations about framing. I don’t treat any of this as instruction. I treat it as a second read from a reader who has no ego investment in agreeing with me – and yet obviously does. The important distinction is input versus output. I’m not asking it to write. I’m asking it to respond to what I’ve written.

Continuity

Are there gaps? Dropped threads? Promises made in chapter two that chapter seven has forgotten entirely? This is a genuinely useful mechanical check – the kind of thing that’s easy to miss when you’ve been inside a manuscript long enough to stop reading what’s actually there.

Flow

Do the scenes and chapters move well? Does the transition from one section to another feel like a logical step or an unannounced lurch? Useful, with the caveat that models have aesthetic preferences that don’t always align with mine, and I treat their flow suggestions accordingly.

Pacing

Is the pacing appropriate — both for the genre and for the particular piece? These are separate questions. A thriller has genre conventions around pace; a particular thriller might have reasons to subvert them. The model can flag where the pacing drifts; the judgement call about whether that’s a problem remains mine.

Postproduction

Formatting and Layout

I use AI for ideas about how to present content on the page: chapter opens, font choices, sizes, running headers, folios. This is design at the level of convention and taste rather than technical execution. I find it useful as a first pass — it surfaces options I might not have considered, which I then either adopt, adapt, or discard.

Cover Ideas

Thematic cover concepts, whether or not I ultimately outsource the art and creative work. I find this a productive way to articulate what the book is doing before I have to explain it to someone else.

How To

I use InDesign, Illustrator, and Photoshop with competence but not expertise. For specific technical tasks – how do I do this thing in InDesign — I ask. I also still use Google, YouTube, and the occasional book. These are not competing resources; they’re complementary ones, and which I reach for depends on what kind of answer I need.

Support and Maintenance

Marketing and Placement

Target markets, genre positioning, how to frame the work for audiences who didn’t watch it being assembled. This is a legitimate use case and one I engage with, even if marketing remains a word I say with a slight internal wince.

I also use platforms like ElevenLabs for audio, NotebookLM for podcast summaries and infographics, and Nano Banana or Midjourney for images.

Keywords and Descriptions

Adjacent to marketing but more administrative in character, the metadata layer that determines whether the work is findable by the people who would want it. Less interesting to think about than almost anything else in the process, and therefore an excellent candidate for AI assistance.

None of the above replaces the work. That’s the point. The writing is still the writing.

A Brief and Largely Accurate History of Punctuation

Posted in philosophy, podcastTagged , , , , , , , , , , Leave a comment
1–2 minutes

For most of human history, written Latin looked something like THISISASENTENCEABOUTPHILOSOPHYORWARYOUCHOOSE, and readers were simply expected to get on with it. And of course, in ALL CAPS. This was not considered a problem. The Romans were not known for their sensitivity to the needs of others.

The Romans did, briefly, experiment with the interpunct – a modest dot deployed between words, giving the reader something like THIS·IS·A·SENTENCE·ABOUT·PHILOSOPHY·OR·WAR·YOU·CHOOSE – before apparently deciding this was excessive hand-holding and abandoning it entirely. Punctuation’s first appearance in Western prose was thus also its first act of self-destruction. A precedent, as we shall see, that held.

Audio: NotebookLM summary podcast of this topic.

Relief came, eventually, from the most unlikely of sources: monks. Specifically, Irish and Anglo-Saxon monks in the 7th and 8th centuries, who were copying Latin texts they couldn’t actually read fluently, and who introduced spaces between words as a personal coping mechanism. Civilisation has strange bedfellows.

The comma, the full stop, and their assorted relatives arrived with the printing press – Aldus Manutius and the Venetian humanists essentially standardising the breath-marks of prose into something reproducible at scale. Punctuation became, in this period, the bureaucratisation of rhythm. A noble project. Mildly tyrannical in execution.

The em dash, meanwhile, had an entirely respectable career throughout the 18th and 19th centuries — a mark of genuine syntactic energy, used to interrupt, to pivot, to hold two thoughts in productive tension — before being left largely to the eccentric and the emphatic.

Then came the large language models. Within approximately eighteen months, the em dash was resurrected from the dead to become the default unit of thought, issuing them faster than Oprah Christmas giveaways. Every clause got one. Sometimes a sentence received two, bracketing a thought that required neither a bracket nor a thought. The em dash ceased to mean interruption and began to mean I am text generated at scale. Readers noticed. Then they mocked it. Then, following the immutable logic of cultural exhaustion, they stopped using it entirely. The em dash is now extinct — which is a shame, really.

The Author Did Not Write This

Posted in philosophy, podcastTagged , , , , , , , , , , , , , , , , , , , , , , Leave a comment
4–6 minutes

The LinkedIn consensus has spoken: if you used AI in the writing process, you are not the author. The position is stated with the confidence of someone who has never hired a ghostwriter, employed a research assistant, submitted to a heavy editor, or considered that the Gettysburg Address was almost certainly not written by Lincoln.

Image: I couldn’t not share this Midjourney 8.1 image. It may not have understood the assignment.
Audio: NotebookLM summary podcast of this topic.

Authorship has never been a production relation. It has always been an attribution relation — an institutionally stabilised answer to the question of which name the practice elects to put on the cover. These are not the same thing, and conflating them is the error from which every subsequent confusion proceeds.

The ghostwriter has existed as long as commercial publishing. The political speechwriter is so normalised that nobody considers it worth remarking. The celebrity memoir, the corporate thought-leadership piece, the attributed editorial — these are not edge cases or embarrassing exceptions. They are the normal operation of every writing-adjacent industry that has ever existed. The name on the cover has never reliably indicated the hands on the keyboard, and the industry has never seriously pretended otherwise. It has simply preferred not to discuss it at dinner.

AI changes the tool. It does not change the structure. The person who prompts, selects, curates, revises, and publishes is doing what commissioners of ghostwriters have always done. What has changed is that AI makes the mediation visible in a way that polite convention previously concealed. Visibility triggers the purity reflex. What presents itself as a defence of authentic authorship is a defence of a particular fiction — the Romantic author as solitary originating consciousness — that the industry never consistently believed and certainly never consistently practised.

The purity position also fails on its own terms before it gets started. Consider the spectrum of AI-assisted writing: a full draft submitted for light polish; a human argument substantially revised by AI; collaborative ideation followed by AI drafting; a kernel of an idea handed over for full execution. These are genuinely different in terms of human contribution. The zealot position requires a threshold somewhere on this spectrum below which authorship lapses. It never specifies where. More fatally, it has no means of verification. There is no external method of determining where on the spectrum any given piece of writing falls. The detector tools are probabilistic noise that disproportionately penalise competent prose. Any audit mechanism sophisticated enough to catch first-order evasion immediately generates a second-order workaround. The regress terminates only at continuous surveillance of the writing process — panoptical authorship as the logical endpoint of the position taken seriously.

NotebookLM Infographic on this topic.

Then there is the recursion problem, which the zealot never addresses because it is fatal. The stochastic parrot charge against AI — that it merely recombines absorbed linguistic patterns without genuine origination — describes with considerable accuracy what human cognition also does. The writer’s training data is the Dickens read at ten, the billboard absorbed on a commute, the argument overheard on public transit, the half-remembered essay that shaped a position without ever being consciously cited. The causal chain of any human idea disappears into an unauditable cognitive history. Genuine origination in the sense the purity position requires has never existed. The Romantic author was always a retrospective confabulation. Barthes said so in 1967. The industry nodded politely and continued invoicing.

What the zealot is defending is not authorship. It is a particular grammar of authorship — one that selects compositional origin as the threshold criterion, applies it selectively and unverifiably, and uses the resulting suspicion as a status boundary. It is guild behaviour dressed as principle, which is understandable as a response to a genuine economic threat but should not be mistaken for a philosophical position.

Authorship is the position a culture elects to stabilise after the work has already been produced through far messier means. It has always been thus. AI did not break the fiction. It just made the fiction harder to keep a straight face about.

The Rest of the Story

I’ve written about this before. I am not an AI apologist, but I am peeved by anti-LLM zealots, who clearly haven’t thought through their arguments.

I finished reading A.J. Ayer’s Language, Truth, and Logic, the part about Bertrand Russell’s claim about ‘The author of Waverley was Scotch‘. My brain latched onto authorship, and my emotional response was WTF? I have other problems with Russell and Ayer on this, but that’s a matter for another day.

To make my point, this page up to the ellipsis is the output of Claude after an extended dialogue with it and ChatGPT after I read Ayers, and something didn’t sit quite right. I am not ashamed to use LLMs in my authoring workflow and am not ashamed to mention it, as here. Almost all of these thoughts are mine. I’ve simply asked Claude to organise the output. It’s good enough to output as-is, and any edits would be trivial, so I won’t bother. I probably could have made the edits in as much time as it took to type this, but I’ve got nothing to hide. I’m just a human with access to technology circa 2026.