I commenced reading another LLM book recommendation, Piranesi. I share my thoughts on the first two chapters on my Substack.
Gemini Pro video representation of the sense of Piranesi.
Susanna Clarke’s Piranesi comes in seven parts. I’ve read two. It’s the second of four recommendations an LLM gave me after I fed it ten books and authors I liked enough to inflict on other people. I wrote about that experiment, and about the first book I chose from the list, Jacqueline Harpman’s I Who Have Never Known Men.
I’ll call the parts chapters, because that’s how I read them, and it feels more natural to me. Clarke isn’t here to object, so this unilateral move happens in absentia.
So far the book is mostly world-building. There is a protagonist, Piranesi, a name he’s given and doesn’t seem particularly convinced is his. A second character is known only as the Other and who visits twice a week, enlisting Piranesi’s help in researching the Great and Secret Knowledge, whatever that turns out to be. I’ve spent two chapters trying to decide whether the Other is a foil, a mentor, an antagonist or something else. Please stand by…
I Who Have Never Known Men by Jacqueline Harpman is neither a self-defence of heteronormativity nor a salacious lesbian romp. I finished reading this over the weekend – and again and again – but the Substack essay got out of hand.
The essay is a critique of the book, which I liked very much and recommend highly, but it had some authorial challenges I felt were addressable. My essay is very much critical theory rather than a literary critique.
I took notes in ChatGPT as I read it – par for the course – and I engaged in dialogue afterwards. I wrote the preamble, and then engaged Claude as ChatGPT and I traded ideas. I bounced the outline and some ideas off of Claude. It suggested that each section should rather be a chapter released as a series. In the end, it appears to have been right. Nonetheless, I tasked Claude with the final output.
As I was adding nuance and endnotes, Substack began straining under the weight, so I stopped before something got corrupted. My goal was to provide more context to English-only readers. I added more inline – among en dashes – but I decided to publish it as-is, perhaps editing these clarifications after the fact.
I had already decided there were limitations to this notion, but I feel it’s still useful. The way I see it, affordance instantiates with an encounter. Stairs don’t have affordances as attributes absent interaction. Generalised, affordance is a function where one needs something like this
A = {subject:object:index:context}
where the subject is the sentient or animate entity, let’s say the chimps in this image; the object is the encountered item, say the bench; the index is the affordance, say sitting or standing, and the context is a salience (or valence or attention) modifier.
In the case of sitting, I may not be tired or particularly interested in sitting. The bench still affords sitting, but I don’t instantiate the act.
Taking stairs as a more common example, if I am a typically abled human adult, stairs afford climbing (ascending and descending), but if I am relegated to a wheelchair, no such affordance relationship exists. This helps to explain why affordance isn’t an attribute of the stairs, but rather of the encounter.
Water to a non-swimmer doesn’t afford swimming any more than a tree affords climbing for a fish – and yet does for a chimp.
I posted this article on Substack: The Philosophers Are Here. Which Philosophers? Some companies are making a big deal about hiring philosophers and ethicists to help then with AI, especially AI companies. Isn’t that grand? My point is that companies only make decisions that they think will support rather than undermine their positions, so they are hiring philosophers sympathetic to their worldviews. They are not hiring, for example, me.
In fact, when I was a management consultant, one of my biggest challenges is that I don’t support capitalism, and I find much of business unethical. To add insult to injury, every year I had to take ethics training and take a test with an attestation that I have not witnessed any unethical behaviour. However, what they qualified as unethical behaviour was in fact illegal behaviour. They didn’t seem to understand the difference – or at least didn’t care very much.
Read the full article on Substack, or watch or listen to a summary on YouTube or Spotify.
Every year, apparently, somebody wins happiness. Recently it has often been Finland. The headlines arrive with the confidence of a football result: Finland is the happiest country in the world. One imagines the Finns receiving a small trophy, perhaps in a sauna.
The claim sounds empirical. It sounds as though something called happiness has been located, measured, aggregated, ranked and compared across national borders. There are, of course, surveys behind such rankings, and those surveys may be perfectly competent at measuring what they actually measure: reported life satisfaction, evaluations of present circumstances, perceived social support, security, income, institutional trust and related variables.
But that is already quite different from saying that Finns possess more happiness than everybody else.
Salience, attention, valence and the construction of disposition
Today I live about 90 minutes out of Boston by car on a good day – and 2 hours on a return trip because I live in Cape Cod, a tourist area, so I tend to have to compete with traffic. Today, it took over 2 and a half hours, a Friday afternoon, trapped among boat and beach seekers. I don’t drive, so I require chauffeurs to get me here and there. Today I driving with a bloke for whom traffic possesses unusual psychological gravity. He seems to have traffic and weather on the brain – and don’t get me started on crime, national security, and petrol prices.
Audio: NotebookLM summary podcast of this topic.
I’ll spare the reader the trip up, which was fairly easy, despite traffic-related side comments whenever inappropriate. On the way back and at the first patch of congestion, he harrumphed and announced that we would certainly be stuck in traffic for the entire journey. I checked the Google Maps. Traffic was intermittent, and I told him so.
Soon afterwards, the road cleared. We travelled freely for miles – no traffic to report – but when we later encountered another queue, however, he declared that his prediction had been vindicated. On arrival, he told others that we had been stuck in traffic ‘the whole way’.
Two people; the same reality; different experiences. I got me thinking – and chatting with ChatGPT because the driver was focused on driving – and of course the misery of traffic.
Here’s the rest of the story: The traffic didn’t occupy the whole journey. To him, it occupied the portions of the journey that counted. The clear road was perceptually available but apparently lacked evidential weight. It was experienced without being recruited into the eventual account.
Noticing the divergence between out perceptions, I tried to reconcile. First, I considered filing this under selective attention and confirmation bias, that great psychological lost-property office into which every untidy inference is now deposited, but this doesn’t feel quite right; it doesn’t seem to capture the essence.
Before a belief can be confirmed by evidence, something must first become evidence, right? So, the more interesting question isn’t merely why this geezer interpreted the journey pessimistically, but why congestion became salient whilst free movement receded into the background. What determines which parts of experience become psychologically substantial?
Here’s how I see it:
I tried to view this through lenses of salience, attention and valence. First, a quick refresher:
Salience concerns what stands out, demands processing or acquires significance.
Attention concerns the allocation and maintenance of cognitive resources.
Valence concerns the positive or negative affective character assigned to something.
Traffic may be salient because it frustrates an aim. Its negative valence may help capture attention. Sustained attention may then produce richer encoding, making the traffic easier to recall. Later recollection may strengthen the prior belief that journeys are usually obstructed, which may increase vigilance for traffic during the next journey. No single element need rule the others.
The relation is recursive:
disposition shapes salience
salience directs attention
attention affects encoding
valence influences interpretation
memory reconstructs the event
reconstruction reinforces disposition
And even this is too linear, but here we are. Attention can alter valence. Bodily arousal can alter salience. Memory can modify present interpretation. Expectations can affect what is noticed before conscious appraisal has properly begun. The arrows run in both directions, and probably sideways as well.
The result isn’t simply a biased opinion about a neutral world. It’s a patterned encounter wherein some features arrive already amplified and others scarcely become legible.
Pessimism: Disposition as an epistemic filter
This got my pondering mental health issue. I’m not a psychologist, and I discount the discipline quite heavily, but I wanted to consider the notion of disposition. How can two people quite literally in the same vehicle on the same journey view the world so differently. Of course, we are all carved by our experiences, but this is a reciprocating function: the more we experience the world positively or negatively, the more the world will reveal itself as such. Confirmation bias drives the nails into the coffin.
Let’s look at the down side: pessimism. I don’t mean some clinical depression, but instead of seeing the world with no glasses, and rose-coloured spectacle, the world is view with dark glasses – I’m envisioning a welder’s visor. One could stare at the sun and proclaim it’s overcast. Eyeore on steroids.
A pessimistic disposition might usually described as a tendency to expect bad outcomes. And whilst this may be true, it’s incomplete. Pessimism may also affect what is noticed, how long attention remains with it, how it is interpreted, and what survives into later recall. This is my thesis – or at least my point. The unpleasant becomes an event. The satisfactory becomes the absence of an event.
Congestion is noticed; free movement is merely used. An insult is retained; an ordinary kindness passes as social administration. A failure requires explanation; success is dismissed as luck, temporary relief or an exception that proves nothing.
This asymmetry can make a disposition appear empirically vindicated. The pessimist remembers abundant evidence for pessimism because pessimistically congruent events were more likely to become salient, receive attention and survive reconstruction. The conclusion is then presented as realism:
I don’t expect things to go badly because I am pessimistic.
I am pessimistic because experience shows that things go badly.
Yet ‘experience‘ here isn’t an untouched archive. It’s already been selected, weighted and narrated.
Research on depression is compatible with this more interactive account. Depression and vulnerability to depression have been associated not only with increased processing of negative material but also with diminished positive biases across attention, interpretation, memory and self-referential thought. The pattern is therefore not merely an excess of negativity. It may also involve a failure of positive information to acquire or retain ordinary psychological weight.
A negatively valenced event may dominate because it attracts unusually strong attention. But an apparently negative world might also arise because positive events are weakly encoded, rapidly discounted or treated as anomalous. The same remembered landscape could therefore emerge through different cognitive routes.
Before I continue, a disclaimer: I am not trying to claim that my perception is correct and his isn’t. I’m merely asking how we end up with distinct narratives. It reminds me of the beginning of Woody Allen’s Annie Hall, where a couple are speaking to counsellors. A therapist asks about their love life:
Alvy Singer (him) and Annie Hall (her) are seeing their therapists at the same time on a split screen
Alvy’s Therapist: How often do you sleep together?
Annie’s Therapist: Do you have sex often?
Alvy: [lamenting] Hardly ever. Maybe three times a week.
Annie: [annoyed] Constantly. I’d say three times a week.
The interesting think here, is that both are disappointed for the same event one because sex is over-indexed and the other because it is under-indexed.
Moving on…
Anxiety and the world as warning
Anxiety offers a related configuration. Threatening material tends to capture attention more strongly in anxious populations, although effect sizes, experimental measures and the direction of attention over time are not perfectly uniform. Some people may orient rapidly towards threat; others may struggle to disengage; still others may shift from vigilance to avoidance. The bureaucracy of diagnosis naturally prefers a single box, whilst cognition persists in behaving like a process.
Interpretation and memory may then participate in the same system. An ambiguous expression becomes disapproval. A delayed reply becomes rejection. A bodily sensation becomes evidence of illness. Each interpretation increases anxiety; increased anxiety makes threat more salient; heightened salience provides further material for threatening interpretation.
Research on anxiety increasingly examines these as interacting biases rather than sealed mechanisms. Attention, interpretation and memory can be associated with one another, though the evidence does not support a single simple causal sequence in every case.
The important point is not that the danger is imaginary. Sometimes the road really is congested. Sometimes the delayed reply does indicate trouble. Sometimes the body is ill. The question is how individual events acquire representative force and how readily contrary events are permitted to revise the emerging model.
When significance itself becomes unstable
In some accounts of psychosis, salience plays a still more direct role. The aberrant-salience hypothesis proposes that neutral or incidental stimuli may acquire excessive significance, leaving the person to explain why ordinary events feel unusually charged. A glance, coincidence, headline or passing remark may become loaded with personal meaning.
This is not a complete theory of psychosis, nor should it be paraded as one. Predictive-processing accounts differ over whether particular symptoms arise from weakened prior expectations, excessively rigid priors, altered prediction-error signalling, or different disturbances at different levels of processing. Hallucinations and delusions may not share one uniform mechanism merely because diagnostic manuals house them in neighbouring rooms.
Still, the idea illustrates the broader point. Salience does not merely determine which established facts receive attention. It participates in determining what feels fact-like, connected or worthy of explanation.
Once significance has been assigned, interpretation begins its work. Once interpreted, the event may be more readily remembered. Once remembered, it joins the accumulating case for the interpretation that helped produce it.
Not merely ‘in the head’
None of this means that dispositions are autonomous cognitive errors floating above the world, and I reiterate that I am not a medical professional. Hell, I’m barely a professional. Salience is affected by history. Someone who has repeatedly encountered danger may reasonably become more attentive to possible danger. Poverty, discrimination, violence, illness and insecurity do not merely supply ‘negative thoughts’; they alter the statistical environment in which attention and expectation develop. What appears clinically as excessive vigilance may have been learned under conditions in which vigilance was adaptive.
Nor is salience purely conceptual. Sleep, stress, hormones, medication, pain and physiological arousal may all affect what attracts attention and what is retained. Experimental research indicates that arousal and valence can make distinct contributions to emotional memory, another reason not to compress the process into the nursery vocabulary of ‘positive’ and ‘negative thinking’. The world acts upon the disposition, and the disposition acts upon the encountered world. Neither term stands outside the relation.
This also guards against the moral obscenity of telling distressed people that they have simply chosen the wrong interpretation. A recursive process may be self-maintaining without being voluntary. One doesn’t escape it merely by deciding to notice flowers, practise gratitude and purchase a notebook with an inspirational slogan embossed upon it. Sorry, pop psychology and psychological self help industry
From bias to rigidity
Everyone relies upon selective attention. Everyone remembers unevenly. Everyone encounters a world already organised by concern, expectation and affect. Complete neutrality would not be enlightenment; it would be cognitive paralysis.
The relevant distinction may therefore not be between biased and unbiased minds. It may be between systems that remain revisable and systems that become rigidly self-confirming.
A disposition becomes pathological, perhaps, not merely when it selects, but when selection becomes pervasive; not merely when it assigns valence, but when valence governs what can count as evidence; not merely when memory reconstructs, but when reconstruction repeatedly erases whatever might disturb the governing account.
Even then, ‘pathology’ isn’t a property of one mechanism. It may emerge from the interaction of degree, duration, distress, social conditions, bodily states, functional consequences and the available means of correction. The traffic anecdote isn’t a diagnosis. It’s an anecdote, a small model of how a world can be cognitively weighted.
The bloke didn’t invent the congestion. He encountered real traffic. But the congestion became representative, whilst the open road became incidental. Later, memory converted this inequality of weight into an equality of duration: traffic here and there became traffic everywhere.
Perhaps certain dispositions operate similarly. They do not simply add a judgement to experience after the fact. They influence which experiences become prominent, which are sustained by attention, which acquire positive or negative force, and which are later recruited into the story of what the world is like. The question is therefore not merely whether a person sees reality accurately. It is which parts of reality become visible enough to count.
A final word
My ex-wife used to complain how we never did anything – and like the Annie Hall story, I’d say we did too much – so every week or two she’d want to do something, and her refrain was always, ‘Let’s do it. We never do anything.’ One can imagine, that this was not a sustainable model – perhaps as evidenced by the ex prefix. And so not to and this as a therapy session, I’ll leave it here.
I’ve shared another article on Substack about how the AI-authorship debate is bollox.
Audio: NotebookLM summary podcast of this topic.
The LLM-versus-human debate has been raging for several years now, and there seem to be two principal camps: human exceptionalists and agnostics. I suppose there may also be a small, misanthropic faction, but its members are either remarkably quiet or indistinguishable from everyone else online. If you are out there, do raise your voice.
If you want to witter on about other negative aspects of AI, have at it, but the authorship debate is weak tea.
I shared another Substack post, but I want to expand it a bit here. On the Substack version, I conspicuously didn’t name certain obvious names, but here, I shall. I also left out a reference to Plato’s Republic.
The current American regime is built on meritocracy, and – this is the part everyone misses – it would not deny the charge. It rather likes the word. It uses the reinterpreted version, the laundered one my other essay was about, in which merit means precisely whatever the sovereign has decided it should mean this morning. Accuse them of meritocracy, and they will thank you for noticing.
Take the man at the centre of it. Whether Donald Trump is meritorious is not, in the end, a question about Donald Trump. It is a question about which ontological grammar you happen to be speaking, because the grammar constitutes the object before a single fact is admitted into evidence. This is the whole Language Insufficiency point in miniature: the frame quietly does the work the facts are later congratulated for.
To Cohort A, he is a charismatic and self-made businessman. The six bankruptcies are not failures but proof of a shrewd operator who knows how to play the game and walk away intact – power-mind, killer instinct, the art of the deal. That he has a decades-long habit of not paying his contractors, and an even longer one of not paying his tax, does not indict him; it endears him. It makes him the everyman who beat the system the rest of us merely resent. A working-class hero, if you can manage it, descending on a golden escalator.
To Cohort B – and I’ll confess my membership, since honesty is cheaper than the pretence of the view from nowhere – he is a less likeable Forrest Gump: a man forever arriving at the centre of history without ever quite grasping how he got there, weaponising a charisma that works only on the already-mesmerised, and rebuilding the world with himself at the origin. He has a certain gravity. It attracts sycophants and rewards them, which is the only merit test his court reliably administers.
Same man. Two grammars. The ‘merit’ resides in neither the man nor the record; it is assigned by the frame and then invoiced to the evidence. Which is exactly the thesis of the parent essay, now wearing a name tag.
The downward stroke, said plainly
The parent essay left the reverse operation as a coy ‘phrase in current circulation’. I’ll say it here without the euphemism: DEI hire. The move is to take career civil servants – people with decades of demonstrable domain competence, whatever their private politics – and broad-brush them as diversity appointments, which is to say, unqualified by definition, arrived by an unauthorised route. The vacancies then fill with those who have pledged fealty and displayed the one indispensable qualification: a flexible integrity. The competent are marked down for being the wrong sort; the pliant are marked up for being biddable. ‘Incompetent’ has been quietly reissued to mean not ours, and ‘qualified’ to mean loyal. Same machine, thrown into reverse – now with a name engraved on the lever.
The cream, and the dimension it rose along
Which brings me, as these things do, to Plato. The Substack piece used his ship of state – the crew who master the mutiny rather than the sea. Here I want the other half of the same complaint: the comfortable conviction that the meritorious rise to the top like cream.
They do rise. That was never in dispute. What the metaphor conceals is the dimension along which they rose. These people are genuinely competent – at working the system, at the acquisition and retention of position – which is a wholly different skill from governing, and precisely the one we so conveniently assume we were measuring. Cream rises; so does scum; the trope declines to specify which, and physics is no help at all.
This is the standing problem of republicanism – and no, before anyone lunges for the comment box, I do not mean the Republican Party, which has its own well-stocked catalogue of failures, as do the Democrats, who would be unwise to feel smug reading this. Nor is any of it uniquely American. If you are somewhere else entirely, enjoying the spectacle from a safe distance, do put the popcorn down. The mechanism by which a society mistakes competence-at-ascending for competence-at-ruling is not a national defect. It is a design flaw in selection itself, and your country installed it too.
I will not, on this occasion, start in on the failings of democracy as a system. Been there; done that. There are only so many sacred cows a man can tip before lunch.
The question, once more
‘Meritocracy’ survives all of this the way it survives everything – by never once being asked the only question that matters: merit at what, judged by whom, to whose benefit? Put it to the present court, and the answer is almost embarrassingly legible. Which is, of course, exactly why the question is never put.
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?
