Reading all the books

On machine learning, Mescaline and the pitfalls of a wooden dildo

Actual picture of me, trying to order my thoughts on machine learning.

This week I wanted to write about Joel — my Joel, this Joel —’s new job and going to see Hamnet, the play. As you may have picked up from previous posts, he’s a software engineer of very respectable pedigree and his new job has something to do with SEO.

I wasn’t initially clear how those two things were connected but it turns out what connects them is a little-known phrase in Aldous Huxley’s 1954 book The Doors of Perception.

The phrase is this:

The consciousness of having read all the books.

I jotted this down years ago and it stuck with me, albeit now divorced from context.

A brief recap: Doors of Perception is Huxley’s famous (less famous than Brave New World but still famous) book documenting his (and his wife’s) mescaline trips and what it did to alter (heighten?) their perceptions of consciousness. I seem to recall he used this particular phrase semi-mockingly, referring to people hyper-aware of their own cleverness — their consciousness of having read all the books — in juxtaposition to his own earthier mescaline-paved path to enlightenment.

I remember a party in Uni in the depths of winter 2004 in a ratty basement off the Cowley Road. People were huddled outside the back door, smoking and stamping their feet, shouting, arguing, roaring with laughter. When I asked what had happened, it turned out that someone’s new boyfriend had committed a massive social transgression.

“He corrected Tim’s grammar. We threw him out”.

I needed the specifics.

“What did he correct?”

Everyone was too drunk to remember. Something about “ain’t”.

“He corrected him twice. We escorted him out in a headlock.”

“Oh but he was right. Ain’t isn’t a word, Tim. You should know better.”

“Fuck off Jill, or I’ll throw you out next.”

Now, Tim was of course right. Nitpicky pompous little shits who correct people’s grammar at 2am on the Cowley Road deserve to get chucked out of the party.

That might have been what Huxley meant by “the consciousness of having read all the books”.

But it’s not what I mean.

I’ll try to explain.

Ted Gioia

recently published a piece about how he made reading books — and not just any books, but the hard books — his life’s work. He talked about aiming for the worthy books, recording what he’s read, identifying gaps where there are still things to be read.

I identify with this (as I suspect do most Substackers) and, while my reading lists are *much* less well-organised than Ted’s, I do it too.

See what I mean?

This is an excerpt from The Notebooks, circa 2007. The year after I finished University, deep in my I-won’t-go-work-in-the-City phase, protest-travelling by train from Riga to Chengdu, pre-Kindle¹ and devouring anything I found on hostel shelves that had the Penguin classic tinge to it (and turning my nose up at anything “lowbrow”/holiday romance/true crime etc).

This is a bit of a digression but my point remains — I hard identify with striving towards “the consciousness of having read all the books”. This is because, and I have written this before, when we read, we are seeing through another set of eyes. The more we read, the more sets of experiences — the more consciousness? — we accrue.

The “consciousness of having read all the books” is exactly that: consciousness, imbued with the perspective and experience of thousands of people — probably hundreds of thousands if you’re as prodigious a reader as Ted.

“If books were my drug, I always was taking the big intense dose that offered the greatest out-of-body experience.” (Ted Gioia)

So, it’s not just the reading of all the books. It’s how we, from our unique body with its unique mind, step into another body (momentarily) and relate to or pick apart or take onboard whatever it is we read.

That act of stepping out of your body and hearing someone else’s voice in your head, describing things you’ve never seen with your own eyes, is powerfully transformative. Books are a way of knowing beyond that which can be seen, felt or experienced in one circumscribed body.

The native replied; when did he (naming myself) see a soul go to Heaven or Hell?

The Sachim again replied: He hath books and writings, and one which God himself made, concerning mens souls, and therefore may well know more than we that have none..²

It’s pretty obvious that this innate power of the written word has been hijacked through the ages, as here recorded by a pastor in New England in the early seventeenth century.

Where I’m getting to is that the consciousness of having read all the books is unique to us, to humans. If an AI programme — ChatGPT, say — has read all the books, it has done so without the consciousness of having that out-of-body experience.

Because it doesn’t have a body and it doesn’t have experiences. Duh.

Back to Joel’s new job. With a new job comes lots of new things to learn about.

As a result, he is reading a lot about the nuts and bolts — and actual mathematics — behind machine learning.

He is also talking to me a lot about the nuts and bolts — and actual mathematics — behind machine learning.

I understand maybe two-fifths of what he says about large language models, so forgive me if this isn’t quite right (and please don’t @me to tell me more about how vectors work) but this is roughly how our conversation goes:

“Basically each word gets expressed as a vector to represent its meaning, so that it can be compared to other words. So, like, a really basic example is ‘King’ minus ‘Man’ plus ‘Woman’ equals ‘Queen’.”

I am — instantly, viscerally — really not into this parcelling up of meaning.

“But… ” I am really not into this and struggling to explain why. “The computer is freezing the meaning of words.”

He is very patient with my cartoonish inability to grasp what he’s saying.

“No, there’s no “freezing”. The programme will go on learning new meanings as the word evolves.”

“No! No. That’s not what I mean.” I am panicky and incoherent. “The computer is deciding what the “right” meaning is. That’s fine for some words, maybe. Maybe easy words, where the stakes aren’t that high, and you can say, Cow, yes or Cow, no. But what about words that mean one thing to one person and an entirely different thing to another? What about words that carry multiple meanings? The computer picks a meaning and it scrubs out any uncertainty. It dissolves the need for any reader-managed resolution of an ambiguous (and perhaps intentionally so) word selection.”

“But that’s not how it works. The programme isn’t deciding meaning; it’s not, like, programmed with a dictionary of meanings. It just knows what words are likely to go with other words, and what other words might in turn go with the next word, and so on. It’s just like a really smart person that reads lots of books. Except it can read the entire Internet. It’s predicting strings of text based on everything it’s read — and it has read everything.”

“That must still mean at some point that the computer is making a decision on what the most common — and therefore right — meaning is?”

We argued in this vein for quite some time; me, righteous, indignant, certain as certain can be that no machine can match my understanding of a written text and him, well, less certain.

That’s because words are important to me. If you’re reading this, I expect they’re important to you too.

I’ve tried to explain it before, in the context (weirdly) of what particularly winds me up about Minecraft. The chaotic spectrum of existence reduced down to its most basic form: “sheep”, “gold”, “enemy”. Black, white and all manner of shades of grey — provided it is Grey 3 or Grey 27 or one of the other officially-sanctioned shades of grey that the computer recognises.

It’s everything I loathe about the world now — restaurants, services — all pared down by a private equity accountant to be visible on a balance sheet, everything in its own row, with a ready value to ascribe.

And now it feels like they’re doing it to my precious words. It’s like Digitopolis has taken over Dictionopolis³.

It turns out that I am not alone in this.

Most people have a hunch—it’s no better than that, really—that no robot (made of silicon and metal and plastic, etc.) could ever be conscious the way we humans are. There is something about our living, breathing, organic bodies and brains that is necessary for consciousness. .. and maybe these people are right. But now that we have some insight into the way our bodies and brains can themselves be seen as made of robots made of robots made of robots, and so on, all the way down to below the neuron level where the motor proteins and other nanobots trudge along making the whole system work, we can see that maybe that hunch is just an artefact of impoverished imagination.

From Daniel C. Dennett, Intuition Pumps and Other Tools for Thinking.

Here we are back to consciousness again.

And I have been called out pretty comprehensively for my naive “hunch” that there’s something about us that no machine can replicate.

But still I am unmoved.

So what is it?

We went to see Hamnet, the new play put on by the RSC based on Maggie O’Farrell’s book of the same name last Friday night. I loved the book — seriously, who didn’t? — and I was very interested to see how it would translate to stage.

At the interval, Joel and I went off to pick up our pre-ordered drinks for the rest of our mates.

On the way, a group of women were discussing the play. As I passed, I heard one of them squeal at another:

“You haven’t read the book?!”

At the drinks counter, there was a problem, which Joel immediately spotted. The pre-ordered drinks were in lovely civilised glasses, not the plastic ones allowed in the theatre proper. The problem was that we wouldn’t be able to take glass back into the play.

We still had our old plastic cups but decanting our new drinks into them was like one of those “how do you get the fox, the sheep and the cabbage across the river” problems.

One of the old cups had an inch of water in the bottom so I thrust a napkin in to try to dab it up (which didn’t work; too much water), just as Joel suggested to drink it.

“You drink it.”

“No, you drink it.”

“No, it’s had a napkin in it.”

When we got back with the drinks, Bex and Helen were discussing a recent article in the Guardian about a prehistoric dildo (actually: Roman and actually: wooden.)

I sipped my napkin-flavoured gin in silence, thinking: splinters.

Talk turned to the play and whether we were liking it. I said I was, but found it a bit … spare. We agreed that what was missing was the description, the thoughts, the staggering narrative powers of Maggie O’Farrell. All that colour around the edges of what is at root a fairly simple story arc. The playwright Lolita Chakrabarti had one hell of a tough job: to take the meat of this book — a book so loaded with powerful descriptive prose and, let’s face it, a not-overly-complex storyline — and scrape and sieve it into an hour and a half of continuous, compelling dialogue. No mean feat.

Helen, it transpired, hadn’t read the book.

Like clockwork, I squealed:

“You haven’t read the book?!”

Then someone asked Joel what books he likes reading.

“Well, I’m currently reading a book about discrete maths.”

“Ooh!” Helen is a maths teacher; she is excited.

Then the lights dimmed and the play resumed.

There’s a line in the play Hamnet that I don’t remember from the book Hamnet. It might be there, but I couldn’t find it.

It’s in a scene where William Shakespeare is stage-directing the well-known Elizabethan actors, Will Kemp and Richard Burbage.

Will Kemp ad-libs a line, brushes it off as unimportant, then tries to justify the changed word to the famous playwright. Shakespeare, insisting on his own words, and his own words exactly, says (and I paraphrase with no irony whatsoever):

“You can’t imagine how much care and attention goes into choosing every word. Every single syllable.”

That isn’t in the book. I think it must be a nod from the playwright who had such a tough job converting this book to play, but any writer will recognise it. Shakespeare would certainly have recognised it. It is the acknowledgement that when we write whatever we write, all the words — every single word and all of its knowable and unknowable parts — matter.

And that is because of how they make the reader feel.

My line manager at work sends back every document she looks at that starts with Dear Sirs. She says it is the hill she will die on because every word matters.

Lester Freeman in the Wire said all the pieces count.

I think the Wire is one of the greatest novels ever written. I think a wooden dildo will give you splinters and I think you cannot spreadsheet experiences. That is the hill I will die on.

A machine might know a lot about how a thing feels. It might have read everything that’s ever been written about how a thing feels. But reading isn’t how we learn about how things feel. You don’t learn about being in love by reading about it. Writing often means the most to us when it sparks a memory of a real lived experience. Reading about being in love can make no sense at all — until you’ve been in love.

“We can pool information about experiences, but never the experiences themselves.” (Huxley, The Doors of Perception)

We can’t pool an experience. We can read about it, sure. If it’s well-written, we might feel like we experienced it.

But we didn’t. And ChatGPT sure as fuck didn’t either.

It might have read all the books, but it doesn’t have the consciousness of having read all the books. If anything, it has the unconsciousness of having read all the books.

Now, I know this is a straw man argument. No one is making the case today that ChatGPT is in any way conscious.

But the point is that, someday, with some other iteration, they might.

And I accept that.

I accept that, when computers can talk to each other about the things that they’ve read, compare notes about what their reading meant to them and how it relates to their own unique and personal experiences, then we may have recreated consciousness.

When they talk amongst themselves.. especially of Books and Letters, they will end thus: Manittôwock, ‘They Are Gods’ ⁴

Basically, when they set up their own Substack.

1. In fact, the very first Kindle was launched — and instantly sold out — in November 2007 while I was travelling. I remember hearing about it and thinking, well fuck. That would have been handy.

2. Roger Williams, A key into the language of America, 1643, in English Linguistics 1500-1800, ed.R.C. Alston, No. 299, The Scholar Press Limited, Menston, England, 1971, p.129

3. See: The Phantom Tollbooth, a.k.a. one of the greatest books ever written.

4. R. Williams, p.127