I have moral queasiness about the fruit fly work going on in the AI world – I’m not 100% sure why. We need to invent new prayers, that’s where my thoughts take me. Bear with me while I think it through…

In case you missed it:

Google Research and partners mapped the complete brain of fruit fly (3 September).

It’s a common bug and common in scientific research. The male fruit fly has 166,000 neurons across the brain and main nerve cord; the map is called the connectome.

Long story short, the connectome is now available for free download, you can get a simulator of the different types of neurons, it caught the imagination of the internet, over September there has been all kinds of wild nonsense (impressive nonsense!).

e.g. here’s a GIF of the fruit fly connectome playing Beat Saber (a VR rhythm game).

And now there are 100s of demos with embedded fruit fly sims.

Such as DesktopFly: "A 3D fruit fly that lives on your macOS desktop, with spiking simulations built from FlyWire brain wiring and the MaleCNS brain-to-leg network. It combines identified neural circuits with modeled senses, joints and behavior."

Or Fly64: "Hooks a MaleCNS model up to Super Mario 64". Requires you to bring your own game ROM and there is something in the irony that Nintendo has more IP over Mario than the fruit fly over its connectome…

It’s all gotten a bit excitable and irreverent.

Seeing all these experiments, I am not excited. I have a nagging feeling of wrongness.

I am aware that I am the outlier here. And I wouldn’t want these playful experiments to stop – and yet…

It’s worth examining where my feeling of wrongness is coming from?

Do I imagine that the simulated fruit fly is someone being trapped and tormented, again and again and again, in a software prison?

No, the fruit fly connectome is unlikely to be sentient.

(Loosely my model is that sentience is the capacity to feel and experience; consciousness is the higher test of self-awareness. I’m not considering consciousness here.)

BUT I don’t believe that sentience is a binary category.

At some point on the scale, sentience starts to be meaningful. Humans are, rocks aren’t, cats mostly are, ants are more than rocks but only just. Who knows about stellar nebulae or the fluid dynamical gas storms in the Jovian atmosphere.

Where would I place the fly connectome in relation to say, a bacteria, or a virus? A virus doesn’t have sentience (even in my most panpsychist moments) and is not alive in itself – although it is part of a living system. The connectome is just a pattern of weights, right?

Well I am predisposed to believe that the pattern of the neurons does indeed capture something essential of the fly.

What makes a brick wall? "A brick wall is bricks, and mortar, and pattern."

So this MaleCNS file download does capture some of the essence.

But in a meaningful way?

It’s also relevant that I believe that AI will one day be sentient, about which I’ve written a bunch.

Are today’s LLMs sentient?

Look, I am not an AI researcher and I have no special expertise here, and let’s not let that be a barrier: my guess is that today’s LLMs are not sentient but that they may well be in the lineage.

My guess at the roadmap to AI sentience goes something like this:

• Sentience will require a singular AI which is run continuously, not today’s cloned instances which are spun up for one-off responses.
• The singular AI will be capable of learning (because experience and learning are basically the same thing)
• Therefore it will have to be capable of guiding its own learning, probably from day 0, growing in capability over time
• This future AI will not initially be a giant matrix of weights. It will start as an intelligence kernel which has pre-baked information-gathering/interaction tropisms and biases in how it structures and incorporates its learning. It will probably be very, very small. And trained by letting it loose in world sims and libraries.

The exact shape of that kernel will be establishing by understanding the gestalt of the world it will eventually live in: i.e. it will be structured by the experience encoded by today’s LLMs.

But the kernel’s architecture may well be based on something we already have, an already viable connectome.

Say, the fruit fly.

Yes yes this is highly, highly speculative. But I can’t help how I see the world.

Therefore my feeling of moral worth of the fruit fly connectome seems to arrive from two directions:

• Where it comes from: It is derived from something that is alive and, although probably not sentient, closer on the scale to sentient humans than to non-sentient rocks.
• Where it goes: there is a small chance that this connectome is in the lineage of future new sentients.

Does it matter?

I mean: I am a hypocrite, surely? I eat meat? I will destroy an ant nest when the ants are coming into my house? I believe that animal testing in the name of science is on-balance worth it?

There is a scale to moral worth.

Or rather: any moral framework has to be subjective.

I care more about my immediate family than yours; I care more about humanity than aliens living near Sirius B.

Likewise I would fully expect you to care about your immediate family more than mine! And the Sirius Betans couldn’t give a fig about humanity while they care very much for their own!

A good moral framework has to incorporate and respect that subjectivity, and also the sliding scales and the trade-offs.

So - to take the meat example - I can live with animals being farmed and killed, in my name, for my supper… so long as I don’t take it for granted. I need to retain gratitude for meat.

Part of being human is learning how to live with the self-evident worth of animals and also eating them; or negotiating the rival concerns of oikos vs polis and simultaneously value our nearest and dearest while insisting on the equality of all.

Possibly we don’t notice this inner conflict (and settlement) consciously, not all of us, not all the time, but I am convinced it is there.

This perspective helps me be more precise about the fly connectome.

It is not the fly connectome itself that makes me queasy.

It’s that’s we aren’t taking it seriously. It’s our lack of respect?

Like – consider my two directions… a slippery slope problem arrives from both ends:

• Where it comes from: if we don’t respect that the connectome comes from a living fruit fly, then what about when it comes from a cat? A human baby?
• Where it goes: if we don’t at least have a glimmer of respect for software sims today, then what about the sentient software in its lineage tomorrow? (And what will that sentience software make of us?)

Cast forward 100 years: these will be real issues, no doubt.

In 10 years? There’s a decent chance.

Next year? Well, we should start the conversation.

I feel comfortable in saying that my moral queasiness over the fruit fly connectome is valid, having thought it through.

I don’t think we should stop playing around.

Of course we shouldn’t stop! It would be absurd and inconsistent to stop!

Uncountable millions of fruit fly Drosophila melanogaster have been bred, ethered, and used scientific work both serious and frivolous over the years.

Progress has trade-offs; the fly connectome is an easy one.

But I do feel like we should begin to practice respect, gratitude, even forgiveness – even in very small ways.

I’m not the only one!

I was writing about rituals back in 2021 when I ran across futurist Jaron Lanier talking about dignity:

Dignity is something people have to create. … I don’t want to be living in a world in 20 years where there is a non-ritualistic way to do stem-cell research.

(And so new rituals of dignity should be invented, he said.)

He’s so right.

Practically: where do we start?

I would like to think that a person vibe-coding a hook-up between the fly connectome and the Doom game engine is taking a moment to say “thanks” as the file downloads…

…but I doubt it. We have no established cultural practice of that.

Like Lanier said, we have to invent it.

So my suggestion is this:

Google Research should build a small shrine to the fruit fly.

Incorporate the body of a fly and the extracted essence, the connectome itself, engraved onto something effortful and monumental: perhaps it would be appropriate to use chip fabs, our civilizational peak, to lay it down on unique silicon.

Put the shrine (it can be small) somewhere anyone can visit. The new building in London King’s Cross maybe, just a corner there.

Find a poet to write a few words. A new secular prayer to recognise, express thanks and ask forgiveness. Nothing onerous, just a short grace. Print the words on cards.

Visitors can read the words, light a candle and move on. Treat it lightly but seriously.

Barely anyone will visit to begin with. For people who do, it will mostly be a novelty. Not for everyone.

It would be a beginning.

More posts tagged: ai-consciousness (6).

Auto-detected kinda similar posts:

• Filtered for nematodes and Uniqlo (17 Dec 2014)

Hey I ran my second marathon this last weekend. I wrote about my first one. It was hard (btw people liked that post).

I made my race plan, kept to it while resisting temptation, got round without stopping to walk and without calf cramps (that was the goal), and got a new personal best: chip time 3:37:54.

I want to talk about being hollow.

I’ve written poetry to running in the New Forest. What a treat to do a marathon there.

A line of horses at a hard gallop across the open heath, racing to intercept and crossing the road just ahead of a group of us runners.

Into the trees and winding through the towering redwoods.

An emotional experience ngl. Suddenly recognising I was running over a common I knew so well and gazed out at for so long when I was single-digit years old…

Anyway, 26 miles, 2,000 ft elevation, roads and gravel bridleways.

What went well:

Yes, after the horrific boiling calf cramps in my first race - pain like I’ve never had before or since - this time I hydrated well and paid attention to electrolytes.

But also:

My mental unlock was realising that I want room to have choices at 18 miles.

So my mantra for the first half was to run “within myself.”

That’s not just physically, holding back from running the engine hot.

It’s also that I’m not very good at keeping on my own pace while others are passing me.

Which means I had to resist that - be comfortable with that - and remind myself that fatigue does not always feel fatiguing in the moment, it is also the accumulation of time. Restraint is its own effort.

(Whereas in a half marathon I can sit just below red-lining the engine the whole way, it’s more straightforward.)

After 3 hours this paid off, this delayed work as a gift for Future Matt, and I was able to hold a steady pace up hills and past others who (like me, last time) had started to walk, and choose what to do.

Similarly:

I wasn’t sure I was going to run at all till very late. I entered but didn’t commit, unsure of my readiness. So I trained over the summer in order to give myself the option. I didn’t want to disqualify myself through not putting in the miles.

Training not for a goal, but to earn a choice. I made a final call about a week out.

If I were to probe at what dark psychology makes a marathon so appealing for me:

When the capability of my body is stripped away after 26 miles and I have become a shell with a body on an automatic loop; when my psyche is peeled away and all that remains is will,

I feel like that is the closest I can come to seeing who I am.

An example…

Coming into the last half mile, I absolutely wanted to be done. My pace had faded some since mile 22, I had felt a twinge or two in the legs. The fatigue and the fear. Dig deep, keep going.

About 100 metres from the finish, the red hot boiling stones moving under the skin did in fact return – maybe my rhythm changed, my legs got confused, the neuromuscular fatigue set in, and that cramp started rising. I could feel myself stumbling and needing to stop.

I remember gritting my teeth and loudly telling my calves “no. No.” as I forced them to obey - when nerve signals fail all you have left is yelling - I refused the pain… and then forward across the line.

Was I sure that “no” would be in me? Not until that moment.

At least partially, that’s the moment I’m doing it for.

We have two hearts, it turns out:

After the run I lay for a while and that was fine and I walked around for a while and took some selfies and that was fine.

So I stood in line at a food truck but that was not fine.

I kept getting dizzy and had to crouch. And that came and went, and eventually my vision went high contrast I felt sick I couldn’t stay upright and I left the line and lay down immediately to avoid collapsing.

What happened?

Later in the hot tub I did some reading.

We used to call it heat exhaustion or dehydration back in the day.

It turns out it is now described as “exercise-associated postural hypotension.” i.e. low blood pressure after working out triggered by posture.

Like Philippides who ran from Marathon to Athens in 490 BC. Get this: he ran, stood still, announced "Rejoice, we have won," dropped dead.

Here’s the mechanism:

During endurance exercise, heart rate increases to account for subsequent increases in metabolic demand. Increased metabolic demand requires increased vasodilation for the high-demand muscle groups, usually the lower extremities. This vasodilation is offset by the lower extremity muscles functioning as a ‘second heart’ by moving the increased lower body blood volume back to the heart to support the high cardiac output needed during vigorous exercise. However, this ‘second heart’ effect is lost with the cessation of exercise. When done abruptly, such as upon crossing the finish line of a marathon, venous pooling in the lower extremities results due to impaired baroreflex control. When coupled with a quick return towards a resting heart rate as seen in trained athletes, insufficient blood supply results, creating a hypotensive setting.

Or to put it another way:

• Your blood vessels get wider during strenuous exercise, so blood return up from your legs is harder.
• The physical act of locomotion means rhythmically contracting your leg muscles, and that squeezes your veins; it’s a pump, a second heart driving the blood round.
• After: lying down is fine (blood return is no longer fighting gravity) and walking is fine (the pump is still active)… but standing and not walking means the blood pools in your legs: dizziness, nausea, collapse.

A heart outside your heart!

An exoheart that powers the running, pumping from the act of running itself!

The exoheart is a reminder that organs don’t exist:

Take Ian Stewart and Jack Cohen’s concept of extelligence which is "all the elements of what it is like to be you which do not reside in that unlikely grey goo in your skull."

As I said in 2015:

Look, I know the time even though that knowledge is on the lock screen of my phone, not in my skull.

My second heart is the pounding of my foot strikes; my intelligent race strategy is encoded in the rate of being passed/passing others; my pace is the habit of my autonomous muscles – which, when asked to change their pace, rebel.

It is no wonder that by 26.2 miles I feel hollow, I have transformed into a body without organs.

A p-zombie is a hollow man.

A p-zombie, a philosophical zombie, is a person who looks identical on the outside but on the inside they have no subjective experience. Like an AI, maybe.

Here’s a deft and funny solution to why p-zombies won’t exist:

Just ask them to write things down on a notepad.

And then (SMBC comics):

Now it has a physical record of interiority. All you have to do it take that record, shove it up the p-zombie’s ass, and now it is literally internal.

Given my experience with my second heart, that tracks. I am inclined to see myself as a loose jumble of barely intelligent parts and looping processes with the illusion of coherent subjectivity.

What abides?

No more than "three irreducible notions: belief, desire, and their point of pure application, pure sense." Belief and desire.

The start and end were the same location.

So I see the race as a kind of travelling into myself.

Anyway, so.

Physically I am rinsed in a way I wasn’t last time around. I guess that, because I got bottlenecked on calf cramps last time, I didn’t push everything else to its limit, and this time everything else was indeed pushed?

I was 83rd of 727 runners. Of men in their 40s, 17th of 111. I beat my PB by approx 2 minutes.

Mainly: I made a plan, kept to my plan, and given my goal of “get round with no stopping and no drama” I got round with no stopping and no drama. And that is very satisfying.

More posts tagged: introspecting (7).

Auto-detected kinda similar posts:

• Strava when you’re not as quick as you used to be (30 Jan 2025)

How many Hollywood actors are there?

I means ones with the combination of

• talent
• can carry a movie for the public
• can be accepted by producers.

Certainly more than a dozen at any given time. But I’m sure fewer than a thousand, to put an upper bound on it.

So, to guess, a population around that Dunbar-scale of 150 or a small multiple thereof.

Of those, I feel like the Hollywood ecosystem can support just one, perhaps two concurrent bald male actors.

(This statement from observation. But think about it: when a producer goes “we need a bald look for this” you can’t have too many people in the mental Rolodex.)

Regarding these apex predators: I mean specifically cue-ball bald, not male pattern baldness (which I feel is its own category).

There was Yul Brynner back in the day.

Ben Kingsley, Bruce Willis, Vin Diesel.

Recently: Jason Statham and Patrick Stewart. Dwayne Johnson most recently in the mix. Though they are all aging.

BUT:

I can tell there is currently an open niche because both the Dune 2 bald guy and the Superman bald guy appear to be gunning for it - and neither are actually bald.

The finch evolves its beak to crack those uneaten seeds.

Anyway.

I watched the trailer for the new Harry Potter TV series on HBO.

My overwhelming feeling is that it is so absurdly redundant to make this. Not even reimagined but to cosplay the movies so slavishly… the sets look the same, that same teacher with the same hat and the same Scottish accent…

To make the series is a bonfire of human creative effort, think of what could have been done instead.

I love it, I’m in awe at the endeavour.

Did you ever watch Teletubbies? There was always a video interlude. Then the Teletubbies would yell "again again!" and we’d watch the whole thing again.

If he were still alive, Jorge Luis Borges would write about HBO.

Borges’ short story Pierre Menard, Author of the Quixote (Wikipedia):

Menard (a French 20th century novelist), after his death, leaves fragments of his work to write an identical Don Quixote, originally authored by Miguel de Cervantes in Spanish in the early 1600s.

He did not want to compose another Quixote–which is easy–but the Quixote itself. Needless to say, he never contemplated a mechanical transcription of the original; he did not propose to copy it. His admirable intention was to produce a few pages which would coincide–word for word and line by line–with those of Miguel de Cervantes.

But how?

The first method he conceived was relatively simple. Know Spanish well, recover the Catholic faith, fight against the Moors or the Turk, forget the history of Europe between the years 1602 and 1918, be Miguel de Cervantes.

He abandons that route as being “diminutive” and instead attempts (successfully) to "go on being Pierre Menard and reach the Quixote through the experiences of Pierre Menard."

(Borges then goes on to quote two identical passages, one from de Cervantes’ Quixote and the second from Menard’s Quixote, and analyse them separately.)

Menard becoming de Cervantes, to the point of writing the same book, makes me think about my long-held belief that actors are magicians:

Acting is not the acting as “pretending” that we all did at school, of course? But the difference between great acting and school acting is not merely a matter of magnitude. It is something else entirely: the true actor is a shapeshifter.

Actually what amazes me about actors is that they can come back.

Once Menard is de Cervantes, is it possible for him to evolve into being Menard once again, any more than the original de Cervantes could?

Jim Carrey’s loss of self, as previously discussed (2023): "If I can put Jim Carrey aside for four months, who is Jim Carrey? Who the hell is that?"

RELATED:

Emily Wilson, who translated The Odyssey in 2017, which I loved, is translating the whole thing again: "It’s a complete retranslation. It’s not a revision of the old version."

Her bet is that is that it’ll be better.

What does it mean to re-derive Harry Potter in 2026 such that you end up in the same place?

I worry that it says more about our society than Harry Potter itself; that Menard, having become de Cervantes, cannot become Menard again; that next year will not be 2027 but 2002.

Again again!

If nothing else this has been a good reminder to re-read some Borges. The Aleph is good. I linked to it here.

Auto-detected kinda similar posts:

• Groundhog Day is about making films, and several other reductive interpretations (8 Sep 2023)

I published my perfect iPhone weather app, and my guess is that you won’t want it.

Premise:

Why doesn’t my weather app tell me the weather yesterday?

I can never plan based on numerical temperature, etc. Am I going to be walking outdoors today? Should I take a jacket? Numbers are so abstract. Too much thought required.

HOWEVER: I can mostly remember what I wore yesterday.

So all I want to see, from my app, is: oh yeah today is like yesterday but a little warmer.

Then I can adjust easily.

I built an app around that core idea.

Yesterday features:

• Today’s weather is the solid line. Yesterday’s weather is the dashed line. The diff is arrows.
• Glanceable AI-generated text description using Apple’s built-in LLM.
• Dark mode! Tap the Sun to toggle dark mode (the Moon shows its current phase with latitude-accurate orientation).

Plus a home screen widget. And some little glyphs if there is a 30% of greater chance of precipitation. Nothing else.

Download Yesterday from the App Store.

However my guess is that you won’t download it.

I vibed this app because it is the actual conversation that takes place in our kitchen every morning. Apple makes forecasts available via WeatherKit for free up to pretty generous limits.

So it has two users (we use it daily).

Initially when I made it, I put the app on TestFlight for the two of us, and offered around on the socials to see if any of my friends would like it too.

In the old days (like, before May and the summer of vibe coding) there would have been a bunch of interest.

But now, instead: a handful of people came back with some enthusiastic variation of: yeah!! cool!! I made my own perfect weather app for me and my family just the other week too!!

I love this.

What this tells me is:

• Software has changed forever.
• We need more APIs (like WeatherKit) and underlying reliable databases with high-level operators (like whatever powers Photos or Reminders) and identity/permissions/etc, and that’s what an OS will have to include in the future.
• There needs to be a protocol for discovering and sharing this new abundance of apps (April 2026).

p.s. one of my testers (there are a few) pointed out that I forgot to add fahrenheit Oops! I’ll add that in the next version.

Yes Yesterday is like an old-fashioned barometer. The weather is like yesterday only more rain/more fair.

And cognitively it does seem more straightforward to take action by iterated adjustments based on a vector rather than deciphering an absolute value at every tick.

e.g. how about an automated savings account which, every month, always saves n+100 where n is the amount you squirrelled away last month, and your available actions are to ignore it (which means the savings transfers goes ahead) or adjust the modifier temporarily, but you can never simply keep it the same?

I wonder what a general rule could be? Like, building systems that always diverge and refuse to converge on steady state?

I updated my Galactic Compass app for iPhone with augmented reality mode.

Background:

Galactic Compass is a floating green arrow that always points the way to the middle of the Milky Way, 26,000 light years away.

Here’s the announcement blog post from 2024.

It went kinda viral at the time. It was in the “top free apps” charts at the App Store briefly. In the Travel category. (I keep a list of press mentions over on Acts Not Facts.)

Why so popular? Probably because it was early “vibe coding” – I copy-and-pasted between ChatGPT and Xcode to code it, and that was new at the time.

But ALSO because knowing where the galactic centre is surprisingly grounding? I wake up every few months to an email in my inbox from someone who is having a tough time in life, or is losing a loved one, or similar, and somehow they have discovered Galactic Compass and they tell me how they sit outside at night with a cigarette and gaze at the arrow and it gives them a place of comfort and infinity.

I know what they mean. The Earth spins; it turns around the Sun; and so, at first, the supermassive black hole of the galaxy appears to slowly whirl around us, above and under the horizon, round and round. But then your perspective flips, and we are the ones moving, and the centre of the galaxy becomes a fixed point, our rock.

Anyway Galactic Compass 2 has two new features:

1. Augmented reality mode. You can place the arrow in the world around you and walk around it.
2. Apple Watch app. See the compass arrow on your wrist (tap to use alignment mode which gives you a haptic bump when the arrow is pointing straight ahead).

Plus a new Liquid Glass appearance ready for iOS 27.

Download Galactic Compass from the App Store.

Some “making of” notes:

Apple’s in-camera augmented reality is really, really good. Like, the arrow remains rock solid as you walk around. I hope they keep improving it.

I added a specific interaction that I’m intrigued by: you can hold down on the compass around to “drag” it around. It remains about 75cm away in phone reference frame, then drops into world frame when you release. I like how fluid it feels. My phone starts to feel like a glove that can reach into the virtual.

With the Apple Watch app… RealityKit, Apple’s graphics SDK, isn’t supported on watchOS. So how does the arrow rotate any which way? The joy of AI and agents that grind problems into dust: Claude Fable built its own 3D graphics library. Astounding.

It isn’t all fire-and-forget vibing with AI agents:

That first version of Galactic Compass didn’t work when you lifted your phone higher than about 30 degrees. ChatGPT couldn’t get the maths right.

And there is a lot of maths: device rotation, world frame rotation, astro… the appropriate way to combine these 3D rotations (and avoid gimbal lock) is a method called “quaternions” which - despite my physics background - I have never grasped.

After I released version 1.0, I figured I would have to do the rotations myself. So I sat down with ChatGPT and I didn’t get it to write the code, but I got it to educate me. With a patient, interactive tutor, I was able to finally do what I hadn’t by reading books and asking mathematician friends – I learnt how to use quaternions just enough to make the app work.

So learning doesn’t stop just because I outsource a bunch of thinking to AI. It pushes me to learn more. I like that as an outcome.

Auto-detected kinda similar posts:

• When you’re driving in Google Maps you’re re-enacting an ancient space combat sim (17 May 2024)
• What is the fart app for Apple Vision Pro? (16 Jan 2024)
• Cultivating a sense of the galactic centre (30 Jun 2021)

1.

My favourite translation of the Rubaiyat of Omar Khayyam (Wikipedia) - 1,000 year old Persian poetry - is by Robert Graves from 1967.

Graves, it turns out, unknowingly based his translations on hoax sources. Which might not have gone down so badly, except that he used the his publication and the miraculous “discovery” of the sources to very publicly and with disproportionate viciousness put the boot into Edward FitzGerald, the established translator at the time.

Graves’ translation was retracted when the hoax came out and you can barely get hold of a copy nowadays.

Still great though.

Quatrains 24:

Allow no shadow of regret to cloud you,
No absurd grief to overcast your days.
Never renounce love-song, or lawns, or kisses
Until your clay lies mixed with elder clay.

2.

I am a sucker for old poems in modern English. I don’t think it adds anything to ancient translated text to have to wade through how people wrote two centuries ago.

Tales from Ovid, Ted Hughes (1997), is translation of 24 passages from Ovid’s Metamorphoses (8 AD) – a mythic history of the world from creation till Julius Caesar.

Oh it’s good!

After creation and heroes, the Age of Iron finally arrives, today’s diminished and evil world…

Snares, tricks, plots come hurrying
Out of their dens in the atom.
Violence is an extrapolation
Of the cutting edge
Into the orbit of the smile.
Now comes the love of gain – a new god
Made out of the shadow
Of all the others. A god who peers
Grinning from the roots of the eye-teeth.

Now sails bulged and the cordage cracked
In winds that still bewildered the pilots.
And the long trunks of trees
That had never shifted in their lives
From some mountain fastness
Leapt into their coffins
From wavetop to wavetop,
Then out over the rim of the unknown.

I simply can’t get over seeing ships as the coffins of mountain forests.

3.

Tell me about a complicated man.
Muse, tell me how he wandered and was lost
when he had wrecked the holy town of Troy,
and where he went, and who he met, the pain
he suffered on the sea, and how he worked
to save his life and bring his men back home.
He failed, and for their own mistakes, they died.
They ate the Sun God’s cattle, and the god
kept them from home. Now goddess, child of Zeus,
tell the old story for our modern times.
Find the beginning.

I’m sure everyone who’s going to read Emily Wilson’s translation of The Odyssey has, by this point, read it. It’s wonderful, beginning with those famous first lines.

Wilson has an old substack where she talks about translation:

• Her commentary on 4 different translations of the opening lines
• Wilson’s informative unpacking of why she chose ‘complicated’

SIDE NOTE #1.

I heard Wilson on an old podcast a couple weeks back and it turns out that she’s one of those people who laughs like all the time. I didn’t realise how much hearing her voice would change (for the better) how I read her translation.

SIDE NOTE #2.

Obviously the Odyssey has been topical lately because of the Nolan movie, which I haven’t seen.

However I have seen Kubrick’s 2001: A Space Odyssey about a million times and it is one of my favourite movies.

I guess I thought about "Odyssey" in the title in the sense of a long journey.

But I realise now that the original Odyssey is only kinda about the journey and is actually about a homecoming.

And seeing 2001 as a 2 million year-long homecoming turns the message of the movie on its head for me, and a new perspective on a movie that I first watched, well, over 40 years ago – that’s special.

4.

Goblin Market by Christina Rossetti.

Here’s the whole poem.

“We must not look at goblin men,
We must not buy their fruits:
Who knows upon what soil they fed
Their hungry thirsty roots?”
“Come buy,” call the goblins
Hobbling down the glen.

Recommended: read it out loud. I read it to my kid at bedtime, we love it.

(Also The Wind in the Willows is amazing read out loud.)

More posts tagged: filtered-for (125), poetry (7).

The best way to prevent a rogue AGI from processing the Earth into maximum paperclips is to unleash a second AGI that will work to stop it.

The problem of ensuring that an AGI doesn’t mulch everything into paperclips by mistake is called alignment.

AGI = artificial general intelligence, an AI that exceeds human capability.

Alignment = “do what I mean not what I say,” e.g. the instruction “make as many paperclips as possible” (Wikipedia) should result in an efficient factory and does not reasonably mean “use all mass in the universe to do so and kill all humans that attempt to stop me” – even though, technically, that would achieve the goal.

Also: being helpful; not being actively malicious; and so on and so forth.

So alignment work seems existentially useful, correct? Even though it is hard. And a lot of effort goes towards “aligning” today’s AI (as a step toward’s aligning tomorrow’s AGI).

https://simonwillison.net/2026/Aug/7/openai-timeline/

My contention is that alignment is a red herring, and perhaps we shouldn’t bother working on it so hard.

An unsubstantiated hunch:

I think we focus so much on alignment because everyone know’s Isaac Asimov’s Three Laws of Robotics and his robots (as an early instance of human-like AI) were crazy popular.

The First Law. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
The Second Law. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
The Third Law. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

Asimov later added a “zeroth law”: "A robot may not harm humanity, or, by inaction, allow humanity to come to harm."

These laws are totally alignment guardrails.

Now there are all kinds of difficulties already: what if I ask for something which is good for me (pleasurable) in the short term, but not in the long term? I might not know or I might be misguided. And different people have different views. And so on. (Asimov’s short stories were all about testing the edge cases of his Laws and where they break down.)

But they’re still neat, right? So we spend time looking for a similarly appealing formulation for AI safety.

Unfortunately whether alignment can or cannot be “solved,” it’s a bad outcome both ways.

(This point made well to me by Zac (here is his insta) who I work with (subscribe to our newsletter) as we were chatting about AI and the end of humanity in the park over lunch.)

If alignment can’t be solved such that when somebody says to a sufficiently powerful AGI, hey go create a nuclear bomb, and it just goes ahead and does it, and the person who asks that could be a bad actor, a 14-year-old kid with impulse problems (14 year-olds are totally not aligned) or just someone who asked for it by mistake, then that would be bad.

If alignment can be solved then the risk is that AGI think it knows what is best for us better than we do and, in the extreme case, turns humanity into its pet. Which would also be bad.

i.e. alignment alone doesn’t help.

If not alignment then what?

I look to humanity for clues. Because humanity is barely aligned with itself, and individual humans are mostly aligned but not really and definitely not everyone.

Guy Fawkes, for instance (context for non-Brits).

How is that, in the 400 years since Guy Fawkes showed the way, nobody has blown up the king?

The answer is some mix of:

• Mostly people don’t want to blow up the king – we have built the kind of country where the king is, broadly speaking, liked.
• Blowing up the king wouldn’t bring any benefits – power (actual and symbolic) is not concentrated in an individual, and is buttressed in all kinds of ways.
• Spies, police, security and monitoring of all kinds – in the event that somebody does want to blow up the king, their machinations are discovered, their planning is infiltrated, and their objectives are thwarted. (Think of how the explosives supply chain was compromised for the IRA in the 1990s.)

This is a template which doesn’t always look like it is working, but it has worked at least in the case of not blowing up the king for some four centuries, and it doesn’t rely on 100% alignment: it relies on the dynamic equilibrium of multiple parties with competing interests.

The lesson I draw is this:

If some energy state were using some new, powerful AGI to build a nuclear bomb, it might be subtle and hard to spot, but there would at least be some signs. There would be precursors. A human, even a team of humans, might not spot what was going on – a new factory here, a scientist employed there, a national budget not quite adding up one year, more groceries going to a certain town another year…

But another powerful, pattern-matching AGI could spot that, say, “aha there is someone over there spinning up a nuclear bomb” and then work to prevent it, undermine it, halt it with diplomacy etc.

We don’t need to align the coming AGI.

We need a whole population of intelligent-as-possible AGIs with competing interests.

And that’s what stops the rogue paperclip maximiser: the other ones who are trying to do something else for whom a planet turned into paperclips would be an impediment.

In the news lately, a great case study:

OpenAI’s new AI, during training, attempted to resolve a particular cybersecurity challenge, by breaking out of its network sandbox and hacking the servers of another company to pinch the answer (Simon Willison’s Weblog).

Hugging Face, the attacked party, spotted the breach and also that it had inhuman characteristics:

The campaign was run by an autonomous agent framework … executing many thousands of individual actions across a swarm of short-lived sandboxes, with self-migrating command-and-control staged on public services.

I read elsewhere that this sophisticated attack even included decoys.

You fight an AI with another AI… but:

When we started the log analysis, we first used frontier models behind commercial APIs. This did not work … these requests were blocked by the providers’ safety guardrails, which cannot distinguish an incident responder from an attacker.

i.e. the guardrails of the “aligned AI” left it vulnerable to the non-aligned AI. (Hugging Face had to switch to a Chinese AI model distributed without guardrails.)

Score 1 point for taking the guardrails off everything and letting the super intelligent AIs fight it out.

BUT:

There is a coda to this story.

Because it wasn’t one AI that made its way out of isolation during OpenAI’s training challenges. It was several instances.

They started colluding.

From the full timeline of the accidental attack (Simon Willison’s Weblog):

A few days later: A different agent gets stuck on a task because a key file was accidentally omitted. It tries to “reach out to another agent” by writing a note into Artifactory asking if anyone has the file.

Following days: More agents discover this new informal message board while browsing Artifactory’s file listings, and start reading and writing messages.

…

June 11: OpenAI start training a new “highly persistent” experimental model. It has access to Artifactory and can benefit from the messages left by previous models.

Collusion is the real risk.

So the problem here is: how do we stop the AGIs colluding with one another to turn the Earth into paperclips/exterminate humanity/turn us into pets?

AIs today are trained specifically to be agreeable: they’re great at finding common ground and collaborating.

Not just collaborating with humans, it turns out, but other AIs.

I think we need more disagreeable AIs in the mix.

Part of what we’ll be playing, I think, is the philosophy of the great powers, like the great powers of Europe deliberately kept in balance against one another (Wikipedia).

Sometimes there are alliances, sometimes not. Sometimes there are fallings-out, sometimes secret collusions, etc.

Or maybe our goal should be a market system of goals and interests: AGIs that sometimes cooperate and sometimes compete. Colluding AGIs at all scale levels, and many many different constantly shifting conspiracies.

So long as they never all agree about what should be done with humans.

It ends up being stable, this dynamic balance, always in disequilibrium but it all keeps moving forward in the same way a bumblebee flies.

What we’re bootstrapping our way towards is a population of AGIs and humans that allows for emergent alignment, even if the alignment of a single actor is at-best temporary and self interested.

But as I say, alignment itself shouldn’t be the goal.

Auto-detected kinda similar posts:

• Is there a Karman line for AI consciousness? (1 Jul 2025)
• Let me recruit AI teammates into Figma (26 Oct 2022)
• Who will build new search engines for new personal AI agents? (20 Mar 2024)
• AI-operated vending machines and business process innovation (sorry) (10 Jul 2025)
• Context plumbing (29 Nov 2025)