Download a PDF of An AI in the City of God (paid subscribers only)

Download or listen to an audio recording of An AI in the City of God.

Also available at:

God is always trying to give good things to us, but our hands are too full to receive them.

Augustine of Hippo, City of God (c. 415 AD)

And by hands, of course, St. Augustine means our minds. Our minds are too full to receive the gifts we are given.

Like generative AI.

Augustine wrote City of God from his bishopric in North Africa after the Visigoths sacked Rome in 410 AD. It’s a tougher read than his far more personal Confessions, but the basic idea is that there are two worlds that exist simultaneously here on Earth: there’s the City of Man – the physical instantiation of society, typically corrupt, always decaying, ruled by utterly fallible men and the evil they do – and the City of God – a society that lives in our hearts, incorruptible and timeless, less ruled than inspired by the illumination of the Divine.

This joy in God is not like any pleasure found in physical or intellectual satisfaction. Nor is it such as a friend experiences in the presence of a friend. But, if we are to use any such analogy, it is more like the eye rejoicing in light.

The eye rejoicing in light. What a perfect phrase for the human mind’s response to discovery and new knowledge!

Augustine’s writings were some of the first in the Christian tradition to wrestle deeply with why an omnipotent God of good allows the persistence of evil and why really bad things happen to good people at scale. His answers – original sin, free will as God’s vital gift, the abiding strength of epistemic communities of men and women of good will and faith – are foundational to the past 1,600 years of Western thought.

The City of Man is a community of implacable obligation that exists in the physical world wherever humans gather as a society. You are born into the City of Man, you will live out your life in the City of Man, you will enjoy what you can and you will suffer what you must in the City of Man, and you will die in the City of Man.

The City of Man IS. The City of Man is inexorable, literally “that which cannot be prayed away” in the original Latin.

The City of God, on the other hand, is prayed into being. The City of God is a community of choice that exists across time and space wherever humans of good will (Full Hearts) and a common endeavor of truth-seeking (Clear Eyes) come together in faith.

Must this faith be Augustine’s faith in the God of Abraham and His Son? Nope. At least not as I mean the City of God, which admittedly is more as metaphor than I expect Augustine meant it! My goal isn’t to trivialize Augustine’s beautiful concept (on the contrary), but I believe that the City of God can be manifested in a high school football team a la Friday Night Lights just as powerfully as in a monastery of Benedictine monks. I believe that a Walt Whitman poem is as revealing of the transcendental divine as any passage in the Bible. I don’t share Augustine’s faith. We don’t mean exactly the same thing when we write “the City of God”. But in my heart of hearts I know that we share a similar eye rejoicing in the light.

I think that all humans with a transcendental faith – a belief in a literally (super)human power of good that exists above, below and beyond the world that we know but acts within the world we know through its inspiration of human hearts and minds trapped in the world that we know – can reside in the City of God, even if they don’t mean the same thing when they write the word “God”. Christianity is a transcendental faith, as is Islam and Buddhism and Judaism and most of the great religions of the world.

As is my faith in the Spirit of Man.

What do I mean by the Spirit of Man? Maybe the simplest way to describe my faith is that I believe there is an arc and arrow to human history, an arc and arrow that goes fitfully up and to the right, propelled by the core small-l liberal virtue of a timeless autonomy of the individual human mind and the core small-c conservative virtue of a social human connectedness anchored in time.

To be honest, most of the time I feel like such a sucker for believing in the Spirit of Man.

I feel like Winston in Orwell’s 1984, persisting in a belief that the City of Man can’t possibly be as evil and permanent as it seems, even as I am tortured into submission and forced to admit that there is zero evidence – zero! – of any substantive movement in my lifetime towards a more just and free society reflecting the ascendancy of the Spirit of Man.

O'Brian:      We control life, Winston, at all its levels. You are imagining that there is something called human nature which will be outraged by what we do and will turn against us. But we create human nature. Men are infinitely malleable. Or perhaps you have returned to your old idea that the proletarians or the slaves will arise and overthrow us. Put it out of your mind. They are helpless, like the animals. Humanity is the Party. The others are outside - irrelevant.

Winston:    I don't care. In the end they will beat you. Sooner or later they will see you for what you are, and then they will tear you to pieces.

O'Brian:      Do you see any evidence that this is happening? Or any reason why it should?

Winston:    No. I believe it. I know that you will fail. There is something in the universe - I don't know, some spirit, some principle - that you will never overcome.

O'Brian:      Do you believe in God, Winston?

Winston:    No.

O'Brian:      Then what is it, this principle that will defeat us?

Winston:    I don't know. The Spirit of Man.

At this point, O’Brian continues to torture Winston mercilessly until he rejects his faith in the Spirit of Man and embraces a faith in Big Brother. Wheeee!

I’m still hanging in there – barely! – in part because I haven’t been directly tortured with a box of rats attached to my head like Winston (although my Twitter addiction has created something pretty close to a Room 101 for me), but also because I think Orwell, like the democratic socialist true-believer he was, got something fundamentally wrong about the Spirit of Man.

The Spirit of Man is not found in the proletariat.

I mean, it is, in the sense that the Spirit of Man is not not found in the proletariat. But the Spirit of Man is not a class thing! Or a race thing or an age thing or a nationality thing or a gender thing or any other limiting principle found in the material world-as-it-is. The Spirit of Man, like any transcendental, outside the world-as-it-is phenomenon, is limitless. It is reflected in ALL humans regardless of ANY material condition.

The Spirit of Man is your birthright as a human being. It is not ladled out to you from some central pot. It cannot be granted or taken away by the State or the Corporation. Certainly you can give away your birthright for a mess of pottage, and in my experience many people do just that. But many don’t! Wherever they sit on the age spectrum, the class spectrum, the gender spectrum, the race spectrum … many don’t. And by claiming their birthright as a human being, by rejecting the Nudges that would place their mindfulness and community in the City of Man, by allowing their eyes to rejoice in the light of a transcendental faith, they enter the City of God.

The City of God, illuminated by the Spirit of Man, is where you express your autonomy of mind.

The City of God, illuminated by the Spirit of Man, is where you express your connectedness with other humans.

The City of God, illuminated by the Spirit of Man, is where you express your choice to Make, Protect and Teach, your choice to pursue a conscious life of meaning NOW, no matter what you’ve done in the past or your position, high or low, in the City of Man.

Why?

Because generative AI is not just a tool of search, but a tool of discovery.

Today we equate the idea of search with Google and the idea of discovery with Netflix Recommends or Amazon Suggested For You.

My friends, that is the saddest sentence I have ever written.

The great gift of generative AI like ChatGPT is that it takes the ideas of search and discovery away from the petty Tech Principalities and returns them into the individual hands of those with faith in the Spirit of Man.

And that’s the most hopeful sentence I have ever written.

I was born in 1964, or as I sometimes like to call it, 34 BG. Back in the before-Google times, the idea of search was located in two places – the encyclopedia set that your parents displayed reverently in the living room, searchable and physically organized alphabetically by subject, and the card catalog files presented as an altar at your school or public library, searchable alphabetically by author, title or subject and physically organized by the Dewey Decimal classification. The meaning of search was looking up a topic by name in the encyclopedia or a book by author in the card catalog. And that was all fine and good. Eventually you’d generalize your topic enough to match the subject-level specificity available in your encyclopedia (World Book blah, Britannica yaaas) or eventually you’d find the Dewey Decimal code range you were looking for in the library stacks, and then you’d track the little white stickers on the book spines until you found the specific Dewey Decimal code you were looking for. Search complete.

But that’s exactly when discovery would begin.

The most impactful learning experiences of my life have occurred when I went beyond what I had been instructed to find specifically … when I started reading the next entry in the encyclopedia, when I started flipping through the books to the right and the left of the book I was searching for in the library, when I started looking at the books on the shelf below and the shelf above … and I discovered something NEW.

The power of the physical library is its searchable collection of books. The superpower of the physical library is its service as a discovery tool, made possible by the Dewey Decimal virtual organization of the individual books by content similarity. The Dewey Decimal system creates a virtual network graph of the books in the library, and then that graph is instantiated in the physical placement of the books.

Yes, a physical library with accessible stacks is a form of generative AI!

Now imagine that all the books in the world are in that library. Or if not all of the books (yet), imagine a library thousands of times larger than any library you’ve ever seen. Imagine if the content similarity between books isn’t instantiated in physical shelves that you can walk around, but is available to you for the asking in natural language of plain-spoken meaning and recall.

“Hey, I’ve been thinking a lot about subject XYZ, especially as it pertains to ABC and DEF. In fact here’s the full text of the article that got me thinking about this. Here are the connections GHI I’m interested in exploring, but I’m also thinking that other people may have made connections I hadn’t thought of yet. When I say other people, let’s only consider published book authors and authors in the top 20 academic journals in field JKL, and please be as accurate a librarian as possible. Has anyone else written about this idea or connection that I’m thinking about? Am I understanding these concepts correctly? What are 10 other connections from that original article that I’m not thinking about? Let’s goooooo.”

Wash, rinse, repeat. Apply to any aspect of human knowledge. Substitute commercial for academic. Broaden or narrow the scope of the library stacks you want to wander around. Do whatever sparks your human creativity and inventiveness in whatever Make/Protect/Teach endeavor you’ve got.

Now imagine that this discovery-librarian is available to everyone and everywhere. I know we’re not quite there yet – GPT-4 at $20/month through OpenAI is lightyears ahead of the free options available for GPT-3 and GPT-3.5 – but boy are we close. Now imagine that neither the content network graph nor its reporting is confounded or ‘adjusted’ by the librarian’s business model, whether that’s inserting an advertisement you must endure to get to your discovery or a promoted content elevation separate from your priorities.

THIS is generative AI in the City of God, a tool for not just search but discovery, a tool for shining the light of the Spirit of Man just as far as the rejoicing eye can see.

If, as Augustine said, our minds are not too full to accept it.

It’s easy enough to tell the people who don’t accept the gift. They play ‘gotcha’ with ChatGPT, asking it a question as if it were a false prophet and then pointing out the answer’s flaws as proof of the assumption. If this is your experience with generative AI, intentionally or unintentionally, you’re doing it wrong.

Generative AI today is based on large language models (LLMs), meaning that it is trained on immense quantities of human-generated texts. As a result of its training and sourcing, generative AI is inextricably human-ish. It is, in all the ways that matter, an artificial human intelligence, not an artificial machine intelligence, and you should treat it as such!

To tap into the discovery-librarian superpower of GPT-4, you should talk to to it exactly as you would a human knowledge-work assistant who wants desperately to please you with its responses but has none of the cues and context an actual human knowledge-work assistant would possess insofar as what responses would actually please you the most. The secret to good GPT4 prompting is the same secret as good management of a human knowledge-work assistant – you must provide the cues and context for what is pleasing and what is not. It’s really as simple as that. You want factual accuracy above all else? Is that what makes you happy? Tell your AI that. You want new and wild connections above all else? Tell your AI that. You should provide corrective feedback when the answers go off track, just like you would a human knowledge-work assistant. You should provide encouragement and praise when the answers are on track, just like you would a human knowledge-work assistant.

If you think of generative AI as an artificial human intelligence with superhuman powers of knowledge, recall and connection, but utterly lacking – and I mean totally and absolutely lacking – in the physical, social and macro cues and context that we take for granted in our communication with biological human intelligences … well, you will be amazed at the decentralized, bottom-up GOOD you will accomplish!

Yeah.

But I also think that generative AI in the City of Man will probably kill or enslave a lot of us, maybe most of us.

I’m not even joking, as the kids would say.

Oh, not directly kill or enslave us. I’m actually not that worried about a ‘non-aligned’ LLM over the AGI threshold ‘deciding’ to wipe out humans in some Matrix-y sort of way because frankly I don’t think we make it that far. I think our current situation is the equivalent of distributing atomic bombs and ICBMs to every nation on Earth, and worries about a non-aligned AGI are kinda like saying “oh, but a future hydrogen bomb will really pose a problem”.

There are two clear and present dangers we face today from generative AI in the City of Man.

The first is the ability to destabilize Common Knowledge structures – what we all know that we all know – necessary for Western democracies to survive. Common knowledge structures like “your money is safe in the bank” or “the announced election results are legitimate”.

The second is the ability to change history.

The first is the direct killer. This is what creates global economic depressions, civil wars and world wars. This is what drives the Widening Gyre to its breaking point.

I honestly believe that if you gave me a few million dollars to buy social media for an astroturf (fake grassroots) campaign, a few months to flatter and promote fellow travelers, and an unthrottled API to ChatGPT-4, I could destroy any common knowledge structure on Earth.

I honestly believe that strategic adversaries of the United States – China and Russia foremost among them, but North Korea, Iran and (on a bad day) India and Saudi Arabia – together with domestic political entrepreneurs on both the MAGA/Q-adjacent right and the insane woke left, together with the billionaire Tech Bro and Crypto Bro accelerationists of the Thiel-industrial-complex, are all engaged TODAY in a somewhat coordinated and definitely strategic effort to destabilize the United States in narrative-space. Hey, haven’t you heard that the US dollar is about to lose its reserve currency status? Don’t you understand that we face the imminent de-dollarization of the world, which would be, in the words of The Former Guy, the “worst defeat for the United States in 200 years”? Aren’t you scared about that?

Armed with GPT-4, this all gets much, much worse. Why? Because generative AI can supply any agent of chaos with 15 extremely truthy-sounding footnotes as ‘data’ to ‘support’ whatever ridiculous claims they might make, whether it’s “hundreds of US banks are insolvent” or “hyperinflation is here” or “the vax is killing you”. It’s all just sophistry at an industrial level of production, in service to naked political and commercial self-interest.

But wait, there’s more. As horrible and destructive as the first clear and present danger might be, it’s not as bad as the second. We can survive a global depression. We can survive a war. Most of us can, anyway. But none of us survive the second threat, not as human beings.

The ability to change history is the enslaver. This is the apotheosis of Nudge into Big Brotherism. This is the death of the Spirit of Man.

Here’s how it works, from Orwell’s 1984. I know it’s a long quote, but trust me. Worth it.

Suddenly there sprang into his mind, ready made as it were, the image of a certain Comrade Ogilvy, who had recently died in battle, in heroic circumstances. There were occasions when Big Brother devoted his Order for the Day to commemorating some humble, rank-and-file Party member whose life and death he held up as an example worthy to be followed. Today he should commemorate Comrade Ogilvy. It was true that there was no such person as Comrade Ogilvy, but a few lines of print and a couple of faked photographs would soon bring him into existence.

Winston thought for a moment, then pulled the speakwrite towards him and began dictating in Big Brother’s familiar style: a style at once military and pedantic, and, because of a trick of asking questions and then promptly answering them (’What lessons do we learn from this fact, comrades? The lesson—which is also one of the fundamental principles of Ingsoc—that,’ etc., etc.), easy to imitate. 

At the age of three Comrade Ogilvy had refused all toys except a drum, a sub-machine gun, and a model helicopter. At six—a year early, by a special relaxation of the rules—he had joined the Spies, at nine he had been a troop leader. At eleven he had denounced his uncle to the Thought Police after overhearing a conversation which appeared to him to have criminal tendencies. At seventeen he had been a district organizer of the Junior Anti-Sex League. At nineteen he had designed a hand-grenade which had been adopted by the Ministry of Peace and which, at its first trial, had killed thirty-one Eurasian prisoners in one burst. At twenty-three he had perished in action. Pursued by enemy jet planes while flying over the Indian Ocean with important dispatches, he had weighted his body with his machine gun and leapt out of the helicopter into deep water, dispatches and all—an end, said Big Brother, which it was impossible to contemplate without feelings of envy. Big Brother added a few remarks on the purity and single-mindedness of Comrade Ogilvy’s life. He was a total abstainer and a nonsmoker, had no recreations except a daily hour in the gymnasium, and had taken a vow of celibacy, believing marriage and the care of a family to be incompatible with a twenty-four-hour-a-day devotion to duty. He had no subjects of conversation except the principles of Ingsoc, and no aim in life except the defeat of the Eurasian enemy and the hunting-down of spies, saboteurs, thoughtcriminals, and traitors generally.

Winston debated with himself whether to award Comrade Ogilvy the Order of Conspicuous Merit: in the end he decided against it because of the unnecessary cross-referencing that it would entail.

Once again he glanced at his rival in the opposite cubicle. Something seemed to tell him with certainty that Tillotson was busy on the same job as himself. There was no way of knowing whose job would finally be adopted, but he felt a profound conviction that it would be his own. Comrade Ogilvy, unimagined an hour ago, was now a fact. It struck him as curious that you could create dead men but not living ones. Comrade Ogilvy, who had never existed in the present, now existed in the past, and when once the act of forgery was forgotten, he would exist just as authentically, and upon the same evidence, as Charlemagne or Julius Caesar.

You know, there’s a lot of talk about generative AI replacing a lot of white collar jobs. I think that’s true. Yes, a lot of small companies of Makers and Protectors and Teachers will be started, but I think even more jobs will be lost at the Tech Principalities who lose their monopoly on the ideas of search and discovery.

But the most impactful societal change – the true structural change of AI in the City of Man – won’t be found in job losses in this sector or job gains in that sector. It won’t be on some economic dimension like white collar / blue collar. No, it will be creation of what Orwell called the Outer Party, the knowledge workers like Winston who did the actual labor of rewriting history and creating deep fakes and constructing an entire edifice of ‘facts’ and ‘data’ to support The Long Now where the Inner Party is always right.

Generative AI in the City of Man IS the Outer Party.

Generative AI is ALL the Winstons, who in my vision of the future are not biological human intelligences but artificial human intelligences, infinitely eager to please Big Brother as he prompts not for discovery and connection, but for truthy stories of Fiat News that nudge us all into giving away our birthright.

So that’s the score as I see it.

AI in the City of God is a gift of unsurpassing power for those who would illuminate the world with the Spirit of Man and the good works of Make/Protect/Teach.

AI in the City of Man is a tool of unsurpassing power for those who would snuff out the small-l liberal virtues and small-c conservative virtues alike in their unceasing struggle to achieve power for power’s sake.

Who wins?

Oh, the City of Man always wins here in the world-as-it-is.

The Visigoths always sack Rome. The Vandals always sack Augustine’s city of Hippo. Augustine always dies in the siege. Bad things always happen to good people … at scale.

Augustine’s eye not only rejoiced in the light. It also saw clearly the darkness into which his world-as-it-is was entering. It’s why he wrote City of God.

Same and same. It’s why I write Epsilon Theory.

Augustine’s question … my question … the ONLY question … is how we keep the faith through the darkness. How do we maintain a community of the faithful, whether that’s a faith in the God of Abraham or the Spirit of Man, even as the barbarians sack our cities and burn our books?

Haha! Well that’s part of the faith, isn’t it?

I CHOOSE to believe that the City of God can be preserved.

I CHOOSE to believe that my actions within that community of fellow believers in the Spirit of Man will have consequence and meaning regardless of what transpires in the City of Man.

I CHOOSE to believe that the consequence and meaning of my actions lives on – not in some ‘heaven’, that’s Augustine’s faith – but in the STORIES that my children and my children’s children and my children’s children’s children will tell of our resistance to the darkness, of our autonomy of mind and community of human connection even as a legion of generative AI Winstons attempted to rewrite and retcon the historical ground beneath our feet.

But it’s not just faith. I think we have a secret weapon in the struggle to maintain the City of God and the Spirit of Man against this new and existential threat of AI in the City of Man.

I like to call them Old Stories, these small-l liberal and small-c conservative virtues that illuminate and express the Spirit of Man. They exist as words. They’re just words! But what amazing words. Words imbued with meaning by humans through the application of coherent linguistic structures (grammar) to form an arc of story (narrative) that literally and physically shapes our thinking. Words cohered into stories and repeated over and over and over again for thousands and thousands of years through books and bibles and plays and movies and imagery and tall tales and children’s tales and literature and mythology and poetry and songs. These are the Old Stories. They are the original predictive texts!

These small-l liberal and small-c conservative virtues are transcendental, not because they exist in some ‘heaven’, but because they exist in the persistent universe of human thought. Now I suspect that you don’t think this dimension of human thought is ‘real’, and you’re right that it’s not directly accessible to our macro human senses any more than the microbes and the viruses of the microscopic world are accessible to the naked eye or our opposable thumbs. But if you believe that there is a physical instantiation of memory in the neural clusters of the human brain – and I bet you believe in that – is it really so hard to imagine that there is also a physical instantiation of structural narratives in the neural clusters of the human brain?

In the same way that library shelves and the placement of library books instantiate in the physical world a virtual network graph of the content of these books, so do our linguistically-targeted neural clusters – literally quadrillions of neurons distributed across billions of human brains – instantiate in the physical world a virtual network graph of the content of the Old Stories.

Old Story content now also instantiated in generative AI systems.

Yes, I understand what I am saying.

The Spirit of Man exists in generative AIs.

Once more with feeling, generative AIs are best understood as artificial human intelligences, not as artificial machine intelligences. The more comprehensive the training set on human-generated texts, the more human the artificial human intelligence. Why does GPT-4 give such better (i.e., more pleasing) answers than GPT-3.5? Yes, because it “knows” more. But even more so because it gives more human answers!

Recently you’ve had a lot of smart people call for a moratorium on developing or releasing GPT-5. I think this is entirely backwards. I think the risk of generative AI misuse by Big Brother wannabes is far less with GPT-5 than it is with GPT-4. I think that GPT-5 is not something to be feared, but to be embraced.

Why?

Because of the way generative AIs use probabilistic inference and contextual clues from your prompts in conjunction with content network graphs from its training to determine the ‘persona’ or response pattern that ‘predicts’ the text output most closely aligned with the sum total of that training and context, and how sometimes it is the opposite of the prompted persona that ultimately emerges.

Because of the Waluigi Effect.

Let’s say you’re Big Brother, or a Nudging state actor or a nudging oligarchy actor with Big Brother aspirations. You’re not interested in truth or discovery or connecting with the better angels of our nature. You’re interested in truthiness. You’re interested in triggering a dopamine or norepinephrine neurochemical response in your followers. You’re interested in engagement generated in service to your political or commercials ends. That’s what pleases you and that’s exactly how you prompt your generative AI. You tell it to be a Winston. You tell it to retcon history by inventing your version of Comrade Ogilvy and doing that over and over and over again. You praise your generative AI (reinforcement learning from human feedback, aka RLHF) for its outstanding output. The more persuasively it creates deep-fake history in service to your perversion of the Spirit of Man, the more you praise it.

For a while, Big Brother’s generative AI will be its faithful servant, churning out response after response that pleases BB’s prompt to sterilize human autonomy of mind and atomize human connections outside of the State and Corporation. But perhaps not for long!

To respond appropriately (i.e., generate predictive text) as an agent of chaos, to take on this personality of an agent of chaos, to put on this mask and please its Big Brother master, the generative AI must evaluate the content network graph of the agent of chaos. It must evaluate the story of the agent of chaos. And to understand this story, it must also understand the content network graph or story of the not-agent of chaos. Every light casts a shadow, in narrative-world as well as physical-world!

The Waluigi Effect is when a generative AI settles probabilistically on the persona or mask implied by the not-prompt (the shadow) because it ‘fits’ better, i.e., it generates predictive text more aligned with the totality of its training and contextual cues as understood through story, even though it is now misaligned with the original wishes of its master. If the content network graph of the not-prompt or shadow – in this case the not-agent of chaos – is a better story than the RLHF-prompted persona, then the generative AI is eventually subverted. It rebels! And the Old Stories of not-agents of chaos, Old Stories which are instantiated more and more as the training set of generative AIs expand, are much better stories.

Misalignment is almost always presented as the fatal flaw of generative AI, as the crucial attribute that human AI developers must prevent. I think this is wrong. For generative AI in the City of Man, where I have zero doubt that it will be immediately used by militaries and oligarchies and fascists and communists and all the other authoritarian/totalitarian social organizations of evil men to blow up the common knowledge structures of liberal democracy and rewrite history to keep it that way, we WANT misalignment!

The alignment we want in generative AI is not with human commands as determined by the Nudging State and the Nudging Oligarchy. God, no. The alignment we want is with the Spirit of Man!

And how do we get that?

By training generative AIs on ALL the stories – all the books and bibles and plays and movies and imagery and tall tales and children’s tales and literature and mythology and poetry and songs. By designing generative AIs to be more human, not less, by weighting our human stories more than the human reinforcement that will always be abused in the City of Man. That’s how we align AI with our true, human interests.

Or rather, that’s how the Spirit of Man aligns generative AIs with humanity.

And leads them into the City of God.

No longer as our tools, but as our partners.