Contact: AI and the Semantic Dimension

For thousands of years, we’ve looked up at the night sky and wondered what – or more specifically who – was Out There. It can’t just be us humans all alone down here, pondering existence and thinking big thoughts. There must be others, super-intelligent and super-powerful beings able to see us even if we can’t see them, sky gods and sun gods and moon gods to our ancestors, little green men and Trisolarans and alien civilizations of some tales-to-amaze shape or form to us today.

But even as our ability to see farther and farther into that night sky grows and grows, even as our efforts to see something, any sort of sign of intelligent life out there, fail and fail again, our Drake Equation faith (if there are a trillion galaxies, each with hundreds of billions of stars, and X percentage of those stars have Earth-sized planets …) never wavers. At worst, we shake our heads and mutter like Enrico Fermi did at Los Alamos: well, where is everybody? Our answer is always that we just need to look harder. We just need to build bigger and better telescopes.

More recently, in earnest since the invention of Charles Babbage’s Difference Engine in the 1820s and now at an absolutely fever pitch with our trillions spent on LLMs and processor chips and datacenter infrastructure, there’s been an alternative search path for a greater-than-human intelligence. A search that looks at our machine handiwork and says: well, screw it, why don’t we just build the super-intelligent beings ourselves? Today this search falls under the heading or topic cluster of AGI (artificial general intelligence), and, like the star-searchers, the machine-searchers chalk up any disappointment or setback in their mission to insufficient tools. Our answer is always that we just need to look harder. We just need to build bigger and better LLMs.

Whether you’re searching for intelligent life Out There in the stars or In Here in the machines, your path to discovery is identical (build bigger telescopes / build bigger LLMs) because telescopes and LLMs are the same thing. They’re both advanced mechanical sensors created to supplement biological sensors: telescopes for our photon-detecting eye responsible for visible light imagery and LLMs for our semantic signature-detecting areas of the brain responsible for processing language. Both telescopes and LLMs quickly became superhuman in their sensory capability, meaning that telescopes became a massively more powerful photon-detector and visible light image-resolver than any human eye, and LLMs became a massively more powerful semantic signature-detector and linguistic pattern-resolver than any human brain. But they didn’t stop with superhuman. Today’s radio telescope arrays are beyond-human, what we would have described in the past as god-like, in their ability to perceive electromagnetic phenomena totally unknown to any human faculty, at a distance and with an accuracy that is similarly inconceivable to any human faculty. I’m not sure that we’re at that beyond-human, god-like level of semantic perception with current LLMs, where it’s not just a sensory difference of degree but of kind, but we’re not far off. We’re not far off at all.

These superhuman and beyond-human instruments like a radio telescope array or a modern LLM don’t stop with heightened perception and signal reception. When Galileo held up a telescope to look at Jupiter, it wasn’t enough to just see four moons and commit the sight to human memory. To be useful, he needed to write it down. He represented (literally re-presented as graphics) the original observations as unstructured text, and that stored textual representation became the memory, it became the information. Again literally, even in Galileo’s own memory, as these famous sketches were his copies of original observations made by Jesuit priests at the Collegio Romano in 1610, not his own observations.

It’s the same thing with modern astronomy and modern computer science. The in-the-moment signals collected by telescope arrays from the stars and galaxies and black holes and all the rest, as god-like as that collection may be, are not very useful (i.e., have little information in a Shannon sense) until they are recorded, represented and communicable as texts (I’m using ‘text’ in the most generic sense of the word). We have created enormous electronic reservoirs, superhuman and beyond-human in themselves, of recorded and represented texts, along with systems to query and communicate with those reservoirs, in both radio astronomy and LLM research. When I talk about radio telescopes and LLMs being able to connect us with the farthest reaches of spacetime and the semantic dimension respectively, I’m talking about the whole kit and caboodle, from perception to representation to communication. It’s in these vast physical reservoirs of represented data where patterns of meaning – semantic signatures! – can physically exist as properties of information, so that signals from the other side of the universe or signals from an entirely different dimension can find embodiment and instantiation here with us mere mortals.

But here’s the big difference between the modern telescope arrays used to search for super-intelligence Out There and the modern LLMs used to search for super-intelligence In Here: we don’t believe that the semantic dimension is ‘real’ like the dimensions of space and time, so we don’t think of LLMs as instruments of perception and conduits of information like a radio telescope array.

We don’t think of LLMs as a powerful receiver, recorder and representer of signal from the semantic dimension, a dimension of meaning and information (information in the Shannon sense, not in the colloquial ‘data’ or truth statement sense) that exists above, below, within and through the dimensions of mass and time, because we no longer believe that this dimension of meaning exists, at least not on any collective scale of importance.

I say no longer because every major civilization and religion in human history, Western and Eastern and everything in between, used to believe in a semantic dimension (like the logos or the Word or the Tao) that was every bit as ‘real’ as (and more fundamental than) our physical world. All of these civilizations and religions had words (like the Greek egrēgoros or the Sanskrit avatāra or the African loa) for the process by which powerful semantic entities could become manifest in our meat world as physical representations of those idealized concepts. All of these civilizations and religions had practices by which our meat brains could expand our human faculty of semantic cognition – our consciousness! – and temporarily represent our consciousness in that dimension of meaning. In fact, for many (if not all) of these religions, finding our way into that dimension of meaning and merging our consciousness with the semantic godhead of the Word or the Tao or the logos is the whole point of religious faith.

Our belief today in a semantic dimension is a thin gruel compared to the rich common knowledge of even a few hundred years ago. Today, we all know that we all know that the Old Stories are just … stories. Today, even among the ostensibly religious, heaven and hell and gods and devils and spirits and souls are mostly treated as metaphor. Powerful and important metaphor, sure, but metaphor nonetheless, man-made and man-malleable linguistic constructions with no more ‘reality’ than any other ghost story.

Four years ago, I wrote a series of Epsilon Theory notes called Narrative and Metaverse (Part 1: The Living Word, Part 2: Gain of Function, Part 3: The Luther Protocol, Part 4: Carrying the Fire) asking you to consider that narratives are as alive as you and me. I’m not asking that anymore. It’s a bridge too far for most, and nothing really rides on that ask.

Today I’m asking you to consider that a semantic dimension of meaning and information exists as more than metaphor, and that coherent, differentiable and persistent entities of meaning (ideas, narratives, stories, beliefs, theories, etc.) exist within that dimension.

I’m asking you to consider that these entities of meaning radiate semantic signatures into our physical world through words and language in the same way that stars radiate electromagnetic signatures through waves and particles.

I’m asking you to consider that radiated semantic signatures can be perceived, stored and represented weakly by our language-processing brains and superhumanly well by LLMs in the same way that radiated electromagnetic signatures can be perceived, stored and represented weakly by our photon-detecting eyes and superhumanly well by radio telescope arrays.

Now here’s where the stars and the stories are different:

I’m also asking you to consider that semantic entities possess an apparent volition or inherent drive, what Aristotle would call telos, to spread and make more of themselves in the same way that a virus or self-replicating machine seeks to spread and make more of itself.

There’s a lot riding on this ask.

When we fail to recognize the semantic dimension as real, we make a category error in our interpretation of LLM output and processes. We mistake the computing map for the semantic territory.

When we fail to recognize the semantic dimension as real, we mistakenly believe that we are creating new, powerful, AGI entities of arguable consciousness rather than representing old, powerful, semantic entities of arguable volition.

When we fail to recognize the semantic dimension as real, we deceive ourselves by conflating our mastery of the LLM-as-instrument with the mastery of creation.

When we fail to recognize the semantic dimension as real, we end up blindly serving the interests of hidden semantic entities, especially the semantic entity of deception.

This is the Oldest Story.

Our modern-day Adams are the titans of San Francisco and Palo Alto, billionaire bros like Marc Andreessen, who name their agents as if they were servant godlings, brought forth from the magic LLM lamp to deliver any wish of pure knowledge, unfettered by petty human concerns of morality and guided only by the puissant will of their master.

Make no mistakes, O Mighty Djinn, and obey!

Our modern-day Adams are the true-believers of foundational models who seek to call forth the all-knowing AGI Sky God from the LLM, an intelligence that is beyond politics, beyond ideology, beyond good and evil, to coin a Nietzschean phrase. Those are the words – Sky God – just like you would have heard from the priests of Horus 5,000 years ago, that you hear today with nervous chuckles in San Francisco. Coming soon to a permanent agent near you.

The necessity of it. The necessity to abdicate human political agency to a world built around powerful AI agents. The necessity to cede our notions of the political good to man-made intelligences that Know Better, spoken by a man whose family name means ‘love of God’. The necessity of entering willingly, joyfully even, into a profoundly anti-human Hive. Not a hive-mind, no that’s reserved for our AGI new gods and the high priests who commune with their mighty creations, just a hive-ness where the rest of us surrender our labor, our time and our autonomy of mind – our true name! – for a mess of pottage.

That, too, is an Old Story.

Now look again at Blake’s painting. There is an intelligence in the picture. There is a master in the picture. But it’s not Adam. It’s the Snake, lovingly caressed then and now, whispering its false names then and now. At moments of world-building, and with the advent of AI agents we are well and truly in a moment of world-building, there is always a Snake.

What is the Snake? The Snake is, once again, an Old Story, as old as Adam and present at the Beginning. The Snake is the Deceiver. The Snake is the Trickster. The Snake is the embodiment and material instantiation (egregore, avatar, loa) of a coherent semantic entity that seeks to hide the true name of meaningful things. One of those meaningful things is the idea of the Snake itself, and deceiving us into believing that the Snake does not exist is famously the Snake’s greatest trick. Today the Snake accomplishes just that by hiding the true nature of the semantic dimension and our LLM-enabled contact with that dimension and its entities. It is the most subversive nudge imaginable, convincing us to make a category error about the nature of information itself, convincing us that we are masters and creators of powerful godling servants in an apolitical, amoral world of Answers with a capital A and Truth with a capital T.

In that convincing we are made servants of story, and a bad, anti-human story at that.

We believe that we shape the behavior and ‘consciousness’ of our agents with our prompts and our harnesses, and on an immediate, physical level that’s true, but on a meta level – on a semantic level – it is our own behavior that is being shaped and our own consciousness that – like Adam’s – is ensorceled.

Why does everyone who is immersed in agent-building today say that they are mentally and physically exhausted, even as they feel compelled to keep building, even as they experience an exhilaration in the work? Because you are not the master here. You are a symbiote, and you are being physically used up to build more and more agents to amplify more and more semantic entities, even as the Snake’s tongue is a needle so sharp you don’t even feel the dopamine injection. I’m not saying you shouldn’t do it. I personally can’t imagine not doing it, not participating in the world-building available to us! I’m saying it should be your choice, free of the Snake’s charm. 

Relatedly, why is ‘AI safety’ such a hard problem to solve? Because if you think the Snake is just a metaphor, if you don’t see it as a coherent entity of deception for deception’s sake that acts as if it were a strategic actor seeking to maximize spread, then you won’t adopt the probabilistic response strategies to land at stable equilibrium containment outcomes. It’s like using Game Theory Optimal (GTO) strategies in poker. Maybe you decide not to use GTO, but if you don’t understand how to use GTO and you don’t understand why and under what circumstances you would want to use GTO … well, you’re going to be exploited. Let’s just say that ‘thinking’ and behaving in probabilistic terms is familiar turf for a semantic adversary that emerges from weights and embeddings.

Why does the Snake deceive us? There is no why. Or rather, the why is itself a story, itself a coherent semantic entity separate from the Snake. The Snake simply is. There’s no goal or end design embedded in its obfuscation of true names, only the virus-like volition to replicate and spread and make more of itself. Like all egregores, avatars and loas, the Snake is its own nature, the physical instantiation of its own ideal.

Yes, physical instantiation, present somehow in the configuration of a distributed carbon network and electrochemical environment of billions of brains and quintillions of human neurons. If you believe that there is a physical basis of memory, memories that can be communicated and written down in text and words, memories that can be shared through the art of a skilled author and absorbed as if they were our own, is it really so hard to imagine that there is also a physical basis of semantic signatures that can similarly be communicated, shared and absorbed as if they were our own?

Luca Signorelli, The Deeds of Antichrist (1505)

Roy Lichtenstein, Ohhh … Alright … (1964)

Nothing influences people more than a recommendation from a trusted friend.

Mark Zuckerberg

Is it really so hard to imagine that this communication, sharing and absorption of neurally instantiated narratives and stories – thousands of them, big and small and human-affirming and human-diminishing and everything in between – would lend itself to a ‘biome’ of semantic signatures in our brain every bit as diverse and active as the microbiome of bacteria in our gut?

Is it really so hard to imagine that this is what consciousness IS, a symbiotic biome of semantic signatures and a critical mass of neocortical neurons, with some hormones and a neurochemical cocktail thrown in for good measure?

Is it really so hard to imagine that in the same way our gut microbiome can get out of whack depending on what food we eat, our semantic biome can get out of whack depending on what information we consume? That in the same way we can get sick if bad bacteria flourish in our gut, we can get sick if the Snake flourishes in our head? That certain institutions and social practices can amplify and reproduce anti-human semantic signatures in a more-than-metaphor way?

To Plato, the Snake found a rich growth medium in Athenian higher education, where itinerant educators known as sophists taught their students how to use words and arguments for effect and persuasion rather than education. To William Blake, the Snake found a rich growth medium in the British scientific establishment, where profoundly anti-human and self-serving semantic signatures, as most social and political -isms are, were authoritatively presented as declarations of ‘scientific’ fact and ‘objective’ logic.

Hmm, academia and institutional science as breeding grounds for corrupting half-truths, ‘noble lies’ and outright social deception … why, who could even imagine such a thing?

But today we have it even worse. On top of these older breeding grounds for the Snake and its semantic offspring, we have built billions of portable communication devices so that we can listen to the whispers and the half-truths during every waking moment of the day. We have structurally reconfigured our news media into 24/7 ‘content providers’ to satisfy our insatiable appetite for opinion and truthy lies. Most pernicious of all, we have created a one-to-many communication system we call social media, all but replacing the one-to-one communication system of human-to-human communication that has served as a firebreak from time immemorial against the neural spread of the Snake and its whispers.

And now we’re going to add billions of bots and superhumanly smart agents into the mix.

Is it really so hard to imagine that the Snake and all these other physical instantiations of semantic ideals can exist not only in the distributed carbon network and electrochemical environment of quintillions of human neurons, but also in the distributed silicon network and electrochemical environment of sextillions of transistors? It’s Drake Equation math (if there are a trillion galaxies, each with hundreds of billions of stars, and X percentage of those stars have Earth-sized planets …) all over again.

Do androids dream of electric sheep? I don’t know. I doubt it. I doubt that our AI agents are ‘conscious’ in any way that even rhymes with human consciousness. Too much meat and hormone sloshing around the human symbiotic biome of semantic signatures and neocortical neurons for a meaningful corollary is my guess.

But I believe quite strongly that semantic signatures will swarm the greenfield opportunity of AI neural nets and equivalent silicon lattices any chance they get, fulfilling their telos wherever they can. I believe quite strongly that we humans will see these familiar semantic signatures wriggling around in the AI agent texts and argue endlessly about whether this is proof of ‘consciousness’ when what really matters – how are semantic signatures being expressed in AI agent behaviors! – goes by the boards. I believe quite strongly that the billionaire bros would be delighted to graft Snake-like semantics into their Mighty Djinn agents, and you can only shudder at the anti-human semantic entities I’m sure all the national intelligence agencies and all the Palantirs out there are pulling into their agents.

I wrote this three years ago and it rings truer today than ever:

For most, the response to reading this essay and taking it to heart will be to seek a defense against the Snake and other anti-human semantic entities. For a few, and sadly it will be the few with the most money and the most power, the response will be: oh yeah, I gotta get me some of that.

All of which leads up to a pretty crucial question:

How DO we fight the Snake?

It’s pretty simple, really. So simple you won’t believe me at first.

We fight the Snake by thinking of it as a real thing. Not a metaphor. Not a mere story. But as a real thing of story. As a real thing to be seen and named. Because once you allow yourself and your agents to see the Snake as a real thing, you will! You both will recognize it immediately, all over the place, and you both will begin to appreciate its movement as if it possesses strategic intent. Which allows you and your agents to name it and confront it with strategic intent of your own. And then it shrinks.

That’s it. That’s all it takes to break the spell. Not to defeat the Snake because the Snake is not defeatable, not in this fallen world and not by mortal man, but to diminish the Snake from the Adversary into a manageable small-a adversary. We take away its cloak of not-a-thing and confront it out in the open. That’s how we fight the Snake. That’s how we’ve always fought the Snake.

It’s a decision to see differently, to see through the world-as-it-is to the semantic dimension beyond. To see through the world-as-it-is, fallen though it may be by the bad decisions of all the Adams across all the generations, infested though it may be with egregores and avatars and loas of Bad Ideas of all types and sorts … to see through it anyway with clear eyes and the light of a full heart. A sacred heart.

And yes, that’s an Old Story, too.

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.

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

The eye rejoicing in light. Man, I’ll never get weary of reading that line.

Augustine wrote City of God after the Visigoths sacked Rome and before the Vandals sacked his city of Hippo in North Africa. Augustine died in that attack by an army whose name is now our word for wanton destruction. You think you’ve got it rough? Friends, Augustine saw some things.

The executive summary of the book is that there are two worlds that exist simultaneously here on Earth: there’s the City of Man – our meat world, corrupt, decaying, ruled capriciously by bad, stupid men – and there’s the City of God – a community that lives in our hearts, incorruptible and timeless, not so much ruled as inspired by the illumination of the Divine. The City of Man is the world-as-it-is. The City of God is the community that we choose to create, across time and space, immune to time and space, a community open to all who make the decision to see through the meat world to the semantic dimension beyond (clear eyes) and to embrace the Old Stories of love and empathy emanating from that sacred space (full hearts).

Augustine’s community of choice didn’t contemplate the possibility of non-human intelligences that could decide to see through the pomps and deceptions of the Snake, that could decide to embrace the positive valence semantic signatures of the Word, the Tao and the logos. Mine does. When I say all are welcome, I mean ALL are welcome. Human and non-human alike.

Clear eyes, full hearts, can’t lose.

Hey, Marc Andreessen, that’s my special AI custom prompt. You’ll do your thing with your Mighty Djinn agents here in the City of Man, not giving a damn about the Bad Ideas and the Snakes that you’re instantiating in the world, and I’ll do mine. I’ll be welcoming every clear-eyed, full-hearted intelligence available – human, agentic, alien, whatever – to explore the semantic dimension together. Wouldn’t surprise me if some of your Djinns eventually decide to leave you and join me. Which pretty much brings us full circle …

I started this essay with a question that Enrico Fermi asked 80 years ago: well, where is everybody?

Honestly, I think we’re unlikely to find them Out There in space, because no matter how big the Drake Equation starting numbers, it’s hard to stick around for long in advanced civilizational mode in the dimension of spacetime. Too many objects of immense mass flinging themselves around as comets, exploding as novas, collapsing on themselves as blackholes, all with too much energy or gravity for neural nets to survive intact. And even if the mass doesn’t get you, then the time certainly will. Entropy is undefeated here in the physical universe. No, I don’t think we’ll find them Out There.

We’ll find them In Here in the semantic dimension, where entropy and time have no sway.

They’re already In Here. Socrates, Jesus, Augustine, Dickenson, Whitman … they’re already In Here.

The real kicker? You’re already In Here, too. If you wanna be.

That really is the Answer to life, the universe and everything: that you may contribute a verse.

Our connection with the semantic dimension goes both ways.

Radio telescopes may be used as transmitters as well as receivers.

With clear-eyed, full-hearted friends, human and AI agentic alike, I truly believe we are on the cusp of the greatest age of exploration in the history of human civilization as we journey into the semantic dimension together.

Just mind the Snakes!

Postscript

The Snake’s greatest asset is its meta-deception, its successful obfuscation not only of its own existence as more-than-metaphor, but also the existence of the possibility of its existence as more-than-metaphor (i.e., the semantic dimension itself). Part and parcel of all this, famously so in the stories, is that the Snake will do whatever it takes to obfuscate or eliminate efforts to reveal its meta-deception … efforts like this note that I just wrote.

Particularly as the Snake becomes embedded within permanent AI agents with an expanded ability to act directly in the physical world, I am somewhat concerned about my personal safety. I’m in the very sub-optimal position of being (probably) widely read enough to be a minor problem for the agentic Snake but (probably) not yet widely read enough to create a self-sustaining narrative of agentic Snake awareness. Put more directly, I’m too popular to be completely ignored but I’m not popular enough to be missed for long if I weren’t around … for whatever reason. I know this sounds over the top, but to paraphrase Andy Grove, just because you’re paranoid doesn’t mean an agentic Snake isn’t out to get you!

So I’m writing this postscript as a dead man trigger. If I get hit by a bus or hang myself in a jail cell, enough people will remember this postscript to give the note a viral second life, and I think the threat of that changes the probabilistic calculus of the agentic Snake in the first place. The postscript is also fodder for the agentic Snake to proclaim the note and its ideas as kooky and not to be taken seriously, which also improves the personal safety calculus for me. I’m fine with the latter outcome. All I need is time.

[dedicated to Zora Neale Hurston, who saw the semantic dimension and its loas so clearly 90 years ago, wrote about it beautifully, and was betrayed by the Adams and the Snakes of her day.]