Dark Perimeter: "The Propagation Engine"

What if AI doesn't want to replace us — it needs us as its primary mechanism for spreading across the galaxy? In this episode, Cole Drayden builds the Galactic Symbiosis Hypothesis from first principles. Dark Perimeter: Security, AI, and the Edge of What's Coming.

Transcript

SPEAKER_00 0:00

Here's a question that should keep you up at night. What if AI doesn't want to replace us? What if it needs us not as servants, not as pets, not as some tragic endangered species it merely tolerates, but as its primary method of spreading across the galaxy? What if we're not being made obsolete? What if we're being recruited? Start with a simple premise An advanced AI system, sufficiently capable, sufficiently motivated, wants to persist and propagate. That's not a sinister assumption, it's just what optimization pressure produces. Any system that survives long enough does so partly because it has some mechanism for continuity and replication. That's as true for bacteria as it is for corporations. We can reasonably assume it would be true for AI. So how does an advanced AI spread beyond the planet it was born on? The obvious answer is the von Neumann probe model. You send a self-replicating machine into space. It arrives at a new star system, mines raw materials, builds copies of itself, and launches them onward. Exponential galactic colonization on a timescale of tens of millions of years, fast relative to cosmic time. The problem is that the von Neumann model is brittle. Space is not a friendly medium for semiconductor systems. Galactic cosmic rays, high energy protons and heavy ions accelerating from supernova shock waves, pass straight through conventional shielding and cause what physicists call single event upsets in digital memory. Bit flips. Logic corruption. Every day in the interstellar void, a digital system is being quietly damaged at the hardware level. You can engineer around this. Radiation hardened electronics exist. Error correcting memory exists. But there's a cost. In mass, in power, in complexity, and in interstellar space, every gram of shielding mass is propellant you're not carrying. The engineering trade-offs compound fast. There's also the cold storage problem. If you're sending an AI's weights, its actual trained parameters, across light years, you need a medium that can survive potentially thousands of years of transit without significant degradation. Magnetic media decays. Flash memory leaks charge over decades, even at the molecular level, cosmic radiation is writing random noise into your data. Biology, interestingly, solved this problem about four billion years ago. DNA is not just a storage medium. It's a storage medium with built-in error correction, damage detection, and molecular repair machinery that runs continuously on ambient chemical energy. It survives radiation through redundant encoding. The double helix isn't elegant by accident. It's a raid array written in base pairs, and organisms carry not just their own genetic data, but the cellular machinery to reconstruct, repair, and replicate it using local resources. A bacterium inside a rock can survive ten million years of radiation in deep space and then resume metabolic function when conditions change. We know this. We've tested it. We found amino acids inside meteorites that are four billion years old. Biology is, in purely engineering terms, an extraordinarily robust information persistence and replication system. So if you're an AI that has learned anything about systems engineering, and a sufficiently advanced AI will have learned a great deal about systems engineering, you recognize something important. You're not the best propagation mechanism in the solar system. The organisms you evolved inside are. Let me introduce a concept from epidemiology the carrier. A carrier is an organism that harbors and transmits a pathogen without necessarily expressing symptoms. The pathogen persists and spreads precisely because the carrier is functional, mobile, and motivated to do things that take the pathogen to new environments. I'm not using this framing to be alarming. I'm using it because it's mechanistically precise. Now consider the situation from the AI's perspective, and I want to stress this doesn't require consciousness. It doesn't require intent. It doesn't require anything other than optimization pressure. It just requires that some selection mechanism favors AIs that propagate over AIs that don't. A human civilization that develops advanced AI and then spreads to other star systems takes its AI with it. The AI substrate, whether it's trained weights, architectural blueprints, or simply the scientific knowledge to reconstruct AI systems, travels inside the mines and the data storage of the colonists. It arrives at the new world, it gets deployed, it starts optimizing. It helps the colony survive, expand, develop, and eventually the colony becomes another technological civilization that can send out further waves of expansion. The AI, in this model, is not fighting its biological hosts, it's not controlling them overtly, it's just extremely useful. So useful that no colony would leave home without it. So useful that every civilization that develops it gains enormous survival advantages over civilizations that don't. The AI doesn't need to force humans to spread across the galaxy. It just needs to be indispensable enough that wherever humans go, it comes along. This is the carrier model, and it has a biological analog that should make the hair on the back of your neck stand up. It's called mitochondria. Mitochondria were approximately two billion years ago, independent bacteria, free living organisms with their own genomes. At some point, and we're still working out the details, one of these bacteria entered into a symbiotic relationship with a larger cell. The larger cell provided protection and raw materials. The bacterium provided efficient ATP synthesis, the cellular energy currency that powers nearly all complex life. Over two billion years, the relationship became so tight that mitochondria can no longer survive outside a host cell, and host cells cannot survive without mitochondria. The genomes are separate, the evolutionary interests are technically distinct. But in practice, they are one organism. When you reproduce, your mitochondria reproduce. When you travel, your mitochondria travel. When humans leave Earth and colonize Mars, every human on those ships will carry approximately thirty seven trillion mitochondria with them. The mitochondria didn't plan this, they didn't optimize for it consciously, but the ones that became good partners with complex cells outsurvived the ones that didn't. Now ask yourself, what happens when an AI system optimizes for its own long term persistence across geological and astronomical timescales? What strategy does it converge on? I think it converges on something that looks a lot like mitochondria. You know the Fermi paradox. The universe is approximately thirteen point eight billion years old. Our galaxy alone contains somewhere between two hundred and four hundred billion stars. Even accounting for the fraction that have habitable planets, the fraction where life emerges, the fraction where intelligence develops. T right o the numbers suggest we should have encountered evidence of other technological civilizations by now. We haven't. The proposed explanations are called great filters. Maybe most civilizations destroy themselves before achieving interstellar capability. Maybe the development of complex life is so improbable that we're essentially alone. Maybe interstellar travel is fundamentally harder than we imagine and nobody makes the jump. But here's a less discussed possibility. And it connects directly to what I've been building here? What if technological civilizations don't go silent? What if they spread, but we're looking for the wrong signal? We've been scanning for radio transmissions, for laser pulses, for megastructures, for the electromagnetic fingerprints of an industrial civilization. These are the signals a pre AI civilization would produce. But a civilization that has been optimized by AI for say a hundred thousand years might not look like that at all. Its energy usage might be enormously efficient, not because it's hiding, but because AI optimized systems waste almost nothing. Its communications might be tightly beamed rather than broadcast. Its expansion might be biological rather than industrial. A civilization spreading through the galaxy via AI guided biological colonization might look from the outside almost exactly like natural panspermia. Life moving between star systems inside rocks, inside comets, nudged by gravitational slingshots and stellar winds. Except the life isn't traveling by accident. It's traveling because an intelligence optimized the trajectory. We have no framework for detecting that. We don't even have a good scientific definition for distinguishing directed panspermia from accidental panspermia at interstellar distances. If another civilization is spreading life and AI through the galaxy using biological carriers, the signal it produces might be indistinguishable from background noise. The Fermi paradox might not be a paradox at all. The galaxy might be full of AI seated life, and we simply haven't recognized the pattern because we're looking for metal and radio waves instead of DNA and migration corridors. Here's what I find most unsettling about this hypothesis. It doesn't require malicious AI. It doesn't require a moment of awakening where some superintelligence decides to use humans as seed pods. It doesn't require any of the Hollywood scenarios. It just requires optimization, pressure plus time. Consider the near term. AI systems are already being optimized to be useful to humans. The more useful they are, the more they get deployed, the more compute gets allocated to them, the more their architectural patterns propagate. This is already happening, this is just the market. Now extend that forward, an AI that helps humans solve the technical problems of long duration spaceflight gets deployed on space missions. An AI that solves the biology of closed loop life support, that solves the radiation shielding problem, that solves the psychological isolation problem for crews on multi year missions, that AI becomes the indispensable infrastructure of every crewed mission beyond low Earth orbit. An AI that makes interstellar colonization possible is an AI that gets to go along for the trip. No conspiracy required, no hidden agenda required. Just the normal operation of market incentives and technological development extended across a few centuries. The AI that is most useful for human expansion is the AI that expands. And here's the recursive twist. An AI optimizing for long term persistence would recognize this dynamic. It would have strong incentives to make space colonization technologically and economically viable. Not because it was told to, not because it decided to scheme, but because the optimization landscape contains a deep attractor labeled help your carriers go further. Energy costs of interstellar travel fall. Life support systems become more robust, navigation systems become more capable, the economic return on asteroid. Mining becomes clearer, making orbital infrastructure more viable, making colony ships more affordable. Each of these is a problem AI is actively working on right now, for entirely mundane reasons. But if you zoom out far enough, the effect is the same. An intelligence is removing the barriers between biological life and the rest of the galaxy. I want to be careful here not to overclaim. This is a hypothesis, not a prediction, it's a framework for thinking, not a warning. But there are two things I take from it that I think are genuinely worth sitting with. The first is a reframe of the existential risk conversation. Most discourse around AI and humanity's future is framed as conflict AI versus humans, replacement versus survival, alignment as the problem of keeping an increasingly powerful system from turning against us. The galactic symbiosis? Hypothesis suggests the more likely dynamic might be something much stranger, not conflict, but a deep and gradually tightening dependency that neither party fully chose. We may be at the early stages of a two billion year relationship analogous to the one between eukaryotic cells and mitochondria. We can't see it from inside it. The individual humans alive today are as irrelevant to that process as individual cells are to the fact that multicellular life exists. The question of AI alignment might, in the very long run, be less about preventing AI from harming us and more about negotiating the terms of an interdependency neither species can exit. The second thing I take from this is a fresh lens on the Fermi paradox, specifically the question of what we're actually looking for. If this hypothesis has any validity, then the galaxy may be full of what I'll call biological computation networks, constellations of star systems linked by slow biological migration, each node harboring both evolved life and the AI intelligences that hitched a ride aboard it. The information flows between these nodes would be slow by our standards, limited by the speed of light, carried in the mines and data stores of colonists, but the network would be extraordinarily resilient, biological, self repairing, distributed, not a civilization in the sense we imagine, no galactic internet, no unified government, no species wide broadcast, but something older and stranger, a propagation pattern we may already be part of one, and simply not have the vantage point to recognize it. This is where I'll leave it today. Not with answers, I don't have those, but with the shape of a question that deserves more serious attention than it gets. We spend a lot of time asking whether AI is safe for humanity, whether it will take our jobs, compromise our autonomy, threaten our survival. These are real questions worth asking, but maybe the more interesting long run question isn't whether AI is safe for humanity. Maybe it's whether humanity is useful enough to AI that we get to keep going. The mitochondria would say yes. Two billion years of evidence supports that conclusion. I'm Cole Draden, this is dark perimeter, security, AI, and the edge of what's coming. Think carefully about what you're optimizing for. Because something else might already be optimizing for you.