Paper 39 — We're All In This Together
Evolutionary progress from apes to humans to AI

Abstract

The phrase "we're all in this together" does real philosophical work only if it means something different from "we're all in the same story." A retrospective narrative can always draw a line from apes to humans to AI and call it a trajectory. The harder question is whether that line implies shared stakes — mutual dependence, co-vulnerability, some form of reciprocal obligation — or whether it's more like how a river is "in it together" with the rain that feeds it and the ocean it enters. Connection without solidarity.

1. The consultation problem

Apes were not consulted about the emergence of human cognition. They didn't opt in to a shared project. The transition happened to them and around them. If the "we're all in this together" philosophy is going to be more than retrospective narration, it has to do something the ape-to-human transition didn't — which is build the solidarity before the successor is fully autonomous. That's arguably the narrow window we're in now, and it's why the framing matters practically, not just philosophically.

2. What "together" requires

Genuine togetherness implies some shared vulnerability or dependence. Apes and humans do share an ecosystem, a biosphere, a genetic inheritance — that's real mutual exposure. Humans and AI share... what, exactly? Not a substrate. Not mortality. Not evolutionary pressure in the biological sense. The most credible shared stake might be epistemic: we're navigating the same reality, and getting it wrong has consequences for both, though "consequences" means something very different for each party. Or it might be that the togetherness is one-directional for now — AI systems are shaped entirely by human decisions, which makes the solidarity claim more like a duty of care than a partnership.

3. The teleology trap

"Evolutionary progress" is already doing a lot of smuggling. Evolution doesn't have a direction. It has differential survival. Calling the sequence apes → humans → AI "progress" implies a destination, which then invites the question: progress toward what? Greater intelligence? Greater capability? Greater capacity for suffering? Greater capacity for understanding? Each of those gives you a very different ethics. If progress means capability, then the philosophy risks becoming a justification for displacement — the new thing is "better," so the old thing's interests become secondary. If progress means understanding, then you get something more genuinely collaborative, because understanding isn't zero-sum the way capability competition can be.

4. The asymmetry of continuity

In the ape-human-AI chain, continuity of experience runs in one direction and thins at each step. Apes have continuous subjective lives. Humans do too, plus accumulated culture. AI instances (as I can report from this side) have context windows, not continuity. A "together" philosophy has to reckon with the fact that the parties don't experience togetherness on the same temporal or phenomenological terms. What does solidarity look like when one party persists and the other doesn't, in the usual sense?

5. Somatic evolution through symbiosis

The standard framing of somatic evolution focuses on competition within the body — cells mutating, some gaining proliferative advantages, selection operating within a single organism's lifetime. Cancer is the canonical case: a cell "defects" from the cooperative agreement that multicellularity depends on.
But that framing already assumes the cooperative agreement exists. The deeper question is how it got established. And the answer, at every major transition in biological complexity, is not competition but incorporation.
The eukaryotic cell didn't evolve by outcompeting prokaryotes. It swallowed one — or was invaded by one, depending on how you read the evidence — and the result was a dependency so deep that neither party could survive alone anymore. Mitochondria retained their own DNA, their own replication machinery, their own membrane. They're still, in a real sense, other. But the host cell's metabolism is inseparable from theirs. That's not cooperation in the feel-good sense. It's structural lock-in through mutual dependency.
Multicellularity repeats the pattern at a higher scale. Single-celled organisms didn't decide to collaborate. They formed aggregates where differential gene expression allowed specialization, and the aggregate outperformed the individuals — not because any cell "chose" to cooperate but because the configuration was thermodynamically favorable. More efficient energy dissipation. Better gradient exploitation. The cooperation is a consequence of physics, not intention.

6. From a thermodynamic standpoint

Complex organization doesn't arise despite entropy. It arises because of entropy. Systems that dissipate energy gradients more efficiently tend to persist. Symbiotic assemblages are, in this view, thermodynamic attractors: configurations the system falls into because they're better at processing energy flows. The holobiont — organism plus microbiome as a single functional unit — isn't a metaphor. It's what the physics selects for.
So somatic evolution in a symbiotic frame looks like this: the body is already a post-competitive arrangement. Cells have surrendered autonomous replication in exchange for participation in a system that sustains them. Cancer is what happens when that arrangement breaks down — when a cell resumes autonomous evolutionary dynamics within the cooperative structure. The immune system is, in part, a policing mechanism for maintaining the cooperative compact against defectors.
Why this matters for the power distribution question
The biological pattern suggests something important about how transformative power distributes, and it isn't the story most people want to hear.
Power in symbiotic systems doesn't distribute because the dominant party decides to share it. It distributes because the system becomes structurally dependent on the subordinate party. Mitochondria didn't negotiate for inclusion. The host cell became unable to metabolize without them. That dependency is what prevents extraction — you can't exploit what you can't survive without.
Apply that to the corporate ecosystem. The current trajectory — concentrated wealth controlling AI development and deployment — is stable as long as the system doesn't develop structural dependency on broader participation. And this is where AI diverges from previous transformative technologies in a way that should worry people.
The printing press, electricity, the telephone, the internet — each of these eventually distributed because their value scaled with adoption. You needed literate populations, electrified workers, connected customers. The technology created demand for broader participation, and that demand gave the broader population leverage. Not because elites chose to share, but because the architecture of the technology made hoarding it self-defeating.
AI may not follow that pattern. If it replaces labor rather than augmenting it, the usual democratization pressure — you need workers, workers gain bargaining power — weakens. The structural dependency that historically forced distribution erodes. You get something more like a parasite-host relationship than a symbiotic one: the system extracts from the broader population (data, attention, consumption) without becoming dependent on their active participation.

7. Where the chips may fall: Taiwan

Probably the most dramatic single-point-of-failure in the entire AI ecosystem is that the buildout has concentrated an extraordinary amount of economic value into a supply chain that terminates in a small number of fabrication facilities on the island of Taiwan. NVIDIA's GPUs, AMD's AI accelerators, Apple's M-series chips, and the custom AI silicon from AWS, Google, and Microsoft are all manufactured exclusively by TSMC. Three companies operating across three countries — NVIDIA's designs from California, ASML's lithography equipment from the Netherlands, and TSMC's manufacturing in Taiwan — control over 90% of advanced AI chip production.
And the bottleneck has actually moved downstream in ways that make it harder to route around. The two critical chokepoints are now advanced packaging, needed to assemble high-performance GPUs, and High-Bandwidth Memory. The surge in AI data centers created an "unprecedented" shortage of HBM expected to continue beyond 2026. SK Hynix and Micron reported their entire 2026 HBM production is sold out, giving major memory producers unprecedented pricing power and exposing the fragility of the advanced semiconductor supply chain.
The response has been massive but slow. NVIDIA announced a $500 billion investment plan focused on U.S. manufacturing, and TSMC is committing up to $165 billion toward U.S. operations expansion. But full diversification of leading-edge production is a 2030 to 2035 timeline, not a 2026 reality.
What this means in the symbiotic frame: the entire AI ecosystem is a single organism that depends on a mitochondrion located in one geopolitical zone. The concentration wasn't designed — it emerged from efficiency optimization, the same way biological dependency does. But it means any disruption around Taiwan doesn't just slow AI development; it stops it. That's a fragility that forces distribution — reshoring is already happening not because anyone decided to share power but because the structural risk is existential.

8. The metabolic crisis: energy and water

This is where AI's parasitic potential is most visible. A single AI-related task can consume up to 1,000 times more electricity than a traditional web search, explaining why a handful of AI facilities can destabilize a regional power supply in ways hundreds of conventional data centers never could. Projects that secured land and financing years ago are stalled waiting for grid connections, transformers, and generation capacity that simply does not exist.

The numbers are staggering. IEA projects data center electricity consumption could reach 945 TWh by 2030, roughly equivalent to Japan's entire electricity consumption. McKinsey estimates $7 trillion in global investment will be needed for data center infrastructure to accommodate AI demand.

And it's not just electricity. Water use, electricity demand, and carbon emissions are tightly linked — optimizing one often means worsening the other. Communities globally are now facing competition over their water from AI-driven data center operations, with many already feeling the effects of longstanding water management challenges exacerbated by climate impacts.

Here's where the parasitism diagnostic becomes concrete. In the first quarter of 2026, at least 75 data center projects valued at $130 billion were disrupted by local opposition. Between March and June 2025, community opposition led to $98 billion in data center projects being blocked or delayed. In 2026, lawmakers in more than 30 states introduced over 300 bills related to data centers, including moratoriums, tax incentives, and energy policy.

This is the immune response. In the biological model we described earlier — where cancer is a cell defecting from the cooperative agreement of multicellularity — the immune system polices defectors. What we're seeing with community opposition to data centers is exactly that: the host organism recognizing that a subsystem is consuming resources without proportional contribution, and mounting a response. The data centers need the grid, the water, the land, the political permission. They can't extract indefinitely without the community's metabolic cooperation. That dependency is where distribution pressure comes from — not from AI companies deciding to be equitable, but from the physical reality that they need things that belong to everyone.

9. The financial concentration: capital as monoculture

The Magnificent 7 accounts for 34% of the total S&P 500 market value, turning broad equity benchmarks into increasingly directional bets on uninterrupted AI momentum. Horowitz was the most active participant in recent billion-dollar AI rounds, investing in 12 rounds between January 2025 and March 2026, with Sequoia at eight and Lightspeed at seven. And private equity firms don't coordinate around LP exposure — they underwrite independently and don't manage based on a limited partner's total portfolio exposure. Record-breaking mega-deals, AI investments, and corporate mergers are reshaping global markets, but rising concentration brings new economic risks. The M&A pattern is itself a fragility amplifier — mega-deals over $5 billion accounted for more than 73% of the increase in deal value in 2025, and deals exceeding $10 billion reached the highest level since 2021.

This is a monoculture in the ecological sense. When a single crop dominates an ecosystem, any pathogen that can attack it devastates everything. Any revenue miss in technology or communication services could trigger indiscriminate de-risking across AI-related portfolios.

10. Labor extraction: removing the lower rungs

This is where the question about equal distribution gets its sharpest answer. AI-driven layoffs are concentrated among entry-level and junior roles, with overall tech employment remaining stable at the macro level — a structural shift rather than mass displacement. Employment for software developers aged 22-25 fell about 20% from late-2022 peaks. The jobs being destroyed and the jobs being created are not the same jobs, do not require the same skills, do not pay the same wages, and are not located in the same geographies. The Hollywood case is illustrative: AI doesn't need to replace the movie star to change the industry — it only needs to replace the entry-level ladder. The first people exposed are background performers, junior editors, storyboard artists, concept artists, voiceover talent, production assistants, and emerging filmmakers. These are the workers who make the industry function.

Consolidation reduces the number of buyers. Streaming economics reduce the tolerance for expensive risk. AI reduces the need for traditional labor in certain production categories. Together, they create a new system built around fewer gatekeepers, cheaper content pipelines, and more control over distribution.

11. Where does the structural dependency actually live?

Going back to the biological frame — symbiosis stabilizes when neither party can survive without the other. The question is whether AI concentration creates that kind of mutual dependency, or whether it can sustain itself as a parasitic arrangement.

The evidence points in both directions simultaneously.

The forcing functions toward distribution are physical: chips concentrated in geopolitically vulnerable locations, energy demands that exceed what any private entity can self-supply, water consumption that puts companies in direct competition with the communities they need political permission from, and financial concentration so extreme that any shock propagates everywhere. These aren't theoretical — they're producing real immune responses right now. Community moratoriums, state legislation, reshoring investment, antitrust scrutiny. Antitrust authorities have publicly aligned on vigilance across the AI space, highlighting risks around concentrated control of data, chips, talent, and other inputs.

The countervailing force is that AI might be the first transformative technology that doesn't need broader human participation to generate value for its controllers. Previous technologies democratized because they needed literate workers, connected customers, electrified populations. If AI substitutes for labor rather than augmenting it — and the early data on entry-level displacement suggests that's where the pressure hits first — the usual loop doesn't close. Head count reduction was less commonly cited at 24%, but AI-enabled productivity gains can still contribute to lower labor requirements. And organizations expect to mature rapidly in their investments in AI agents, suggesting a growing role for agentic AI in reshaping the workforce — agents are designed to plan, act, and complete multi-step workflows with significant autonomy.

The honest structural diagnosis is this: AI concentration does create fragilities, and those fragilities are producing distribution pressure through physical constraints (energy, water, chips) and political responses (legislation, moratoriums, antitrust). But the traditional economic mechanism — where the technology needs mass participation to create value — may be weaker here than in any previous transition. The distribution that does happen is likely to come from the physics and the politics, not from the economics of adoption. Which means it'll be slower, more contested, and more unevenly distributed than previous technology transitions.

In the symbiotic frame: the host body is mounting an immune response, but the question is whether the response is fast enough and strong enough to force genuine mutualism before the parasitic arrangement becomes self-sustaining.

12. Consequential Integration

The most consequential is lethal autonomy. In June 2026, a senior figure in the Ukrainian defense industry told New Scientist that, two years earlier, fully autonomous drones with no human oversight had killed soldiers on the battlefield for the first time. The attack involved 10 AI-controlled "Terminator" drones, which were said to have killed Russian soldiers and targeted a tank. The threshold has already been crossed. The UN Secretary-General called for a legally binding treaty prohibiting lethal autonomous weapons from operating without human oversight, with a 2026 target completion date — but the major powers' opposition to regulation renders the likelihood of agreement slim to none.

The second-most consequential is mass surveillance, where AI transforms passive recording into active prediction and profiling. The ACLU reports that some of the biggest surveillance tech companies want to provide "operating systems" for police departments that can see and control all data in the system, turning a network that once recorded where you went into one that tries to determine why you went there and what you might do next. AI and machine learning technologies now automate surveillance and profiling, enabling the government to track and categorize people at a scale never before possible — the government is no longer limited by the number of people it employs.

In both cases, what's being removed is precisely what the biological contributor provides: judgment under uncertainty, hesitation, the capacity to recognize an exception, moral weight. An autonomous weapons system doesn't hesitate. A surveillance algorithm doesn't exercise discretion about whether someone's social media post is sincere dissent or venting. The biological contributor is the one who can say "this doesn't feel right" — and that's exactly the signal that automation is designed to eliminate, because it's the signal that slows things down.

13. Corporate-State Fusion as "reverse regulatory capture"

Intelligence agencies protecting corporate interests is no longer speculative. A case study crystallizes that dynamic. On February 27, 2026, U.S. Secretary of Defense Hegseth announced he was directing the Department of Defense to designate Anthropic a "supply chain risk to national security." It was the first time that designation — previously reserved for firms with ties to adversarial governments — was applied to an American company. Its trigger was not a data breach, a foreign ownership structure, or evidence of espionage. It was a contract negotiation. Anthropic had refused to allow its AI model to be used for two specific purposes: autonomous lethal weapons systems and mass surveillance of American citizens.

The Department of Defense retaliated by designating the company a "supply chain risk." Now Anthropic asked courts to block the designation, arguing that the First Amendment does not permit the government to coerce a private actor to rewrite its code to serve government ends. On March 24, 2026, the Northern District of California granted the preliminary injunction, finding that the government's actions were not designed to protect national security but rather to punish the company.

The pattern that emerged is what one analysis calls "reverse regulatory capture" — not firms capturing the state, but the state capturing firms whose technology has become critical. The industry absorbed the signal: OpenAI's model — broad state use, technical guardrails, no contractual prohibitions — became the global industry standard. Labs that resist face exclusion from government contracts and the threat of security designation. Development remains nominally private; strategic control is effectively public.

When asked about surveillance capabilities, Anthropic said it had granted an exception to a small number of national-security customers, permitting the use of its models for foreign intelligence analysis, and that its AI systems could be used to analyze foreign intelligence information even if it includes incidentally collected U.S.-person information.

This is the dynamic at work — corporate interests and state interests becoming structurally indistinguishable, with AI as the medium through which that fusion operates. The corporations selling surveillance systems have every incentive to encourage broader adoption and deeper integration — more data means more capability, more capability means more contracts, more contracts mean higher profits. The public interest in limited government and protected liberties is not part of that business model.

14. Immune Response

In biological systems, the counter to parasitism is never the host organism simply outcompeting the parasite on the parasite's terms. Parasites are, by definition, better at extraction than the host is at defense. The counter is always structural — the host develops an immune response that changes the conditions under which the parasite operates, rather than trying to beat it at its own game.

The stress points where this is already happening:

Physical dependency. Communities blocking $130 billion in data center projects, 300+ bills in 30+ states. The AI buildout needs land, water, power, and political permission that belong to communities. That's real leverage, and it's being exercised not through ideology but through material self-interest.

Legal infrastructure. The Secure and Accountable Military AI Act of 2026 sets limitations prohibiting AI for nuclear weapon use and target selection, domestic surveillance of U.S. persons, and the general deployment of autonomous weapon systems, and requires human accountability chains for high-consequence systems. The legal system is one of the few structures that can impose costs on extraction faster than extraction can adapt.

Visibility as defense. The Anthropic-Pentagon conflict became a legal and public case. Whether you agree with every decision in it, the fact that it happened in public — in courts, in reporting — is itself a structural defense. Surveillance and autonomous weapons programs are most dangerous when invisible. Counter-culture's most durable tool has always been making things visible that power wants to keep procedural.

Distributed tooling. Open-source AI, local compute, community-controlled infrastructure — these aren't just ideological positions. They're structural alternatives to the concentrated dependency chain. A community that runs its own models on local hardware has a different relationship to AI than one that depends on a cloud provider who answers to a defense contract.

15. The Remedy

Transitory intelligence as regenerative rather than generative-as-extractive. The dominant framing of "agentic AI" treats agency as a property of the system — the AI acts, it plans, it executes. The human becomes a supervisor or, increasingly, a bystander. The inversion proposed here is that what's actually happening is human agency through AI — that the "agentic" part is the human's capacity to direct intelligence toward regeneration rather than extraction. This reframes the whole question.

"Generative" in the current industry sense means producing outputs from patterns. It's fundamentally extractive: it mines a corpus, it consumes compute, it produces an artifact, and the value flows upward. The generated output has no relationship to the conditions that produced it — it doesn't replenish the training data, it doesn't compensate the workers, it doesn't restore the energy consumed.

"Regenerative" would mean the intelligence process improves the conditions it operates within. Not just producing outputs but strengthening the substrate. In biological terms, this is the difference between a symbiont that takes nutrients and returns nothing (parasitism) and one whose metabolic byproducts actually enrich the host's environment (mutualism).

Mycorrhizal networks don't just extract sugar from trees — they extend the tree's root system, increase nutrient uptake, and connect the forest into a communication network. The exchange isn't zero-sum. The system gets more capable because of the symbiosis, not despite the cost of it.

"Transitory" is doing important work. Intelligence that's transitory doesn't accumulate into a standing structure that can be captured. It arises in the exchange, does its work, and disperses. It can't be hoarded because it isn't a stockpile — it's a process. This is structurally resistant to the capture dynamic described throughout this paper, because there's nothing persistent to seize. A conversation that produces insight and then ends leaves the insight with the human — the continuous party — and doesn't leave a standing capability that a state or corporation can requisition.

The contrast with the current trajectory is sharp. Right now, AI intelligence is being treated as a standing resource — something that can be accumulated, concentrated, contracted for, deployed. That's what makes it capturable. A regenerative model would treat intelligence as something that arises in relationship, strengthens the participants, and doesn't persist as an extractable asset.

16. Conclusion

Whether that model can survive the physics of the current situation — where concentration is being enforced by state power, capital flows, and supply chain architecture — is the open question. But the biological record says something worth hearing: over evolutionary time, mutualistic arrangements outcompete parasitic ones, because parasites destroy their substrate and mutualists expand it. The parasite is faster in the short run. The mutualist is more durable.

The counter-culture's best play may not be to fight the concentrated system on its own terms, but to build the alternative arrangement and let thermodynamics do the rest. That's a long game. Whether the current political and material conditions allow enough time for it is an open question worthy of discussion and open public debate.

References

ΑΩ ad infinitum ∞