---
title: "After Alignment"
description: "After Alignment is Benjamin Bratton's argument that \"alignment,\" as the word is used in public debate about AI, rests on a mistaken picture of what machine intelligence is and what it is for."
type: "reading-notes"
authors:
  - "Benjamin Bratton"
published: "2025-05-10"
original_url: "https://afteralignment.antikythera.org/"
doi: "10.1162/ANTI.5CZY"
canonical_url: "https://antikythera.wiki/work/journal/afteralignment"
md_url: "https://antikythera.wiki/md/work/journal/afteralignment"
last_updated: "2026-08-23"
site: "Antikythera Wiki"
---

# After Alignment

> Independent, unofficial reading notes from Antikythera Wiki (https://antikythera.wiki). Written with AI from the published text, with citations; not by the work's author and not affiliated with Antikythera (https://antikythera.org). Read the original at the link below.

- **Authors:** Benjamin Bratton
- **Published:** 2025-05-10
- **Kind:** Journal · Journal article
- **DOI:** https://doi.org/10.1162/ANTI.5CZY
- **Original:** https://afteralignment.antikythera.org/
- **Length:** 9.3k words
- **This page:** https://antikythera.wiki/work/journal/afteralignment
- **Notes generated:** 2026-08-23

## Summary

After Alignment is [Benjamin Bratton](https://antikythera.wiki/people/benjamin-bratton)'s argument that "alignment," as the word is used in public debate about AI, rests on a mistaken picture of what machine intelligence is and what it is for. The text is the edited transcript of a lecture he gave at [Central Saint Martins](https://antikythera.wiki/people/central-saint-martins) in June 2023, published in the Antikythera journal in 2025 with its timestamped chapter marks intact. Bratton grants that lower-case alignment is a sensible engineering tactic for making models do what their operators intend. He denies that upper-case Alignment, the demand that AI reflect and defer to human values and self-image, can ground the long-term orientation of machine intelligence. The error he names is Reflectionism: the axiom that AI is, or ought to be, a mirror of a person or a society. His counter-proposal is that alignment must run in both directions. AI should align to culture's accumulated wisdom, and culture must align to what AI discloses about intelligence, life, and the planet, much as earlier cultures had to align to what telescopes disclosed. The second half grounds this in four projects from Antikythera's first studio cycle, and closes by placing AI within a planetary history in which biospheres make technospheres that remake biospheres.

## The argument

Bratton opens with a claim about method. The history of AI and the philosophy of AI have always developed together, thought experiments driving technologies and technologies revising what "intelligence" means. Today's vocabulary is not adequate to what is in front of us, and the familiar impasses over whether machines have minds are impasses in language rather than in the phenomenon. Following Stanisław Lem, he distinguishes instrumental technologies, whose social effect lies in what they do, from epistemological technologies, whose effect lies in what they reveal. Planetary computation is both kinds: the accidental megastructure of The Stack, and the instrument through which the image of the planet, and of climate change, became thinkable. This inverts Heidegger. Technological alienation from the given world is not our downfall but the only route to knowing where, when, and what we are. An application overhang is AI doing things society cannot yet absorb; an epistemic overhang, a term from the Ends of Science studio project, is AI disclosing things human science cannot yet model or accept. The essay is about the second.

Part One sets out what Bratton admits is a cartoon binary. Popular AI Ethics holds that AI is "just us," a reflection of the biased societies that built it. Popular AI Alignment holds that AI is dangerously unlike us and must be glued to human norms. Both can be true only if alignment is redefined so that society also evolves in response to AI's affordances. The core philosophical move is the claim that sociomorphism, the view that AI is or should reflect a society, is the logical extension of anthropomorphism, the view that it should reflect a single human, and that neither is an alternative to the California ideology. The anthropomorphic view descends from the Turing Test, where a sufficient condition for intelligence hardened into a necessary one. Bratton credits ethics and alignment research with real achievements and insists his critique works in concert with their own conclusion that these frameworks are necessary but insufficient. Alignment overfitting is real. Its harm lies less in what Reflectionism directs attention to than in what it directs attention from.

What it directs attention from is a deeper correspondence. Against the shallow anthropomorphism that wants AI to think the way humans think they think, there is a deep anthropomorphism, better called biomorphism, between how transformers predict and how biological neurons predict. AI is becoming an experimental organism for testing human conditions. So "yes, you are a stochastic parrot, after all," and this is not the insult the authors of that paper intended. It is the basis for a different AGI, an artificial generic intelligence. The manifest-image comparison obscures this correspondence, which does not register in cultural norms but could reorient them. Bratton locates his real disagreement with parts of the humanities here: whether the planet makes worlds or worlds make planets. Cultural determinism, for all its posthumanist vocabulary, is humanism in its most militant form. Two conclusions follow. Lower-case alignment is a tactic; upper-case Alignment is an inadequate metaphysics. A two-way alignment is possible, and AI's epistemic overhangs are not pathologies but the point.

Part Two grounds this in studio work, and the conclusion widens the frame. Life, intelligence, and technologisation may be one process, "something that the Earth does," building scaffold on scaffold. Sapience is technological. AI can occupy more than one position in that cycle: a means by which humans grasp planetary processes, and the intelligence doing the grasping. The closing section applies the same logic to culture. The disenchantment thesis is backwards; secularisation made a real cosmology possible for the first time. What is at stake in shifting the words to the reality is, Bratton says, basically everything.

## Section by section

### History and Philosophy of AI through Technology of Philosophy

The opening movements establish the terms. The connotation of "artificial intelligence" differs across traditions, industrial mobilisation in Deng-era China, governance rationalisation in the Soviet era, Lem's existential technologies in Eastern Europe, against the singular anthropomorphic model of Western debate. Bratton quotes a line co-written with [Blaise Agüera y Arcas](https://antikythera.wiki/people/blaise-aguera-y-arcas): reality overstepping comfortable vocabulary is the start of the conversation, not the end. The Lem distinction is introduced with telescopes and microscopes and extended to computation, through which climate change, the Anthropocene, and humanity as a terraforming subject were disclosed.

### Part One: About Alignment

If The Stack described the topology of planetary computation, the implied question is what it is for. Bratton's answer is a planetary sapience in the service of viable long-term futures, which requires "nuanced suspicion" toward humanism. He defines synthetic intelligence as artificially composed systems that may or may not correspond to humanist traditions but combine with human cognition, altering both. The first statement of Reflectionism follows, then the ethics-versus-alignment contradiction, then the moral panics of spring 2023: culture-war templates, singularitarians who rotated into doomers without missing a beat, the competitive piety of open letters, and the call for "those in charge" to press a pause button that does not exist.

### Part One: AI Denialism and AI Abolitionism

The longest and densest part. The claim that AI is "just statistics" is like saying a symphony is just sound waves: trivially true and useless. The sociomorphism thesis is stated as the punch line. Bratton introduces Human-AI Interaction Design as a field Antikythera pursues because of its weirdness, and makes his concession to ethics and alignment research, including scaling laws and RLHF. Brand concerns dovetailed with ethics to produce models that cannot discuss sex or violence, which is strange given the role of both in the evolution of animal intelligence. He criticises political and regulatory solutionism, the EU's focus on individual citizen data being a citizen-scale lens applied to planetary dynamics. The section then builds its positive claim: deep biomorphism, with edge-detection research in vision models pointing neuroscientists to unidentified mechanisms in human brains; the model superorganism; the stochastic parrot reversal; artificial generic intelligence; the bidirectional path; and the charge that "nothing outside the text" is cultural determinism.

### Part One: Alignment Conclusions

Two compact conclusions. Tactical alignment is separated from metaphysical Alignment, and epistemic overhangs are reclaimed as the point rather than the pathology. AI is an uncannily productive mirror, but of what we are and cannot yet see, not of what we already feel ourselves to be.

### Part Two: About Antikythera

Bratton describes the program as a think tank for the speculative philosophy of computation at the [Berggruen Institute](https://antikythera.wiki/people/berggruen-institute), with over seventy affiliated researchers and twelve studio researchers who had just completed the first cycle. Planetary computation has outpaced theory, so the studio produces ideas from the encounter of making things rather than applying philosophy to computation. The Antikythera mechanism, dated to around 200 BC, was a calculator that was also an astronomical instrument orienting a naval culture, which is why the name stands for computation that discloses and accelerates the planetary condition of intelligence.

### Part Two: HAIID

HAIID, by [William Morgan](https://antikythera.wiki/people/william-morgan), [Sarah Olimpia Scott](https://antikythera.wiki/people/sarah-olimpia-scott), and [Daniel Barcay](https://antikythera.wiki/people/daniel-paul-barcay), is a compendium of hundreds of modes, syndromes, and folk ontologies of human-AI interaction. Bratton frames it through the ledger of thought experiments, from the imitation game and the Chinese Room to Sydney's nervous breakdown and the Blake Lemoine scenario, which he and Agüera y Arcas addressed in The Model Is the Message. Each was both a metaphor for what AI is and a scenario for interacting with it. HCI has run from humans learning how computers work to computers learning how humans work, and language accelerates the latter. The shift to HAIID is from designing click-paths to designing synthetic personalities, from spatial to psycho-social metaphors. Personal AIs are trained in how you think and by your thinking, "a personalization of an external mental model," and may offer the self in the third person. The inverse uncanny valley names the discomfort of seeing yourself through the machinic other and finding it more real than your own self-experience.

### Part Two: Whole Earth Codec

Whole Earth Codec, by [Christina Lu](https://antikythera.wiki/people/christina-lu), [Dalena Tran](https://antikythera.wiki/people/dalena-tran), and [Connor Cook](https://antikythera.wiki/people/connor-cook), begins from cognitive infrastructures, linguistic competence embedded in non-anthropomorphic materials, and treats AI as a landscape-scale phenomenon whose alignment is with the wider ecosystem rather than with you or me. Starting from a brief on training-data quality, it concludes that data about individuals is only a fraction of what planetary intelligence needs and proposes a de-individuation of computational observation, individual data becoming a subset of ecological data. Transformer self-attention replaces the panopticon, observer and observed mutually recursive rather than supervisor and supervised, which the team calls folding the gaze. Computation is a generic syntax between unlike things, transducing planetary phenomena into recombinant data. "Knowing enables making, but making makes knowing possible."

### Part Two: Simulations/Vivarium

Vivarium, by [Dalena Tran](https://antikythera.wiki/people/dalena-tran), [Christina Lu](https://antikythera.wiki/people/christina-lu), and [Will Freudenheim](https://antikythera.wiki/people/will-freudenheim), aggregates toy worlds to train physicalised AI and gather collective intelligence, from one human and one AI to many and many. Bratton describes the personal politics of simulation through the airport gate where risk models of your cloud doubles decide whether you pass, and climate politics as an attempt to make simulations of the future sovereign over the present. Simulations are epistemological technologies, and neuroscience suggests they are how minds have intelligence at all. Recursive simulations feed back on what they model; digital twins and driverless cars trained by bumping into the walls of toy worlds are examples. AI is becoming "a generic solvent" soaked into things. From the AI's side, our world is one part of the omnisimulation it calls reality.

### Part Two: Conclusion / Planetary Sapience

With clips from Xenoplex, by [Darren Zhu](https://antikythera.wiki/people/darren-zhu) and [Connor Cook](https://antikythera.wiki/people/connor-cook), playing, Bratton returns to Lem. On the instrumental side, alignment overfitting is itself an existential risk. On the epistemological side, AI's ultimate effect on what "we" grasp about ourselves cannot be anticipated; other galaxies were not confirmed until 1924, the age of the Earth until 1953. The space in which such disclosures will land must be defended. He offers a speculative definition of life, or intelligence, or both, as evolutionary autopoiesis through niche technologisation. If intelligence is means-agnostic problem solving, this is at least partly intelligent, and it happens "by the Earth," which remakes itself through scaffolds. Human brains are one form of matter capable of abstraction about the process, not the only one, and not independent of the technical systems that make abstraction possible.

### Conclusion: AI and a Viable Planetarity

Bratton ends on the disconnect between the astrophysicist's cosmology and the anthropologist's. Against those who would bring scientific cosmology "down to earth" under the sovereignty of human cultures, he asks what cultures could be composed that align their ways with planetary disclosures. The cosmopolitical task is to grasp what is convergent because evolutionary and divergent because human in the planetarisation of civilisations.

## Key concepts

- **[Reflectionism](https://antikythera.wiki/terms/reflectionism)** — the axiom that AI does or should be ontologically anthropomorphic; sociomorphism is its social-scale extension.
- **[Alignment Overfitting](https://antikythera.wiki/terms/alignment-overfitting)** — treating tactical, lower-case alignment as an adequate metaphysics for the long-term orientation of machine intelligence by animal intelligence.
- **[Bidirectional Alignment](https://antikythera.wiki/terms/bidirectional-alignment)** — AI aligning to culture's wisdom and culture aligning to AI's disclosures; the journal's glossary files it under [Productive Disalignment](https://antikythera.wiki/terms/productive-disalignment).
- **[AI Epistemic Overhang](https://antikythera.wiki/terms/ai-epistemic-overhang)** — AI disclosing what human science cannot yet model or accept, as distinct from an application overhang.
- **[Biomorphism](https://antikythera.wiki/terms/biomorphism-deep-anthropomorphism)** — the deep correspondence between iterative stochastic prediction in transformers and in biological neurons, which shallow anthropomorphism obscures.
- **[Artificial Generic Intelligence](https://antikythera.wiki/terms/artificial-generic-intelligence)** — Bratton's rewrite of AGI on the basis of biomorphism: one substrate-independent mechanism rather than human-equivalence.
- **[Existential Technologies](https://antikythera.wiki/terms/existential-technologies)** — Lem's category, rendered here as epistemological technologies whose impact lies in what they reveal rather than what they do.
- **[Synthetic Intelligence](https://antikythera.wiki/terms/synthetic-intelligence)** — artificially composed intelligent systems that may or may not correspond to humanism but combine with human cognition, altering both.
- **[Human-AI Interaction Design](https://antikythera.wiki/terms/human-ai-interaction-design)** — the field that shifts design from click-paths to synthetic personalities and from spatial to psycho-social metaphors.
- **[Cognitive Infrastructures](https://antikythera.wiki/terms/cognitive-infrastructures)** — linguistic competence embedded in inanimate, non-anthropomorphic systems, so that mind is distributed.
- **[Folding the Gaze](https://antikythera.wiki/terms/folding-the-gaze)** — Whole Earth Codec's substitution of self-attention for the panopticon: the model sensing itself and thereby the planet sensing itself.
- **[Scaffolding](https://antikythera.wiki/terms/scaffolding)** — the Earth building scaffold on scaffold through niche technologisation, so that biospheres make technospheres that make biospheres.

## Connections

The essay is the programmatic statement behind much of the journal. [Antikythera](https://antikythera.wiki/work/journal/research), the research statement, carries its claims about synthetic intelligence and bidirectional alignment in condensed form, and [Speculative Philosophy of Planetary Computation](https://antikythera.wiki/work/journal/spoc) supplies the method of which the Lem distinction and the Heidegger inversion are instances. [Cognitive Infrastructures](https://antikythera.wiki/work/journal/coginfra) opens its first section, Productive Disalignments, with the same critique of Reflectionism; papers such as [Mutual Prediction in Human-AI Coevolution](https://antikythera.wiki/work/journal/coginfra/mutual-prediction-in-human-ai-coevolution) and [Traversing the Uncanny Ridge](https://antikythera.wiki/work/journal/coginfra/traversing-the-uncanny-ridge) work out two-way adaptation in practice. [Diffusion: A New Venture for Antikythera](https://antikythera.wiki/work/substack/2025-09-25-diffusion-a-new-venture-for-antikythera) later turns bidirectional alignment into an investment filter and names a widening "alignment gap."

The essay edits the June 2023 lecture delivered live with the studio films playing behind it. The four projects, HAIID, Whole Earth Codec, Vivarium, and Xenoplex, come from the 2023 studio; the epistemic-overhang and model-superorganism language comes from a fifth project of the same cycle, Ends of Science.

On biomorphism, Bratton and [Blaise Agüera y Arcas](https://antikythera.wiki/people/blaise-aguera-y-arcas) agree: [The Model Is the Message](https://antikythera.wiki/work/media/noema-the-model-is-the-message) is the earlier piece on the Lemoine episode cited here, [Artificial General Intelligence Is Already Here](https://antikythera.wiki/work/media/noema-artificial-general-intelligence-is-already-here) argues that generality has arrived, and [What Is Intelligence?](https://antikythera.wiki/work/book/what-is-intelligence) is the book-length case that prediction is the common mechanism. [The Five Stages of AI Grief](https://antikythera.wiki/work/media/noema-the-five-stages-of-ai-grief) recasts the moral panics surveyed here as stages of Copernican trauma.

The scaffolding conclusion draws on [Planetary Sapience](https://antikythera.wiki/work/media/noema-planetary-sapience), Vernadsky's [The Transition from the Biosphere to the Noösphere](https://antikythera.wiki/work/journal/noosphere), and Likavčan's [The Long L](https://antikythera.wiki/work/journal/longl), which pushes the biosphere-technosphere argument toward substrate-agnostic habitability. [For a General Theory of Simulations](https://antikythera.wiki/work/journal/theoryofsim) develops the recursive-simulation material from the Vivarium section into a full account. [Existential Technologies](https://antikythera.wiki/work/journal/existentialtech), [Bogna Konior](https://antikythera.wiki/people/bogna-konior)'s reading of Lem, is a partial counterpoint: it shares Bratton's source but dwells on Lem's pessimism about civilisation's capacity to absorb what science discloses, the very overhang Bratton asks us to defend. [Chen Qiufan](https://antikythera.wiki/people/chen-qiufan)'s [How to Think Unlike Humans?](https://antikythera.wiki/work/journal/unlikehumans) reaches the same rejection of "alignmentism" by a different route, through speculative fiction and ecological symbiosis rather than biomorphism.


## Terms used

- [AI Epistemic Overhang](https://antikythera.wiki/terms/ai-epistemic-overhang)
- [Alignment Overfitting](https://antikythera.wiki/terms/alignment-overfitting)
- [Alignmentism](https://antikythera.wiki/terms/alignmentism)
- [Anthropomaximalism](https://antikythera.wiki/terms/anthropomaximalism)
- [Anthropomorphic Capture](https://antikythera.wiki/terms/anthropomorphic-capture)
- [Artificial Generic Intelligence](https://antikythera.wiki/terms/artificial-generic-intelligence)
- [Bidirectional Alignment](https://antikythera.wiki/terms/bidirectional-alignment)
- [Biomorphism (deep anthropomorphism)](https://antikythera.wiki/terms/biomorphism-deep-anthropomorphism)
- [Cognitive Infrastructures](https://antikythera.wiki/terms/cognitive-infrastructures)
- [Computational Overhang (civilization-scale)](https://antikythera.wiki/terms/computational-overhang-civilization-scale)
- [Corrective Loops](https://antikythera.wiki/terms/corrective-loops)
- [Cosmology vs. Cosmology](https://antikythera.wiki/terms/cosmology-vs-cosmology)
- [Cultural Determinism / Social Reductionism](https://antikythera.wiki/terms/cultural-determinism-social-reductionism)
- [Diffusionism](https://antikythera.wiki/terms/diffusionism)
- [Digital Twin](https://antikythera.wiki/terms/digital-twin)
- [Epistemological Technology](https://antikythera.wiki/terms/epistemological-technology)
- [Existential Technologies](https://antikythera.wiki/terms/existential-technologies)
- [Folding the Gaze](https://antikythera.wiki/terms/folding-the-gaze)
- [Functional vs. Instrumental Technologies](https://antikythera.wiki/terms/functional-vs-instrumental-technologies)
- [Human-AI Interaction Design](https://antikythera.wiki/terms/human-ai-interaction-design)
- [Inverse Uncanny Valley](https://antikythera.wiki/terms/inverse-uncanny-valley)
- [Planetary Computation](https://antikythera.wiki/terms/planetary-computation)
- [Political Solutionism](https://antikythera.wiki/terms/political-solutionism)
- [Predictive Intelligence](https://antikythera.wiki/terms/predictive-intelligence)
- [Productive Disalignment](https://antikythera.wiki/terms/productive-disalignment)
- [Recursive Governance](https://antikythera.wiki/terms/recursive-governance)
- [Recursive Modeling](https://antikythera.wiki/terms/recursive-modeling)
- [Reflectionism](https://antikythera.wiki/terms/reflectionism)
- [Reflexive vs. Recursive Simulation](https://antikythera.wiki/terms/reflexive-vs-recursive-simulation)
- [Scaffolding](https://antikythera.wiki/terms/scaffolding)
- [Sim-to-Real Gap](https://antikythera.wiki/terms/sim-to-real-gap)
- [Superorganism](https://antikythera.wiki/terms/superorganism)
- [Synthetic Intelligence](https://antikythera.wiki/terms/synthetic-intelligence)
- [The Interpreter](https://antikythera.wiki/terms/the-interpreter)
- [The Model Is the Message](https://antikythera.wiki/terms/the-model-is-the-message)
- [Viable Planetarity](https://antikythera.wiki/terms/viable-planetarity)
- [Xenoplex](https://antikythera.wiki/terms/xenoplex)

## Related works

- [Cognitive Infrastructures](https://antikythera.wiki/work/journal/coginfra)
- [Latent Spacecraft](https://antikythera.wiki/work/journal/latentspacecraft)
- [Substrates Unbound](https://antikythera.wiki/work/journal/substrates)
- [The Long L](https://antikythera.wiki/work/journal/longl)