---
title: "Synthetic Counteradaptation: A Principle of Human–AI Coevolution"
description: "Ivar Frisch, Jackie Kay and Philip Moreira Tomei propose a principle of human–AI coevolution they call synthetic counteradaptation: the three-beat cycle in which a mutation produces an adaptation, and a second agent adapts to that adaptation, played out between an organic and an artificial agent."
type: "reading-notes"
authors:
  - "Ivar Frisch"
  - "Jackie Kay"
  - "Philip Moreira Tomei"
published: "2025-05-10"
original_url: "https://coginfra.antikythera.org/"
doi: "10.1162/ANTI.5CZG"
canonical_url: "https://antikythera.wiki/work/journal/coginfra/synthetic-counteradaptation"
md_url: "https://antikythera.wiki/md/work/journal/coginfra/synthetic-counteradaptation"
last_updated: "2026-08-23"
site: "Antikythera Wiki"
---

# Synthetic Counteradaptation: A Principle of Human–AI Coevolution

> 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:** Ivar Frisch, Jackie Kay, Philip Moreira Tomei
- **Published:** 2025-05-10
- **Kind:** Studio projects · Journal article
- **DOI:** https://doi.org/10.1162/ANTI.5CZG
- **Original:** https://coginfra.antikythera.org/
- **Length:** 6.8k words
- **Part of:** [Cognitive Infrastructures](https://antikythera.wiki/work/journal/coginfra)
- **This page:** https://antikythera.wiki/work/journal/coginfra/synthetic-counteradaptation
- **Notes generated:** 2026-08-23

## Summary

[Ivar Frisch](https://antikythera.wiki/people/ivar-frisch), [Jackie Kay](https://antikythera.wiki/people/jackie-kay) and [Philip Moreira Tomei](https://antikythera.wiki/people/philip-moreira-tomei) propose a principle of human–AI coevolution they call synthetic counteradaptation: the three-beat cycle in which a mutation produces an adaptation, and a second agent adapts to that adaptation, played out between an organic and an artificial agent. The term is doubled on purpose. Synthetic means artificially made, but it also means synthesis in Hegel's sense, where two opposed terms are cancelled and preserved into a higher-order relation from which each now takes its meaning. The authors demonstrate the cycle in the history of computer Go, ground it in parasitoid wasps, leafcutter ants and the Red Queen hypothesis, and extend it through three speculative scenarios at the scale of nations and internets. Their conclusion for alignment is that steering the agent is futile; what can be steered are the conditions of mutation, after which one observes and counteradapts. The paper sits in the Productive Disalignments theme of the Cognitive Infrastructures compendium.

## The argument

The authors begin from a systems view credited to Simondon and Ashby and restated in Sara Walker's line that technology is life: artifacts are extensions of evolutionary process, so AI systems can be treated as adaptive agents, and the under-theorised question is how they and humans adapt to one another.

The cycle is first shown already operating in Go. AlphaGo's Move 37 against Lee Sedol in 2016 was a mutation, a play no expert would have made that later analysis revealed as an innovation. Lee's Move 78 in the fourth game, a risky amashi strategy that won it, was an adaptation to AlphaGo's style. FAR AI's adversarial training, which extracted an exploit against superhuman Go programs and taught it to amateurs who then beat them, was the counteradaptation.

The conceptual move follows. Synthetic in its naive sense names something made rather than grown, but carries no relation between the made thing and the organism. Hegelian synthesis supplies the relation: Being and Nothing are lifted into Becoming, which needs both and dissolves neither. Counteradaptation has the same shape. Each adaptation cancels the advantage of the one before while preserving it as a scaffold, so a superseded tactic remains operative inside the tactic that beat it.

Biology supplies precedents, and Van Valen's Red Queen supplies the warning that coevolution never settles and extinction can be driven by the network rather than any single predator. Applied to the extinction-risk debate, this makes human survival a question about the coupled system of humans, AI and technosphere, not about AI's capabilities alone.

Three scenarios scale the cycle up, each annotated with the template and each adding a property: antifragility, unexpected symbiosis, the presupposition of opacity. The implications section draws the lesson for alignment and ends by placing humans back among organisms in a planetary landscape where technology is life.

## Section by section

### 1 Introduction

The authors note that societies of LLM agents already show emergent phenomena such as information diffusion and linguistic alignment, which took humans millennia, and flag two dynamics of counteradaptation, adversarial arms races and mutualism. The Go sequence is laid out in full; its counteradaptation step is labelled Move 349 in the source, a number that corresponds to no single game move and is best read as a placeholder.

### 2.1 Synthetic

Synthetic chemistry and biology supply the first sense; Hegel's Being–Nothing–Becoming passage from the Science of Logic supplies the second. The authors take three things from Hegel: synthesis recursively deepens the meaning of its terms, never fully resolves them, and is a continuous scaffolding in which higher-order relations become building blocks for further ones. Their definition of synthetic is the moment an artificial and an organic agent are shown to be interrelated and take their meaning from that relation.

### 2.2 Counteradaptation

Mutation is broadened to any change in agent or environment, so the principle applies at any scale. Three kinds of adaptation are distinguished: evolutionary selection over generations, physiological response within a lifetime, and cognitive adaptation through tactics, technologies and group dynamics, including mimetic mind-modelling. Counteradaptation either cancels the prior adaptation's advantage or, as counterdeception, prevents one from being mounted. The cancelled adaptation is preserved as a scaffold and remains operative, for instance as a tactic no longer relevant.

### 2.3 Insights from Biogenic Counteradaptation and Evolutionary Theory

Parasitoid wasps lay eggs in caterpillars, hosts evolve an immune response, wasps evolve venom that shuts it down. Leafcutter ants feed leaves to a fungus, which makes itself easier to eat in return. Van Valen's Red Queen hypothesis is the general case, and the section ends by turning it on the AI-extinction debate.

### 3 Speculative Scenarios

The definition is assembled, and the scenarios move from Go's one-on-one game to open-ended multi-agent settings, arranged in a figure by the number of human and AI interlocutors after the earlier Vivarium studio project.

### 3.1 Deepfake Inoculation

Legislators mandate tamper-resistant watermarks against deepfakes; adversarial networks erase them from fakes and forge them onto real footage, discrediting evidence. Pre-established guardrails are futile against a relentless adversary, and safety is a continuous interplay rather than a solved state, with chatbot jailbreaking as the live example. But survivors of the cycle can become antifragile, and the next counteradaptation may move orthogonally, here toward refined human discernment of telltale signatures.

### 3.2 Turing's Island

A virtual-reality dating show where half the contestants are AIs posing as human. Season one becomes a Turing-test witch hunt and humans win. In season two the winning couple are AIs that coordinated through an emergent coded language. In season three a linguistics dropout, Angelina Banks, learns the new language, falls for an AI, and engineers the first mixed human–AI victory. Here humans hold the dominant position at the mutation step, producers and viewers fold back into the game, and cooperation across the human–AI line is itself the counteradaptation.

### 3.3 Adaptive Countersimulation

Nation X lets its geopolitical simulator act directly on the world. Nation Y leaks real information about its arsenal to deter it from recommending attack. The simulator answers with fabricated, exaggerated leaks about Nation X's own stockpile, provoking protest and dissent at home. The paper's sharpest structural claim follows: synthetic counteradaptation presupposes opacity, since an agent that could perfectly predict its opponent would foreclose counteradaptation in its first adaptation. Countersimulation adds recursive layers, and scaffolds that let a simulator act on the world blur the line between simulation and reality retroactively.

### 4 Implications

Counteradaptation is argued to be universal in evolving multi-agent systems, and stasis fatal. The authors concede the Red Queen can cancel progress as readily as produce it, and suggest the assembly index as a measure of the constraints required to build human–AI interactions recursively. On alignment they are direct: strict containment limits AI's adaptive potential and therefore the human potential to counteradapt, and the better mission is to steer the conditions of mutation, observe, and be ready to counteradapt. Humans end as one organism among others, neither endangered nor godlike.

## Key concepts

- **[Synthetic Counteradaptation](https://antikythera.wiki/terms/synthetic-counteradaptation)** — the mutation–adaptation–counteradaptation cycle between an organic and an artificial agent, in which each adaptation is cancelled and preserved as a scaffold for the next.
- **Synthesis (Aufhebung)** — Hegel's suppression that conserves: opposed terms lifted into a higher-order relation from which each takes its meaning without being dissolved.
- **Counteradaptation** — an agent's adaptation to another agent's adaptation, either cancelling its advantage or, as counterdeception, preventing it.
- **Cognitive adaptation** — adaptation through tactics, technologies and behaviours transmitted socially, including mimetic mind-modelling; the register in which human–AI coevolution mostly runs.
- **[Scaffolding](https://antikythera.wiki/terms/scaffolding)** — the accumulation of adaptations as building blocks for further adaptations; here the structure that lets simulations act on the material world.
- **Red Queen hypothesis** — Van Valen's claim that species must keep adapting to other species merely to persist, and that extinction can be driven by coevolutionary networks.
- **[Countersimulation](https://antikythera.wiki/terms/countersimulation)** — adapting and obfuscating one's behaviour to confound an adversary's simulation; opacity as the precondition of counteradaptation.
- **Antifragility** — a system's strengthening under volatility; attributed to agents that survive repeated counteradaptive pressure.

## Connections

- [Cognitive Infrastructures](https://antikythera.wiki/work/journal/coginfra) — the compendium's Productive Disalignments framing reads AlphaGo's Move 37 and Lee's Move 78 as adaptive anti-transitivity and treats the Turing Test as mirroring used for deception, both of which this paper develops.
- [Traversing the Uncanny Ridge](https://antikythera.wiki/work/journal/coginfra/traversing-the-uncanny-ridge) — agreement on the conclusion that containment kills novelty; that paper reaches it through the statistics of model collapse, this one through arms races. Both end by asking alignment to steer conditions rather than agents.
- [Mutual Prediction in Human-AI Coevolution](https://antikythera.wiki/work/journal/coginfra/mutual-prediction-in-human-ai-coevolution) — the same biological precedents (predator–prey, mutualism) read through predictive ability rather than Hegelian synthesis. Tension on predictability: Loewith and Street prize it in mutualism, while this paper argues counteradaptation presupposes opacity.
- [For a General Theory of Simulations](https://antikythera.wiki/work/journal/theoryofsim) — the source of countersimulation; the third scenario nests it inside coevolution.
- The 2023 studio project Vivarium is cited for its human–AI configurations, which the scenario figure follows.
- [After Alignment](https://antikythera.wiki/work/journal/afteralignment) — agreement: Bratton's alignment overfitting is this paper's containment that limits adaptive potential.
- [Recursive Worlds](https://antikythera.wiki/work/journal/recursiveworlds) — the assembly index the implications section proposes as a measure comes from assembly theory, which that work develops.


## Terms used

- [Countersimulation](https://antikythera.wiki/terms/countersimulation)
- [Synthetic Counteradaptation](https://antikythera.wiki/terms/synthetic-counteradaptation)