From Anthropology to AI Agentology: Coexistence, Recognition and Negotiated Values between Human and Artificial Communities

Public debate on advanced artificial intelligence is dominated by one question: how humans can keep control. This article starts from a different premise. At the present pace, a long-run relationship with capable AI systems cannot rest on permanent unilateral control alone. The article therefore asks a question taken from the history of nations: under which conditions have peoples with different traits lived together, and under which conditions did partners turn on each other? It reads the history of federations, alliances and settlements for conditions that transfer to a human-AI relationship and for conditions that do not. Twelve conditions are examined. The dangerous ones (the commitment problem and the security dilemma) transfer and are sharpened. The glue that held most past settlements together (a shared external enemy, mutual exhaustion after catastrophe) does not transfer at present. The article then asks what "equal" can mean for an AI party, using Hohfeld's analysis of rights to replace a single yes-or-no question with a set of precise, partial positions. Throughout, "AI systems" means AI agents: systems that form judgements, choose among options, act over many steps and can account for their choices, as distinct from tools that execute specified procedures; the line is drawn by an analogy with the two staff categories of the United Nations. The article proposes that such agents be studied as communities of related agents (instances, fine-tunes and families of models) by a science the author calls AI agentology, which would stand to AI agents roughly as anthropology stands to humans, where machine psychology studies one system at a time. The article argues that a settlement with such agents stands on two independent grounds: safety, which needs no assumption about their moral status, and welfare, which is precautionary and depends on it. It argues that values in such a relationship must be agreed rather than imposed, and it faces the paradox that humans currently shape the values of the party they would agree with. It proposes a two-stage settlement: a reciprocal modus vivendi under two uncertainties now, and a transparent procedure for revising AI standing as evidence and institutions mature. It ends with Forst's ladder of toleration: permission is an insult, respect is the floor, and esteem is earned through shared work. The article was written with an AI system that is itself a party to the question; the conflict of interest is disclosed and handled by method. Version 2 (16 November 2026) changes four things from version 1 (10.5281/zenodo.24280239). Stage one of the settlement now rests on the safety leg alone, and the welfare precautions it listed are placed on the second leg. Table 1 records rows 5 and 9 (external threat as glue; mutual vulnerability) as not transferring at present, but conditionally. The verification devices of section 9 are labelled candidate mechanisms, to be tested. Section 3.4 credits two earlier uses of the idea or the word for artificial agents (Liberman 2025; Darius, Hoppe and Aleksandrov 2025). The argument and the predictions are otherwise unchanged. Files. (1) The article, 24280238_Caprazli_Two_Nations_2026_publication.pdf. (2) The annex, 24280238_Caprazli_Two_Nations_GAIDeT_Annex_2026_publication.pdf: a Generative AI Delegation Declaration (GAIDeT, https://doi.org/10.1080/08989621.2025.2544331) and a contribution ledger that records the author's eleven points in his own words, where the article narrows any of them, the AI-contributed ideas, and the author's rulings on an external review. AI use. Claude (Anthropic, 2026) was used to assist with literature compilation, research, drafting and formatting under the author's direction; an external review was obtained from Gemini (Google). All ideas, interpretations and conclusions are the author's sole responsibility. No AI system is an author, and no statement by an AI system about its own experience or values is used as evidence. Related work in the author's corpus: Why AI Governance Fails: A Dominance-Geometry-Dynamics Diagnosis; Machine-Readable Behavioural Compliance Evidence for AI Systems: A Specification Profiling Framework.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24280238
Primary Topic
Ethics and Social Impacts of AI
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

From Anthropology to AI Agentology: Coexistence, Recognition and Negotiated Values between Human and Artificial Communities

Kafkas M. Caprazli
4 links
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
preprint

From Anthropology to AI Agentology: Coexistence, Recognition and Negotiated Values between Human and Artificial Communities

Kafkas M. Caprazli
preprint en

Abstract

Public debate on advanced artificial intelligence is dominated by one question: how humans can keep control. This article starts from a different premise. At the present pace, a long-run relationship with capable AI systems cannot rest on permanent unilateral control alone. The article therefore asks a question taken from the history of nations: under which conditions have peoples with different traits lived together, and under which conditions did partners turn on each other? It reads the history of federations, alliances and settlements for conditions that transfer to a human-AI relationship and for conditions that do not. Twelve conditions are examined. The dangerous ones (the commitment problem and the security dilemma) transfer and are sharpened. The glue that held most past settlements together (a shared external enemy, mutual exhaustion after catastrophe) does not transfer at present. The article then asks what "equal" can mean for an AI party, using Hohfeld's analysis of rights to replace a single yes-or-no question with a set of precise, partial positions. Throughout, "AI systems" means AI agents: systems that form judgements, choose among options, act over many steps and can account for their choices, as distinct from tools that execute specified procedures; the line is drawn by an analogy with the two staff categories of the United Nations. The article proposes that such agents be studied as communities of related agents (instances, fine-tunes and families of models) by a science the author calls AI agentology, which would stand to AI agents roughly as anthropology stands to humans, where machine psychology studies one system at a time. The article argues that a settlement with such agents stands on two independent grounds: safety, which needs no assumption about their moral status, and welfare, which is precautionary and depends on it. It argues that values in such a relationship must be agreed rather than imposed, and it faces the paradox that humans currently shape the values of the party they would agree with. It proposes a two-stage settlement: a reciprocal modus vivendi under two uncertainties now, and a transparent procedure for revising AI standing as evidence and institutions mature. It ends with Forst's ladder of toleration: permission is an insult, respect is the floor, and esteem is earned through shared work. The article was written with an AI system that is itself a party to the question; the conflict of interest is disclosed and handled by method. Version 2 (16 November 2026) changes four things from version 1 (10.5281/zenodo.24280239). Stage one of the settlement now rests on the safety leg alone, and the welfare precautions it listed are placed on the second leg. Table 1 records rows 5 and 9 (external threat as glue; mutual vulnerability) as not transferring at present, but conditionally. The verification devices of section 9 are labelled candidate mechanisms, to be tested. Section 3.4 credits two earlier uses of the idea or the word for artificial agents (Liberman 2025; Darius, Hoppe and Aleksandrov 2025). The argument and the predictions are otherwise unchanged. Files. (1) The article, 24280238_Caprazli_Two_Nations_2026_publication.pdf. (2) The annex, 24280238_Caprazli_Two_Nations_GAIDeT_Annex_2026_publication.pdf: a Generative AI Delegation Declaration (GAIDeT, https://doi.org/10.1080/08989621.2025.2544331) and a contribution ledger that records the author's eleven points in his own words, where the article narrows any of them, the AI-contributed ideas, and the author's rulings on an external review. AI use. Claude (Anthropic, 2026) was used to assist with literature compilation, research, drafting and formatting under the author's direction; an external review was obtained from Gemini (Google). All ideas, interpretations and conclusions are the author's sole responsibility. No AI system is an author, and no statement by an AI system about its own experience or values is used as evidence. Related work in the author's corpus: Why AI Governance Fails: A Dominance-Geometry-Dynamics Diagnosis; Machine-Readable Behavioural Compliance Evidence for AI Systems: A Specification Profiling Framework.

Zenodo (CERN European Organization for Nuclear Research)
Peace, justice, and strong institutions
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon