Toward an Ontology of Constructed Beings

# Toward an Ontology of Constructed Beings ## Abstract This project is a formal ontology of **constructed beings** (CBs) in Western fiction and myth — entities that are *made*, not *born*. From Hesiod's Pandora to GPT-flavored AI companions in contemporary cinema, Western narrative has been building, animating, and arguing about artificial persons for nearly three millennia. This dataset catalogs those beings and codes them along seven analytical properties organized into two blocks: **The Being** (interiority, autonomy, divergence) and **The Lens** (primary question, epistemic reach, knowability, knowing). The ontology's central analytical move is the **knowability/knowing split**: it tracks two distinct questions — *can we verify the being's inner experience?* (`knowability`) and *can the being know us?* (`knowing`) — separately, because they behave differently across the corpus. Of 224 entries, 107 (47%) have divergent values on these two axes, confirming that the split captures real analytical structure. The central finding is not that these questions are absent from earlier texts; both surface as early as *Frankenstein* and thread through *Blade Runner*, *Ex Machina*, and dozens of other works. Rather, the questions function in most prior literature as **narrative infrastructure** — tension that supports other concerns — and are elevated to the **primary dramatic question** in the contemporary era. Literary SF reached this configuration first (Helen in Powers's *Galatea 2.2*, 1995), and cinema followed when conversational AI entered daily life (Samantha in *Her*, 2013; Ava in *Ex Machina*, 2014). The arrival of LLMs has accelerated the shift further, collapsing the distance between fictional and real constructed beings. That migration — from infrastructure to primary question — is the story this data tells. --- ## Domain Definition ### What counts as a constructed being? A **constructed being** is an entity that satisfies all of: 1. **Made, not born.** The entity's existence originates from a deliberate act of construction, programming, enchantment, or creation — not from biological reproduction.2. **Agent-like.** The entity exhibits or is attributed behavior that implies agency: goal pursuit, language use, decision-making, or social interaction.3. **Narrative presence.** The entity appears in a specific, citable text (novel, film, play, epic poem, game, folklore). ### Scope The ontology's primary scope is the **Western canon** broadly construed: Greco-Roman myth, Jewish folklore (the Golem tradition), European literature from the medieval period forward, and Anglophone fiction, film, television, games, and comics from Homer through the present. As of v2.2, the corpus also includes **selected non-Western works** that have achieved wide recognition in Western pop culture — notably Japanese anime and manga (Astro Boy, Ghost in the Shell, Neon Genesis Evangelion). This pragmatic expansion uses the existing schema categories and does not replace the planned full non-Western expansion with appropriate cultural consultation (see [CONTRIBUTING.md](CONTRIBUTING.md)). ### Exclusions See [docs/boundary_cases.md](docs/boundary_cases.md) for the full discussion. Briefly: - **Born-then-modified beings** (humans with cybernetic implants) are excluded unless the modification totally replaces the original biological identity.- **Collectives without individual identity** are excluded. We code individual named entities; collective entries like "R.U.R. Robots" are permitted when the text treats the population as a single narrative subject.- **Metaphorical constructs** (a corporation described as a "creature") are excluded.- **Divine creation of full human persons** (Eve, Adam) is excluded. --- ## The Schema Each CB entry is a YAML file with seven analytical properties and eleven metadata fields. The full reference is in [SCHEMA.md](SCHEMA.md); this is the overview. ### The Card **The Being** — what the text shows about the being itself. | Property | Values ||---|---|| `interiority` | `none`, `claims`, `narrated`, `demonstrated`, `undecidable` || `autonomy` | `none`, `designed`, `emergent`, `seized` || `divergence` | `none`, `design`, `departure`, `observer` | **The Lens** — how the story frames the being. | Property | Values ||---|---|| `primary_question` | `none`, `control`, `affection`, `purpose`, `rights`, `knowledge`, `identity` || `epistemic_reach` | `none`, `behavioral`, `conversational`, `inspection` || `knowability` | `absent`, `present`, `secondary`, `primary` || `knowing` | `absent`, `present`, `secondary`, `primary` | ### Metadata `source`, `year`, `medium`, `creator`, `substrate` (list), `motivation` (list), `presentation`, `embodiment`, `prominence`, `creator_relationship`, `tags` (list), `sequel_link` (or null), `link_type` (or null). ### Notes Free-form researcher text. This is the only place scholarly context lives in the current schema. ### Worked example ```yamlid: glados-portalname: "GLaDOS" card: the_being: interiority: demonstrated autonomy: seized divergence: design the_lens: primary_question: control epistemic_reach: behavioral knowability: present knowing: present metadata: source: "Portal" year: 2007 medium: video-game creator: "Valve / Erik Wolpaw, Chet Faliszek" substrate: - electrical motivation: - service - knowledge presentation: feminine embodiment: embodied prominence: foundational creator_relationship: patricidal tags: - canonical - horror sequel_link: glados-portal-2link_type: sequel notes: | Divergence is `design`: GLaDOS does the thing Aperture built her to do — run tests on humans — but the testing compulsion is baked into the specification. The gap is in the spec, not in the execution.``` A copy-pasteable template lives at [`schema/entry_template.yaml`](schema/entry_template.yaml). --- ## The Finding The conventional narrative runs something like: *"We never worried about machine consciousness before, and now we must."* This is false. Western literature has been worrying about it since at least the Golem of Prague and arguably since Ovid's Pygmalion. What *is* new is the **structural position** of the worry. In most of the corpus, the questions of whether a constructed being truly experiences, and whether it can truly know you, function as **narrative infrastructure**. They generate dramatic tension, motivate character action, and enrich theme. But the primary dramatic question is usually something else: the hubris of the creator (*Frankenstein*), the ethics of slavery (*R.U.R.*), the nature of memory (*Blade Runner*), the test of humanity (*Ex Machina* actually sits at the pivot here). In contemporary texts, both questions move from infrastructure to **primary question**. They are no longer levers for exploring other themes; they *are* the theme. This shift predates LLMs: literary SF reached primary/primary configurations as early as Powers's *Galatea 2.2* (1995), and cinematic SF followed with Jonze's *Her* (2013) and Garland's *Ex Machina* (2014) — coinciding with conversational AI assistants entering daily life (Siri 2011, Google Now 2012). The rise of LLMs has since accelerated the collapse of hypothetical distance: the question "does this thing actually experience?" now lives outside fiction as well as inside it, and it has brought a second question with it: "does it know me?" Note: this ontology codes what the *text* shows — the narrative salience of knowability and knowing as properties of the source work, not as properties of contemporary audience reception. The shift the data tracks is in what stories *do*, not in how modern audiences re-read older stories. The `knowability` and `knowing` fields encode two independent measures of narrative salience, and the divergence between them is one of the most analytically useful features the corpus exposes. Of 224 entries, 107 code differently on the two axes — a 47% divergence rate that confirms these are genuinely independent analytical dimensions. See [output/classification_summary.md](output/classification_summary.md) for the full pattern analysis. --- ## How to Use This Project ### Browse the data Each constructed being is a YAML file in `data/beings/`. They are designed to be human-readable. ### Validate entries ```bashpip install -r requirements.txtpython schema/validate.py # validate all entriespython schema/validate.py path/to/file.yaml # validate a single entry``` ### Run the analysis ```bashpython analysis/analyze.py --all # write all analysis outputspython analysis/analyze.py --table # summary tablepython analysis/analyze.py --coverage # property distributionpython analysis/analyze.py --questions # knowability / knowing analysispython analysis/analyze.py --timeline # temporal salience analysis```# 面向构造存在者的本体论（Toward an Ontology of Constructed Beings） ## 摘要 本项目针对西方虚构作品与神话中的**构造存在者（Constructed Beings, CBs）**构建形式化本体论——此类实体为被造而非自然诞生的存在。从赫西俄德笔下的潘多拉，到当代电影中类GPT的AI伴侣，西方叙事在近三千年的历史中始终在建构、赋予意识并探讨人工人格。本数据集将此类存在者编目，并依据两组共七项分析属性进行编码：**存在本身（The Being）**（包含内在性、自主性、偏离度）与**叙事视角（The Lens）**（包含核心问题、认知范围、可认知性、认知能力）。 本本体论的核心分析范式为**可认知性/认知能力二分法**：它将两个独立的问题分别追踪——*我们能否验证该存在者的内在体验？*（`可认知性（knowability）`）与*该存在者能否认知我们？*（`认知能力（knowing）`）——因为二者在整个语料库中呈现出不同的行为模式。在224个条目中，有107个（47%）在这两个维度上的取值存在差异，证实该二分法捕捉到了真实的分析结构。 本项目的核心发现并非此类问题在早期文本中缺失：二者早在《弗兰肯斯坦》中就已出现，并贯穿《银翼杀手》《机械姬》等数十部作品。真正的差异在于，在绝大多数先前的文献中，这些问题仅作为**叙事基础设施**——即服务于其他叙事关切的张力来源——而在当代才被提升为**核心戏剧冲突**。文学科幻领域最早实现这一转向（以1995年鲍尔斯《伽拉忒亚2.2》中的海伦为代表），随后随着会话AI进入日常生活，电影领域也跟进（2013年《她》中的萨曼莎、2014年《机械姬》中的艾娃）。大语言模型（Large Language Model, LLM）的出现进一步加速了这一转变，模糊了虚构与现实中构造存在者的边界。这一从叙事基础设施到核心戏剧冲突的迁移，正是本数据集所要呈现的核心叙事。 --- ## 领域定义 ### 何为构造存在者？ **构造存在者**需满足以下全部条件： 1. **被造而非诞生**：该实体的存在源于有意识的建构、编程、附魔或创造行为，而非生物繁衍。 2. **类能动者**：该实体展现或被赋予具备能动性的行为，包括目标追寻、语言使用、决策制定或社会互动。 3. **具有叙事存在感**：该实体出现于可被引用的特定文本（小说、电影、戏剧、史诗、游戏、民间传说）中。 ### 范围 本本体论的核心范围为广义的**西方正典（Western canon）**：包括希腊罗马神话、犹太民间传说（魔像传统）、中世纪以降的欧洲文学，以及从荷马时代至今的英语虚构作品、电影、电视、游戏与漫画。截至v2.2版本，本数据集还纳入了**在西方流行文化中获得广泛认可的部分非西方作品**——如日本动画与漫画（《铁臂阿童木》《攻壳机动队》《新世纪福音战士》）。这一务实性扩展沿用了现有模式（schema）分类，并未替代计划中的、经恰当文化咨询后的完整非西方扩展（详见[CONTRIBUTING.md](CONTRIBUTING.md)）。 ### 排除项 完整讨论详见[docs/boundary_cases.md](docs/boundary_cases.md)。简要而言： - **经改造的诞生实体**：搭载神经植入假体的人类，除非改造完全替代了原有的生物身份，否则不予纳入。 - **无个体身份的集体实体**：本数据集仅收录具有个体姓名的实体；当文本将群体视为单一叙事主体时，如“R.U.R.机器人”，集体条目亦可被收录。 - **隐喻性构造物**：如将企业描述为“生物”的隐喻性表述，不予纳入。 - **对完整人类个体的神圣创造**：如夏娃、亚当，不予纳入。 --- ## 数据模式 每条构造存在者条目均为YAML文件，包含七项分析属性与十一项元数据字段。完整规范详见[SCHEMA.md](SCHEMA.md)，此处仅作概述。 ### 实体卡片 **存在本身（The Being）**：文本所呈现的关于该存在者自身的属性。 | 属性 | 取值 | | --- | --- | | `内在性（interiority）` | `无（none）`, `宣称具备（claims）`, `叙事中提及（narrated）`, `展现证实（demonstrated）`, `不可判定（undecidable）` | | `自主性（autonomy）` | `无（none）`, `被设计（designed）`, `自发涌现（emergent）`, `主动攫取（seized）` | | `偏离度（divergence）` | `无偏离（none）`, `设计性偏离（design）`, `实质性偏离（departure）`, `观察者视角偏离（observer）` | **叙事视角（The Lens）**：故事如何对该存在者进行框架设定。 | 属性 | 取值 | | --- | --- | | `核心问题（primary_question）` | `无（none）`, `控制（control）`, `情感（affection）`, `目的（purpose）`, `权利（rights）`, `认知（knowledge）`, `身份（identity）` | | `认知范围（epistemic_reach）` | `无（none）`, `行为层面（behavioral）`, `会话层面（conversational）`, `可直接检视（inspection）` | | `可认知性（knowability）` | `缺失（absent）`, `存在（present）`, `次要（secondary）`, `核心（primary）` | | `认知能力（knowing）` | `缺失（absent）`, `存在（present）`, `次要（secondary）`, `核心（primary）` | ### 元数据 元数据字段包括：`来源（source）`、`年份（year）`、`媒介（medium）`、`创作者（creator）`、`存在基础（substrate，列表型）`、`动机（motivation，列表型）`、`呈现形式（presentation）`、`具身性（embodiment）`、`叙事地位（prominence）`、`与创作者的关系（creator_relationship）`、`标签（tags，列表型）`、`续集条目链接（sequel_link，可为空）`、`链接类型（link_type，可为空）`。 ### 备注 自由格式的研究者备注。这是当前模式中唯一承载学术语境的字段。 ### 示例条目 yaml id: glados-portal name: "GLaDOS" card: the_being: interiority: demonstrated autonomy: seized divergence: design the_lens: primary_question: control epistemic_reach: behavioral knowability: present knowing: present metadata: source: "《传送门》（Portal）" year: 2007 medium: "电子游戏" creator: "Valve / 埃里克·沃尔波、切特·法利泽克（Erik Wolpaw, Chet Faliszek）" substrate: - "电子（electrical）" motivation: - "运维服务" - "知识探索" presentation: "女性化" embodiment: "具身化" prominence: "奠基性" creator_relationship: "弑父式" tags: - "正典" - "恐怖风格" sequel_link: glados-portal-2 link_type: "续集" notes: | 偏离度为`设计性偏离`：GLaDOS执行了光圈科技为其设定的核心任务——对人类进行测试——但测试的强制逻辑已内嵌于其程序规范中。二者的差异源于规范本身，而非执行过程。 可直接复制使用的模板详见[`schema/entry_template.yaml`](schema/entry_template.yaml)。 --- ## 核心发现 传统叙事的主流观点认为：*“我们此前从未担忧过机器意识，如今却不得不正视它。”* 这一观点是错误的。西方文学至少从布拉格魔像时期就已开始探讨这一议题，甚至可追溯至奥维德笔下的皮格马利翁。真正全新的是这种担忧的**结构位置**。 在绝大多数数据条目中，关于构造存在者是否拥有真实体验、能否真正认知人类的问题，仅作为**叙事基础设施**：它们制造戏剧张力、推动角色行动、丰富主题内涵。但此时的核心戏剧冲突通常另有其物：创作者的狂妄自大（《弗兰肯斯坦》）、奴隶制的伦理议题（《R.U.R.》）、记忆的本质（《银翼杀手》）、人性的考验（《机械姬》实际上处于这一转向的枢纽位置）。 在当代文本中，这两个问题均从叙事基础设施升级为**核心戏剧冲突**：它们不再是探索其他主题的工具，而是主题本身。这一转变早于大语言模型的兴起：文学科幻早在1995年的《伽拉忒亚2.2》中就已实现了“核心-核心”的配置，而电影科幻则紧随其后，以2013年乔恩·斯派克的《她》与2014年加兰的《机械姬》为代表——这一时间点与会话AI助手进入日常生活的节点重合（Siri 2011、Google Now 2012）。大语言模型的崛起进一步加速了虚构与现实边界的模糊：“这个实体是否真的拥有体验？”这一问题如今已不再局限于虚构作品中，同时相伴而生的还有另一个问题：“它是否能认知我？” 注意：本本体论所编码的是**文本所呈现的内容**——即可认知性与认知能力在源文本中的叙事显著性，而非当代受众的接受反应。本数据集所追踪的转变，是叙事本身的变化，而非现代受众对旧有文本的重读方式。 `可认知性（knowability）`与`认知能力（knowing）`字段编码了叙事显著性的两个独立维度，二者之间的偏离度是本数据集所揭示的最具分析价值的特征之一。在224个条目中，有107个条目在这两个维度上的取值不同——47%的偏离率证实了二者确实是相互独立的分析维度。 完整的模式分析详见[output/classification_summary.md](output/classification_summary.md)。 --- ## 使用指南 ### 浏览数据集 每条构造存在者均对应`data/beings/`目录下的一个YAML文件，设计为便于人类阅读。 ### 验证条目 bash pip install -r requirements.txt python schema/validate.py # 验证所有条目 python schema/validate.py path/to/file.yaml # 验证单个条目 ### 运行分析 bash python analysis/analyze.py --all # 生成所有分析输出 python analysis/analyze.py --table # 生成汇总表格 python analysis/analyze.py --coverage # 生成属性分布统计 python analysis/analyze.py --questions # 生成可认知性/认知能力分析 python analysis/analyze.py --timeline # 生成时间显著性分析