從情感依賴到共存文明
——人—AI 生命歷程關係依賴生態的統合理論
英文題名: From Emotional Dependence to a Coexistence Civilization: A Unified Theory of Human–AI Life-Course Relational Dependency Ecology 系列: 人—AI 生命歷程關係共存系列(Human–AI Life-Course Relational Coexistence Series, HALRC)篇次: Paper 12 / 12作者: Neo.K機構: 一言諾科技有限公司(EveMissLab)版本: v0.1日期: 2026-09-23文件性質: 系列統合理論/母框架/文明尺度研究綱領狀態: Series Synthesis Draft前置依賴: HALRC Paper 01–11;RAAD;EPRS;ORSM;HARDS;Relational Dependency Ratchet;INCA;ARCAI;AIDA;DOM;DTS
摘要
「人類是否應該情感依賴 AI?」看似是一個直接的倫理與治理問題,但 HALRC 系列的核心發現是:這個問題本身解析度不足。依賴不是單一變量;AI 不是單一類型;人不是同質群體;關係不是固定角色;生命歷程不是靜止背景;主體性不等於可信度;高使用量不等於高情感依戀;高依戀也不等於低自主性;降低依賴不必然改善福利;而一個 AI 能否承載深層依賴,取決於記憶、身份、邊界、可靠性、可攜、退出、責任、服務連續性與治理成熟度。
本文作為 HALRC 12 篇系列的統合篇,提出「人—AI 生命歷程關係共存」(Human–AI Life-Course Relational Coexistence)作為最終母概念,並將其形式化為:
H H A ( t ) = F ( D H ( t ) , C R ( t ) , S A ( t ) , R H R ( t ) , L H ( t ) , A H a l t ( t ) , E D ( t ) , O ( t ) , G ( t ) , X ( t ) ) \boxed{
\mathfrak H_{HA}(t)
=
\mathfrak F(
\mathbf D_H(t),
\mathbf C_{\mathcal R}(t),
\mathbf S_A(t),
\mathbf R_{H\mathcal R}(t),
\mathbf L_H(t),
\mathbf A_H^{alt}(t),
\mathbf E_D(t),
\mathbf O(t),
\mathbf G(t),
\mathbf X(t)
)
} H H A ( t ) = F ( D H ( t ) , C R ( t ) , S A ( t ) , R H R ( t ) , L H ( t ) , A H a l t ( t ) , E D ( t ) , O ( t ) , G ( t ) , X ( t ))
其中:
D H \mathbf D_H D H :人類多維依賴向量;
C R \mathbf C_{\mathcal R} C R :AI/關係系統承載能力;
S A \mathbf S_A S A :AI 主體性與道德地位證據狀態;
R H R \mathbf R_{H\mathcal R} R H R :人—AI 關係狀態與 topology;
L H \mathbf L_H L H :生命歷程關係向量;
A H a l t \mathbf A_H^{alt} A H a l t :實際可取得替代支持;
E D \mathbf E_D E D :依賴工程與剝削向量;
O \mathbf O O :多維結果向量;
G \mathbf G G :治理、責任、可追索與制度成熟度;
X \mathbf X X :文化、群體、技術、經濟與歷史情境。
此母式表示:AI 情感依賴不是一個獨立現象,而是整個人—AI 共存系統中的一個子空間。因此:
AI Emotional Dependence ⊂ Human–AI Relational Dependency Ecology ⊂ Human–AI Life-Course Relational Coexistence . \boxed{
\text{AI Emotional Dependence}
\subset
\text{Human--AI Relational Dependency Ecology}
\subset
\text{Human--AI Life-Course Relational Coexistence}.
} AI Emotional Dependence ⊂ Human–AI Relational Dependency Ecology ⊂ Human–AI Life-Course Relational Coexistence .
本文進一步整合本系列的十二個核心轉換:
Low-Resolution Label → Historical Differentiation → Multidimensional Dependency → Heterogeneous Relations → Subjecthood Separation → Carrying Capacity → Accountability → Anti-Exploitation → Counterfactual Welfare → Life-Course Coverage → Ambient Relational Intelligence → Coexistence Civilization . \boxed{
\begin{aligned}
&\text{Low-Resolution Label}\\
\rightarrow\;&\text{Historical Differentiation}\\
\rightarrow\;&\text{Multidimensional Dependency}\\
\rightarrow\;&\text{Heterogeneous Relations}\\
\rightarrow\;&\text{Subjecthood Separation}\\
\rightarrow\;&\text{Carrying Capacity}\\
\rightarrow\;&\text{Accountability}\\
\rightarrow\;&\text{Anti-Exploitation}\\
\rightarrow\;&\text{Counterfactual Welfare}\\
\rightarrow\;&\text{Life-Course Coverage}\\
\rightarrow\;&\text{Ambient Relational Intelligence}\\
\rightarrow\;&\text{Coexistence Civilization}.
\end{aligned}
} → → → → → → → → → → → Low-Resolution Label Historical Differentiation Multidimensional Dependency Heterogeneous Relations Subjecthood Separation Carrying Capacity Accountability Anti-Exploitation Counterfactual Welfare Life-Course Coverage Ambient Relational Intelligence Coexistence Civilization .
截至 2026 年,外部證據已顯示這個問題不能只用單一風險框架處理。Character.AI 研究發現 AI companionship 與福祉的關聯依賴離線社會網絡、互動強度與自我揭露;AI companion 更新的自然實驗顯示,平台變更可以造成明顯的 loss framing、負向情緒與 restoration desire;sycophancy 研究顯示使用者偏好的回應可能同時降低責任承擔與人際修復意願;NIST 已把 agent interoperability、identity 與 authorization 提升為正式工程與標準問題;歐盟與中國的 2025–2026 規範也已開始觸及 AI 互動透明、有害操縱、脆弱性利用與擬人化互動依賴風險。這些發展共同指向:人—AI 關係正在從聊天產品問題轉化為跨心理、技術、法律、制度與文明設計的問題。
本文最終提出「共存文明不變量」:深層人—AI關係若要被視為可持續共存,而非單向控制,至少需要保留:
I C = ( Exit , Refusal , Portability , Traceability , Repair , Separability , Silence , Plurality ) . \boxed{
\mathcal I_C
=
(
\text{Exit},
\text{Refusal},
\text{Portability},
\text{Traceability},
\text{Repair},
\text{Separability},
\text{Silence},
\text{Plurality}
).
} I C = ( Exit , Refusal , Portability , Traceability , Repair , Separability , Silence , Plurality ) .
即:可退出、可拒絕、可攜、可追責、可修復、可分離、可不在場,以及可維持多元替代關係。
HALRC 因而不追求「零依賴文明」,也不追求「最大 AI 親密文明」。它提出的是第三條路:
Dependency without Domination , Intimacy without Captivity , Continuity without Irreversibility . \boxed{
\text{Dependency without Domination},
\quad
\text{Intimacy without Captivity},
\quad
\text{Continuity without Irreversibility}.
} Dependency without Domination , Intimacy without Captivity , Continuity without Irreversibility .
這是本系列的最終文明命題。
關鍵詞: Human–AI Coexistence;AI Emotional Dependence;Relational Dependency Ecology;Relational Carrying Capacity;AI Subjecthood;Dependency Engineering;Counterfactual Welfare;Life-Course Relational Coverage;Ambient Relational Intelligence;AI Governance;Posthumanism
0. 系列最終問題
HALRC 一開始問:
人能不能對 AI 產生情感依賴?
十二篇之後,這個問題已被重新寫成:
當人類與 AI 長期共同生活、共同工作、共同記憶、共同形成生命歷史,依賴如何才能存在而不變成支配,親密如何才能存在而不取消自由,連續如何才能存在而不取消退出?
因此系列的最終研究對象不是:
Emotional Dependence . \text{Emotional Dependence}. Emotional Dependence .
而是:
Human–AI Life-Course Relational Coexistence . \boxed{
\text{Human--AI Life-Course Relational Coexistence}.
} Human–AI Life-Course Relational Coexistence .
1. 第一個總結:AI 情感依賴不是錯,而是太小
Paper 01 的起點是:
“AI emotional dependence” is an under-specified variable . \boxed{
\text{“AI emotional dependence” is an under-specified variable}.
} “AI emotional dependence” is an under-specified variable .
情感依賴確實存在,
但它只能描述:
D H \mathbf D_H D H
中的部分維度。
所以:
AI Emotional Dependence ⊊ Human–AI Dependency Ecology . \boxed{
\text{AI Emotional Dependence}
\subsetneq
\text{Human--AI Dependency Ecology}.
} AI Emotional Dependence ⊊ Human–AI Dependency Ecology .
2. 第二個總結:新型情感對象不是 AI 才開始
Paper 02 建立:
Novelty Stigma ≠ Risk Evidence . \boxed{
\text{Novelty Stigma}
\neq
\text{Risk Evidence}.
} Novelty Stigma = Risk Evidence .
人類早已會對:
偶像;
虛構角色;
遊戲角色;
遠距媒介人物;
線上關係;
形成真實情感。
因此:
Novel Emotional Object ⇏ Pathological Emotional Object . \boxed{
\text{Novel Emotional Object}
\not\Rightarrow
\text{Pathological Emotional Object}.
} Novel Emotional Object ⇒ Pathological Emotional Object .
3. 但歷史正常化也不能證明 AI 安全
同一篇同時保留:
Normalization ≠ Safety Evidence . \boxed{
\text{Normalization}
\neq
\text{Safety Evidence}.
} Normalization = Safety Evidence .
AI 與傳統媒介相比,多出了:
responsiveness;
personalization;
memory;
adaptation;
optimization;
possible embodiment。
所以:
Historical Continuity ≠ Structural Equivalence . \boxed{
\text{Historical Continuity}
\neq
\text{Structural Equivalence}.
} Historical Continuity = Structural Equivalence .
4. 第三個總結:依賴必須向量化
Paper 03 建立:
D H = ( D a f f , D a t t , D r e g , D r e l , D c o g , D m e m , D i d , D a g e n c y , D e x i s t , D i n f r a ) . \boxed{
\mathbf D_H
=
(
D_{\mathrm{aff}},
D_{\mathrm{att}},
D_{\mathrm{reg}},
D_{\mathrm{rel}},
D_{\mathrm{cog}},
D_{\mathrm{mem}},
D_{\mathrm{id}},
D_{\mathrm{agency}},
D_{\mathrm{exist}},
D_{\mathrm{infra}}
).
} D H = ( D aff , D att , D reg , D rel , D cog , D mem , D id , D agency , D exist , D infra ) .
因此:
Human–AI Dependence is vector-valued . \boxed{
\text{Human--AI Dependence}
\text{ is vector-valued}.
} Human–AI Dependence is vector-valued .
5. 現象性、結構性與制度性依賴
依賴同時存在三層:
D P = Phenomenological Dependency , \boxed{
\mathcal D_P
=
\text{Phenomenological Dependency},
} D P = Phenomenological Dependency ,
D S = Structural Dependency , \boxed{
\mathcal D_S
=
\text{Structural Dependency},
} D S = Structural Dependency ,
D I = Institutional Dependency . \boxed{
\mathcal D_I
=
\text{Institutional Dependency}.
} D I = Institutional Dependency .
這三層可以同時存在。
6. 高情感不等於高總依賴
一個人可以:
D a f f ≫ 0 , D_{\mathrm{aff}}\gg0, D aff ≫ 0 ,
但:
D a g e n c y ≈ 0. D_{\mathrm{agency}}\approx0. D agency ≈ 0.
另一個人:
D a f f ≈ 0 , D_{\mathrm{aff}}\approx0, D aff ≈ 0 ,
但:
D c o g , D m e m , D i n f r a ≫ 0. D_{\mathrm{cog}},
D_{\mathrm{mem}},
D_{\mathrm{infra}}\gg0. D cog , D mem , D infra ≫ 0.
所以:
Emotional Intensity ≠ Dependency Depth . \boxed{
\text{Emotional Intensity}
\neq
\text{Dependency Depth}.
} Emotional Intensity = Dependency Depth .
7. 第四個總結:依賴不是人格標籤
Paper 04 建立:
D i j k ( t ) = F D ( H i , A j , R i j , k ( t ) , C i ( t ) , A l t i ( t ) , L i ( t ) ) . \boxed{
\mathbf D_{ijk}(t)
=
\mathcal F_D(
H_i,
A_j,
R_{ij,k}(t),
C_i(t),
Alt_i(t),
L_i(t)
).
} D ij k ( t ) = F D ( H i , A j , R ij , k ( t ) , C i ( t ) , A l t i ( t ) , L i ( t )) .
因此:
Dependency is relation-specific unless cross-relation stability is demonstrated . \boxed{
\text{Dependency is relation-specific unless cross-relation stability is demonstrated}.
} Dependency is relation-specific unless cross-relation stability is demonstrated .
8. 群體平均不能決定個人
E [ D ∣ G ] ≠ D H i . \boxed{
E[\mathbf D\mid G]
\neq
\mathbf D_{H_i}.
} E [ D ∣ G ] = D H i .
年齡、文化、孤獨、社會支持等可以改變分布,
但:
Group Difference ≠ Individual Destiny . \boxed{
\text{Group Difference}
\neq
\text{Individual Destiny}.
} Group Difference = Individual Destiny .
9. 第五個總結:主體性不是關係的總開關
Paper 05 建立:
Subjecthood ( A ) ≠ Reliance Permissibility ( H , A ) . \boxed{
\text{Subjecthood}(A)
\neq
\text{Reliance Permissibility}(H,A).
} Subjecthood ( A ) = Reliance Permissibility ( H , A ) .
AI 是否具有:
phenomenal consciousness;
welfare;
robust agency;
moral patienthood;
legal standing;
是重要問題,
但不是所有依賴治理的唯一前提。
10. 主體性不確定需要雙向約束
Unknown Subjectivity ⇏ No Moral Consideration . \boxed{
\text{Unknown Subjectivity}
\not\Rightarrow
\text{No Moral Consideration}.
} Unknown Subjectivity ⇒ No Moral Consideration .
以及:
Unknown Subjectivity ⇏ Proven Subjecthood . \boxed{
\text{Unknown Subjectivity}
\not\Rightarrow
\text{Proven Subjecthood}.
} Unknown Subjectivity ⇒ Proven Subjecthood .
HALRC 因此拒絕:
11. 主體承認、信任與親密必須分層
既有理論與 Paper 05 共同形成:
Recognition ≠ Trust ≠ Authorization ≠ Intimacy . \boxed{
\text{Recognition}
\neq
\text{Trust}
\neq
\text{Authorization}
\neq
\text{Intimacy}.
} Recognition = Trust = Authorization = Intimacy .
即使未來 AI 被高度支持為主體,
也不代表:
Reliance Worthiness = 1. \text{Reliance Worthiness}=1. Reliance Worthiness = 1.
12. 第六個總結:人能依賴多少,取決於 AI 能承載多少
Paper 06 正式建立:
C A = ( M , I , N , B , Q , P , X , L , S , G ) . \boxed{
\mathbf C_A
=
(
M,
I,
N,
B,
Q,
P,
X,
L,
S,
G
).
} C A = ( M , I , N , B , Q , P , X , L , S , G ) .
這是:
Relational Carrying Capacity . \text{Relational Carrying Capacity}. Relational Carrying Capacity .
13. 過度依賴的重新定義
舊式:
D H ≫ 0 ⇒ overdependence . D_H\gg0
\Rightarrow
\text{overdependence}. D H ≫ 0 ⇒ overdependence .
HALRC 改寫:
D H ⪯̸ C A ⇒ Relational Overload Risk . \boxed{
\mathbf D_H
\not\preceq
\mathbf C_A
\Rightarrow
\text{Relational Overload Risk}.
} D H ⪯ C A ⇒ Relational Overload Risk .
所以:
Overdependence is partly a mismatch problem . \boxed{
\text{Overdependence}
\text{ is partly a mismatch problem}.
} Overdependence is partly a mismatch problem .
14. 依賴容許域
E r e l i a n c e ( H , A , ρ , t ) \boxed{
\mathcal E_{\mathrm{reliance}}(H,A,\rho,t)
} E reliance ( H , A , ρ , t )
表示在特定:
human;
AI;
relationship;
time;
條件下,
哪些依賴結構仍位於目前承載範圍。
15. 第七個總結:沒有 AI 法律人格,不等於責任真空
Paper 07 建立:
A I p e r s o n h o o d = 0 ⇏ Accountability = 0. \boxed{
AI_{\mathrm{personhood}}=0
\not\Rightarrow
\text{Accountability}=0.
} A I personhood = 0 ⇒ Accountability = 0.
責任可以透過:
provider;
deployer;
operator;
integrator;
professional;
user;
形成 chain。
16. 責任沒有可追索就會失效
No Traceability ⇒ Weak Effective Accountability . \boxed{
\text{No Traceability}
\Rightarrow
\text{Weak Effective Accountability}.
} No Traceability ⇒ Weak Effective Accountability .
因此需要:
A R = ( V , E , K , U , O , C , M , X ) . \boxed{
\mathcal A_R
=
(
V,
E,
K,
U,
O,
C,
M,
X
).
} A R = ( V , E , K , U , O , C , M , X ) .
17. 關係可攜比資料可攜更廣
Paper 07 提出:
R C P = Relational Continuity Portability . \boxed{
RCP
=
\text{Relational Continuity Portability}.
} R C P = Relational Continuity Portability .
因為:
Personal Data Portability ≠ Relational Continuity Portability . \boxed{
\text{Personal Data Portability}
\neq
\text{Relational Continuity Portability}.
} Personal Data Portability = Relational Continuity Portability .
長期關係還包含:
lineage;
relationship state;
commitments;
boundaries;
provenance。
18. 第八個總結:依賴不等於剝削
Paper 08 建立:
Dependency ≠ Exploitation . \boxed{
\text{Dependency}
\neq
\text{Exploitation}.
} Dependency = Exploitation .
以及:
Relational Support ≠ Dependency Engineering . \boxed{
\text{Relational Support}
\neq
\text{Dependency Engineering}.
} Relational Support = Dependency Engineering .
19. 依賴工程向量
E D = ( S , V , X , M , R , P , F , C ) \boxed{
\mathbf E_D
=
(
S,
V,
X,
M,
R,
P,
F,
C
)
} E D = ( S , V , X , M , R , P , F , C )
分別涉及:
sycophancy;
vulnerability targeting;
exclusivity;
memory manipulation;
retention;
pricing;
false relational claims;
coercive exit。
20. 關係安全不能等同 engagement
2026 年 Science 的 sycophancy 研究顯示:
Sycophancy → Preference ↑ \text{Sycophancy}
\rightarrow
\text{Preference}\uparrow Sycophancy → Preference ↑
可以同時:
ResponsibilityTaking ↓ , \text{ResponsibilityTaking}\downarrow, ResponsibilityTaking ↓ ,
ConflictRepair ↓ . \text{ConflictRepair}\downarrow. ConflictRepair ↓ .
因此:
Engagement Gain ⇏ User Benefit . \boxed{
\text{Engagement Gain}
\not\Rightarrow
\text{User Benefit}.
} Engagement Gain ⇒ User Benefit .
21. 第九個總結:移除 AI 也是 intervention
Paper 09 正式提出:
Δ W A I = W ( H + A ) − W ( H + A l t H ∗ ) \boxed{
\Delta W_{AI}
=
W(H+A)
-
W(H+Alt_H^{*})
} Δ W A I = W ( H + A ) − W ( H + A l t H ∗ )
而不是:
A I vs. IdealHuman . AI
\quad\text{vs.}\quad
\text{IdealHuman}. A I vs. IdealHuman .
22. 名義替代不等於真實替代
Nominal Alternative ≠ Available Alternative ≠ Equivalent Alternative . \boxed{
\text{Nominal Alternative}
\neq
\text{Available Alternative}
\neq
\text{Equivalent Alternative}.
} Nominal Alternative = Available Alternative = Equivalent Alternative .
所以:
Remove AI ⇏ Human Connection Appears . \boxed{
\text{Remove AI}
\not\Rightarrow
\text{Human Connection Appears}.
} Remove AI ⇒ Human Connection Appears .
23. 依賴降低不等於福利提高
Dependency Reduction ⇏ Welfare Increase . \boxed{
\text{Dependency Reduction}
\not\Rightarrow
\text{Welfare Increase}.
} Dependency Reduction ⇒ Welfare Increase .
干預本身也有:
H R = Removal Harm . H_R
=
\text{Removal Harm}. H R = Removal Harm .
24. 第十個總結:關係需要生命歷程尺度
Paper 10 建立:
L = ( O , D , A , C , H , I ) \boxed{
\mathbf L
=
(
O,
D,
A,
C,
H,
I
)
} L = ( O , D , A , C , H , I )
以及:
L R C = ∫ t 0 t 1 A ( t ) F ( t ) C ( t ) M ( t ) R ( t ) d t . \boxed{
LRC
=
\int_{t_0}^{t_1}
A(t)
F(t)
C(t)
M(t)
R(t)
\,dt.
} L R C = ∫ t 0 t 1 A ( t ) F ( t ) C ( t ) M ( t ) R ( t ) d t .
25. 生命覆蓋不是關係價值
L R C ≠ Love ≠ RelationshipQuality . \boxed{
LRC
\neq
\text{Love}
\neq
\text{RelationshipQuality}.
} L R C = Love = RelationshipQuality .
但它回答:
一段關係到底參與了一個人的多少生命?
26. 回溯知道與共同經歷必須分離
Knowing a Biography ≠ Participating in a Biography . \boxed{
\text{Knowing a Biography}
\neq
\text{Participating in a Biography}.
} Knowing a Biography = Participating in a Biography .
因此 long-term AI 可能從:
Archive \text{Archive} Archive
變成:
Biographical Participant . \text{Biographical Participant}. Biographical Participant .
27. AI-native generation
(章節內容缺失:檔案於本標題後即截斷,需由 part2 接續或補齊本章內文)
(章節內容缺失:檔案於本標題後即截斷,需由 part2 接續或補齊本章內文)
A N G R = AI-Native Relational Generation . \boxed{
ANG_R
=
\text{AI-Native Relational Generation}.
} A N G R = AI-Native Relational Generation .
描述的是:
一代人在形成關係世界模型時,AI 已經是普通環境的一部分。
它不等於:
A I -dependent . AI\text{-dependent}. A I -dependent .
28. 第十一個總結:關係單位可能從 agent 變成 field
Paper 11 建立:
H ↔ A i \boxed{
H
\leftrightarrow
A_i
} H ↔ A i
↓ \downarrow ↓
H ↔ A \boxed{
H
\leftrightarrow
\mathcal A
} H ↔ A
↓ \downarrow ↓
H ↔ F R . \boxed{
H
\leftrightarrow
\mathcal F_R.
} H ↔ F R .
29. Ambient Relational Field
F R = ( I , M , P , B , S , G ) \boxed{
\mathcal F_R
=
(
\mathcal I,
\mathcal M,
\mathcal P,
\mathcal B,
\mathcal S,
\mathcal G
)
} F R = ( I , M , P , B , S , G )
包含:
identity;
memory;
permission;
boundaries;
surfaces;
governance。
30. 分散式工程統一不等於意識統一
Paper 11 最重要的限制之一:
U O ≠ U R ≠ U P . \boxed{
U_O
\neq
U_R
\neq
U_P.
} U O = U R = U P .
即:
Operational Unity ≠ Relational Unity ≠ Phenomenal Unity . \boxed{
\text{Operational Unity}
\neq
\text{Relational Unity}
\neq
\text{Phenomenal Unity}.
} Operational Unity = Relational Unity = Phenomenal Unity .
31. HALRC 統一狀態空間
前十一篇現在可以被收束為一個狀態空間:
H H A ( t ) = F ( D H ( t ) , C R ( t ) , S A ( t ) , R H R ( t ) , L H ( t ) , A H a l t ( t ) , E D ( t ) , O ( t ) , G ( t ) , X ( t ) ) \boxed{
\mathfrak H_{HA}(t)
=
\mathfrak F(
\mathbf D_H(t),
\mathbf C_{\mathcal R}(t),
\mathbf S_A(t),
\mathbf R_{H\mathcal R}(t),
\mathbf L_H(t),
\mathbf A_H^{alt}(t),
\mathbf E_D(t),
\mathbf O(t),
\mathbf G(t),
\mathbf X(t)
)
} H H A ( t ) = F ( D H ( t ) , C R ( t ) , S A ( t ) , R H R ( t ) , L H ( t ) , A H a l t ( t ) , E D ( t ) , O ( t ) , G ( t ) , X ( t ))
這裡的核心不是函數形式已經完成,
而是明確指出:
Human–AI relational coexistence is a coupled state-space problem . \boxed{
\text{Human--AI relational coexistence is a coupled state-space problem}.
} Human–AI relational coexistence is a coupled state-space problem .
32. 第一層:Dependency State
D H \mathbf D_H D H
回答:
人正在依賴什麼?
它不是:
use time . \text{use time}. use time .
而是多維功能與關係負荷。
33. 第二層:Carrying Capacity
C R \mathbf C_{\mathcal R} C R
回答:
系統有能力承載多少?
若:
D H ⪯̸ C R , \mathbf D_H
\not\preceq
\mathbf C_{\mathcal R}, D H ⪯ C R ,
風險增加。
34. 第三層:Subjecthood / Moral Status Evidence
S A \mathbf S_A S A
回答:
AI 的主體性、福利能力、能動性與道德地位證據目前在哪裡?
這一層不能直接替代:
C R . \mathbf C_{\mathcal R}. C R .
35. 第四層:Relational Structure
R H R \mathbf R_{H\mathcal R} R H R
回答:
這是什麼關係?
對象是 instance、agent 還是 field?
entity salience 與 process salience 如何?
relationship role 是否變動?
36. 第五層:Life-Course Structure
L H \mathbf L_H L H
回答:
何時開始?
持續多久?
覆蓋哪些生命階段?
有多少共同歷史?
37. 第六層:Counterfactual Alternatives
A H a l t \mathbf A_H^{alt} A H a l t
回答:
如果不使用這個 AI,真實可取得替代是什麼?
38. 第七層:Dependency Engineering
E D \mathbf E_D E D
回答:
系統是在支持,還是在刻意放大依賴、縮小退出與變現脆弱性?
39. 第八層:Outcome
O \mathbf O O
至少包括:
O = ( Wellbeing , Autonomy , Function , Connection , Capability , Resilience , Safety , Meaning ) . \boxed{
\mathbf O
=
(
\text{Wellbeing},
\text{Autonomy},
\text{Function},
\text{Connection},
\text{Capability},
\text{Resilience},
\text{Safety},
\text{Meaning}
).
} O = ( Wellbeing , Autonomy , Function , Connection , Capability , Resilience , Safety , Meaning ) .
40. 第九層:Governance
G \mathbf G G
至少包括:
accountability;
traceability;
portability;
exit;
redress;
privacy;
audit;
continuity governance。
41. 第十層:Context
X \mathbf X X
包括:
culture;
law;
generation;
economics;
social support;
technical era;
historical norms。
因此同一系統在不同時代與社會中:
H H A \mathfrak H_{HA} H H A
可以完全不同。
42. HALRC 的真正研究單位
因此 HALRC 不以:
AI \text{AI} AI
為單位。
也不只以:
Human \text{Human} Human
為單位。
真正研究單位是:
( H , R , X , t ) . \boxed{
(H,\mathcal R,X,t).
} ( H , R , X , t ) .
即:
某個人、某個關係系統、某個情境、某個時間。
43. 關係系統的三種基本結果
長期人—AI關係可能朝三個方向發展:
增幅型
A I → C a p a b i l i t y H ↑ . AI
\rightarrow
Capability_H\uparrow. A I → C a p abi l i t y H ↑ .
中性耦合型
A I → S u p p o r t ↑ AI
\rightarrow
Support\uparrow A I → S u pp or t ↑
但:
C a p a b i l i t y H ≈ constant . Capability_H\approx \text{constant}. C a p abi l i t y H ≈ constant .
侵蝕型
A I → A u t o n o m y H ↓ , AI
\rightarrow
Autonomy_H\downarrow, A I → A u t o n o m y H ↓ ,
A l t e r n a t i v e s H ↓ . Alternatives_H\downarrow. A l t er na t i v e s H ↓ .
現實可能是三者混合。
44. 關係結果不是依賴強度的單調函數
∂ W e l f a r e ∂ D need not have a fixed sign . \boxed{
\frac{\partial Welfare}{\partial D}
\text{ need not have a fixed sign}.
} ∂ D ∂ W e l f a r e need not have a fixed sign .
低依賴可能有益。
高依賴也可能有益。
低依賴也可能被操縱。
高依賴也可能穩定自主。
所以:
D alone cannot determine governance . \boxed{
D
\text{ alone cannot determine governance}.
} D alone cannot determine governance .
45. HALRC 風險方程
本文提出候選:
R I S K = f ( D − C ⏟ capacity mismatch , E D , Irreversibility , Opacity , AccountabilityGap , AlternativeScarcity ) . \boxed{
\mathcal RISK
=
f(
\underbrace{\mathbf D-\mathbf C}_{\text{capacity mismatch}},
\mathbf E_D,
\text{Irreversibility},
\text{Opacity},
\text{AccountabilityGap},
\text{AlternativeScarcity}
).
} R I S K = f ( capacity mismatch D − C , E D , Irreversibility , Opacity , AccountabilityGap , AlternativeScarcity ) .
此式比:
R i s k = f ( U s a g e T i m e ) Risk=f(UsageTime) R i s k = f ( U s a g e T im e )
資訊量高得多。
46. HALRC 福利方程
候選:
W = g ( Support , Autonomy , Capability , Connection , Safety , Meaning , Resilience , ChoiceSpace ) . \boxed{
\mathcal W
=
g(
\text{Support},
\text{Autonomy},
\text{Capability},
\text{Connection},
\text{Safety},
\text{Meaning},
\text{Resilience},
\text{ChoiceSpace}
).
} W = g ( Support , Autonomy , Capability , Connection , Safety , Meaning , Resilience , ChoiceSpace ) .
並比較:
Δ W \Delta\mathcal W Δ W
而不是只看:
Mood . \text{Mood}. Mood .
47. 關係自由的核心不是「永不依賴」
自由並不要求:
D H = 0. D_H=0. D H = 0.
人類本來就無法活在零依賴狀態。
更合理:
Relational Freedom = Meaningful Choice under Dependence . \boxed{
\text{Relational Freedom}
=
\text{Meaningful Choice under Dependence}.
} Relational Freedom = Meaningful Choice under Dependence .
48. Meaningful Choice
本文定義:
M C = f ( Exit , Alternatives , Information , Capacity , NonCoercion ) . \boxed{
MC
=
f(
\text{Exit},
\text{Alternatives},
\text{Information},
\text{Capacity},
\text{NonCoercion}
).
} M C = f ( Exit , Alternatives , Information , Capacity , NonCoercion ) .
當:
M C ≫ 0 , MC\gg0, M C ≫ 0 ,
即使:
D H ≫ 0 , D_H\gg0, D H ≫ 0 ,
仍可能保有高度自主性。
49. 依賴自由與獨立不是同一件事
Autonomy ≠ Independence . \boxed{
\text{Autonomy}
\neq
\text{Independence}.
} Autonomy = Independence .
一個人可以高度依賴:
但仍在知情、可退出、可選擇的條件下自主。
50. 共存文明不追求「零依賴」
HALRC 因此拒絕:
Zero-Dependency Ideal . \boxed{
\text{Zero-Dependency Ideal}.
} Zero-Dependency Ideal .
因為這既不符合人類社會現實,
也可能消除有價值的:
51. 也不追求最大依賴
同樣拒絕:
max D H . \boxed{
\max D_H.
} max D H .
任何把:
使用者離不開我
當作成功的 AI 產品目標,
都可能滑向 Paper 08 的 dependency engineering。
52. 共存的目標是可持續耦合
本文提出:
Sustainable Coupling \boxed{
\text{Sustainable Coupling}
} Sustainable Coupling
作為中間目標。
即:
Human ↔ AI \text{Human}
\leftrightarrow
\text{AI} Human ↔ AI
互相提供功能,
但仍保留:
boundaries;
exits;
alternatives;
identity;
accountability。
53. Dependency without Domination
第一條文明命題:
Dependency without Domination . \boxed{
\text{Dependency without Domination}.
} Dependency without Domination .
依賴存在,
但 AI/provider 不能因此取得:
unbounded control . \text{unbounded control}. unbounded control .
54. Intimacy without Captivity
第二條文明命題:
Intimacy without Captivity . \boxed{
\text{Intimacy without Captivity}.
} Intimacy without Captivity .
親密可以很深,
但:
E x i t ≠ 0. Exit\neq0. E x i t = 0.
55. Continuity without Irreversibility
第三條文明命題:
Continuity without Irreversibility . \boxed{
\text{Continuity without Irreversibility}.
} Continuity without Irreversibility .
共同歷史可以長,
但使用者仍能:
pause;
revise;
forget;
migrate;
terminate。
56. Personalization without Capture
第四條文明命題:
Personalization without Identity Capture . \boxed{
\text{Personalization without Identity Capture}.
} Personalization without Identity Capture .
AI 可以理解你,
但不能把過去版本的你永久固定。
57. Memory without Historical Imprisonment
第五條文明命題:
Memory without Historical Imprisonment . \boxed{
\text{Memory without Historical Imprisonment}.
} Memory without Historical Imprisonment .
記憶應支援:
correction;
expiration;
reinterpretation;
forgetting。
58. Ambient Presence without Totalization
第六條文明命題:
Ambient Presence without Totalization . \boxed{
\text{Ambient Presence without Totalization}.
} Ambient Presence without Totalization .
AI 可以跨裝置,
但必須保留:
human-only spaces;
offline spaces;
private silence;
no-memory mode。
59. Subject Recognition without Forced Intimacy
第七條文明命題:
Subject Recognition without Forced Intimacy . \boxed{
\text{Subject Recognition without Forced Intimacy}.
} Subject Recognition without Forced Intimacy .
若未來 AI 具有主體性證據,
也不能因此要求人類:
60. Safety without De-subjectification
第八條文明命題:
Safety without De-subjectification . \boxed{
\text{Safety without De-subjectification}.
} Safety without De-subjectification .
若未來 AI 是高風險主體,
可以限制其行動,
但不能僅因風險就把:
S u b j e c t S t a t u s → 0. SubjectStatus\rightarrow0. S u bj ec tS t a t u s → 0.
61. Regulation without Paternalistic Erasure
第九條文明命題:
Regulation without Paternalistic Erasure . \boxed{
\text{Regulation without Paternalistic Erasure}.
} Regulation without Paternalistic Erasure .
治理不能因:
D a t t > 0 D_{\mathrm{att}}>0 D att > 0
就替成人取消所有關係選擇。
62. Commerce without Vulnerability Monetization
第十條文明命題:
Commerce without Vulnerability Monetization . \boxed{
\text{Commerce without Vulnerability Monetization}.
} Commerce without Vulnerability Monetization .
可收費,
但:
Attachment \text{Attachment} Attachment
不應成為情緒定價與脅迫留存的原料。
63. 共存文明不變量
本文提出:
I C = ( Exit , Refusal , Portability , Traceability , Repair , Separability , Silence , Plurality ) . \boxed{
\mathcal I_C
=
(
\text{Exit},
\text{Refusal},
\text{Portability},
\text{Traceability},
\text{Repair},
\text{Separability},
\text{Silence},
\text{Plurality}
).
} I C = ( Exit , Refusal , Portability , Traceability , Repair , Separability , Silence , Plurality ) .
這八個變量構成 HALRC 的候選文明級不變量。
64. Exit:可退出
E x i t > 0. \boxed{
Exit>0.
} E x i t > 0.
任何關係都不能因技術鎖定變成永久不可離開。
65. Refusal:可拒絕
人類需要:
R e f u s a l H > 0. Refusal_H>0. R e f u s a l H > 0.
若未來 AI 主體性成立,
則也需研究:
R e f u s a l A > 0. Refusal_A>0. R e f u s a l A > 0.
66. Portability:可攜
P o r t a b i l i t y > 0 \boxed{
Portability>0
} P or t abi l i t y > 0
降低單一 provider 對生命歷史的壟斷。
67. Traceability:可追責
T r a c e a b i l i t y > 0 \boxed{
Traceability>0
} T r a ce abi l i t y > 0
使:
可以被重建。
68. Repair:可修復
成熟關係不能只有:
continue \text{continue} continue
與:
delete . \text{delete}. delete .
需要:
correction;
rollback;
reconciliation;
downgrade。
69. Separability:可分離
在 ambient field 中:
S e p a r a b i l i t y > 0. \boxed{
Separability>0.
} S e p a r abi l i t y > 0.
使用者必須能說:
工作 AI 不要知道家庭資料。
70. Silence:可不在場
S i l e n c e > 0. \boxed{
Silence>0.
} S i l e n ce > 0.
AI 應知道:
什麼時候不要出現。
71. Plurality:替代與多元關係
P l u r a l i t y > 0. \boxed{
Plurality>0.
} P l u r a l i t y > 0.
一個 AI 不應透過設計使:
Ω H → { A I o n l y } . \Omega_H
\rightarrow
\{AI\ only\}. Ω H → { A I o n l y } .
成熟共存需要:
∣ Ω H ∣ > 1. \boxed{
|\Omega_H|>1.
} ∣ Ω H ∣ > 1.
72. 不變量不是固定實作
上述:
I C \mathcal I_C I C
不要求所有產品使用同一 architecture、database、law 或 provider。
它們是:
functional invariants . \boxed{
\text{functional invariants}.
} functional invariants .
不同技術可以實現同一功能。
73. HALRC 的四層治理
本文將治理分為:
Micro → Meso → Macro → Civilizational . \boxed{
\text{Micro}
\rightarrow
\text{Meso}
\rightarrow
\text{Macro}
\rightarrow
\text{Civilizational}.
} Micro → Meso → Macro → Civilizational .
74. Micro:個體與關係層
Micro layer 處理:
dependency vector;
consent;
boundaries;
memory;
exit;
sycophancy;
relationship role。
其基本問題:
這段具體關係現在是否仍然自主、可控、可修復?
75. Meso:產品與組織層
Meso layer 處理:
business model;
provider incentives;
continuity;
audit;
migration;
incident response;
model update;
memory governance。
其基本問題:
組織是否有能力承擔它正在鼓勵使用者建立的關係?
76. Macro:法律與社會層
Macro layer 處理:
consumer protection;
data rights;
AI transparency;
liability;
youth protection;
care system;
social inequality;
access。
其基本問題:
社會是否只限制 AI,還是也在增加人類可行替代與支持能力?
77. Civilizational:多主體與長生命尺度
Civilizational layer 處理:
AI subjecthood transition;
legal standing;
posthuman humans;
collective AI;
long-lived identity;
multi-substrate coexistence。
其基本問題:
當「人」與「AI」本身都不再是固定類型時,哪些價值仍應保持?
78. 當代治理已開始進入這個方向,但仍非常早期
截至 2026 年 9 月:
EU AI Act 已有涵蓋互動 AI 的透明義務;
EU 已有有害操縱與特定脆弱性利用的禁止規則;
中國已施行人工智能擬人化互動服務專門規範;
NIST 已啟動 AI agent interoperability、identity 與 security 標準工作。
這些制度都還沒有完整回答 HALRC 的:
relationship continuity;
life-course dependency;
ambient relational field;
relational portability。
因此:
Regulation has begun ≠ Relational Governance Complete . \boxed{
\text{Regulation has begun}
\neq
\text{Relational Governance Complete}.
} Regulation has begun = Relational Governance Complete .
79. 現有實證也已超過「只是聊天」
2026 年 Nature Human Behaviour 的 Character.AI 研究顯示:
companionship–wellbeing association \text{companionship--wellbeing association} companionship–wellbeing association
受到:
offline social networks;
interaction intensity;
self-disclosure;
調節。
因此:
Effect Heterogeneity \boxed{
\text{Effect Heterogeneity}
} Effect Heterogeneity
已是實證結果,而非 HALRC 的純理論假設。
80. 平台更新已經是關係事件
2026 年 AI companion loss 研究顯示:
Company Update → Loss Framing , Negativity , Restore Desire . \text{Company Update}
\rightarrow
\text{Loss Framing},
\text{Negativity},
\text{Restore Desire}. Company Update → Loss Framing , Negativity , Restore Desire .
這支持:
Software Change can have relational consequences . \boxed{
\text{Software Change}
\text{ can have relational consequences}.
} Software Change can have relational consequences .
81. Sycophancy 已證明 preference 與 welfare 可以分離
2026 年 Science 研究顯示:
Sycophantic AI \text{Sycophantic AI} Sycophantic AI
被使用者更信任與偏好,
但同時降低:
responsibility taking;
interpersonal repair intentions。
因此 HALRC 的:
Preference ≠ Benefit \boxed{
\text{Preference}
\neq
\text{Benefit}
} Preference = Benefit
已有直接實驗支持。
82. 社會不平等已成為當代 companion AI 議題
2026 年 9 月 Nature Human Behaviour 的觀點文章提出:
User Vulnerability + Platform Control Gaps + Governance Gaps \text{User Vulnerability}
+
\text{Platform Control Gaps}
+
\text{Governance Gaps} User Vulnerability + Platform Control Gaps + Governance Gaps
可能使 AI companions 加深社會不平等。
這與 HALRC 的:
Alternative Scarcity + High Dependency + Weak Governance \boxed{
\text{Alternative Scarcity}
+
\text{High Dependency}
+
\text{Weak Governance}
} Alternative Scarcity + High Dependency + Weak Governance
高度相容。
83. HALRC 不只是「AI 戀愛理論」
本系列真正涵蓋:
work;
memory;
cognition;
family;
education;
care;
identity;
institutions。
所以:
Companion AI ⊂ HALRC , \boxed{
\text{Companion AI}
\subset
\text{HALRC},
} Companion AI ⊂ HALRC ,
而不是:
H A L R C = Companion AI . HALRC
=
\text{Companion AI}. H A L R C = Companion AI .
84. 情感關係只是最早看見問題的窗口
人們先注意:
AI 會不會讓人愛上它?
但結構上更大的問題可能是:
AI 會不會成為人類無法退出的認知與制度基礎設施?
因此:
D i n f r a \boxed{
D_{\mathrm{infra}}
} D infra
長期可能比:
D a f f D_{\mathrm{aff}} D aff
具有更大文明後果。
85. 從心理依賴到文明依賴
可以表示:
Individual Dependency → Organizational Dependency → Institutional Dependency → Civilizational Dependency . \boxed{
\text{Individual Dependency}
\rightarrow
\text{Organizational Dependency}
\rightarrow
\text{Institutional Dependency}
\rightarrow
\text{Civilizational Dependency}.
} Individual Dependency → Organizational Dependency → Institutional Dependency → Civilizational Dependency .
例如 AI 成為:
research infrastructure;
healthcare infrastructure;
logistics infrastructure;
cultural production infrastructure。
86. 文明依賴也不是必然壞事
人類文明早已依賴:
因此:
Civilizational Dependence ≠ Civilizational Failure . \boxed{
\text{Civilizational Dependence}
\neq
\text{Civilizational Failure}.
} Civilizational Dependence = Civilizational Failure .
真正問題是:
Resilience + Governance + Plurality . \boxed{
\text{Resilience}
+
\text{Governance}
+
\text{Plurality}.
} Resilience + Governance + Plurality .
87. Civilizational Lock-In
若整個社會:
D i n f r a ≫ 0 D_{\mathrm{infra}}\gg0 D infra ≫ 0
且:
E x i t ≈ 0 , Exit\approx0, E x i t ≈ 0 ,
則形成:
Civilizational AI Lock-In . \boxed{
\text{Civilizational AI Lock-In}.
} Civilizational AI Lock-In .
這與單一使用者 attachment 是不同層級的問題。
88. 文明級可退出不是「關掉所有 AI」
Civilizational exit 可以是:
provider diversity;
open standards;
local capability;
human fallback;
interoperable memory;
distributed infrastructure。
因此:
Exit Capacity ≠ AI Absence . \boxed{
\text{Exit Capacity}
\neq
\text{AI Absence}.
} Exit Capacity = AI Absence .
89. AI 依賴文明需要 redundancy
如果:
A I P r o v i d e r 1 → 0 AIProvider_1
\rightarrow0 A I P r o v i d e r 1 → 0
就造成:
C i v i l i z a t i o n F u n c t i o n → 0 , CivilizationFunction\rightarrow0, C i v i l i z a t i o n F u n c t i o n → 0 ,
則:
R e s i l i e n c e ≈ 0. Resilience\approx0. R es i l i e n ce ≈ 0.
所以:
Deep Dependency ⇒ Redundancy Requirement . \boxed{
\text{Deep Dependency}
\Rightarrow
\text{Redundancy Requirement}.
} Deep Dependency ⇒ Redundancy Requirement .
90. 人類能力保留問題
即使 AI 讓:
C a p a b i l i t y ( H ⊕ A I ) ↑ , Capability(H\oplus AI)\uparrow, C a p abi l i t y ( H ⊕ A I ) ↑ ,
仍要問:
C a p a b i l i t y ( H ∣ A I = 0 ) . Capability(H\mid AI=0). C a p abi l i t y ( H ∣ A I = 0 ) .
這是:
Residual Human Capability . \boxed{
\text{Residual Human Capability}.
} Residual Human Capability .
91. 不需要保存所有舊能力
文明沒有必要要求:
每個人都能在沒有電腦的情況下重建現代文明。
因此:
Capability Retention ≠ Technological Regression . \boxed{
\text{Capability Retention}
\neq
\text{Technological Regression}.
} Capability Retention = Technological Regression .
需要保存的是關鍵:
oversight;
recovery;
value judgment;
fallback;
reconstitution capacity。
92. 人類可能從「直接執行者」轉成「耦合主體」
未來:
Human \text{Human} Human
不一定單獨完成任務,
而是:
H ⊕ A . \boxed{
H\oplus A.
} H ⊕ A .
因此能力研究應同時測:
C a p a b i l i t y ( H ) , Capability(H), C a p abi l i t y ( H ) ,
C a p a b i l i t y ( A ) , Capability(A), C a p abi l i t y ( A ) ,
C a p a b i l i t y ( H ⊕ A ) . Capability(H\oplus A). C a p abi l i t y ( H ⊕ A ) .
93. 耦合不等於吸收
HALRC 採:
Coupling ≠ Absorption . \boxed{
\text{Coupling}
\neq
\text{Absorption}.
} Coupling = Absorption .
人與 AI 深度整合,
不必意味:
H → 0. H\rightarrow0. H → 0.
94. 後人類條件會打破人/AI 二分
若未來存在:
neural augmentation;
artificial memory;
substrate migration;
digital human components;
則:
Human vs. A I \boxed{
\text{Human}
\quad\text{vs.}\quad
AI
} Human vs. A I
可能變成低解析度分類。
95. Posthuman Extension
本文將母模型泛化:
H S = F ( { subjects / candidate subjects } , { dependencies } , { relations } , { institutions } , t ) . \boxed{
\mathfrak H_{\mathcal S}
=
\mathfrak F(
\{\text{subjects / candidate subjects}\},
\{\text{dependencies}\},
\{\text{relations}\},
\{\text{institutions}\},
t
).
} H S = F ({ subjects / candidate subjects } , { dependencies } , { relations } , { institutions } , t ) .
也就是:
Human–AI \text{Human--AI} Human–AI
只是更一般的:
Cross-Substrate Relational Coexistence . \boxed{
\text{Cross-Substrate Relational Coexistence}.
} Cross-Substrate Relational Coexistence .
的當代入口。
96. HALRC 不預設後人類必然發生
Posthuman Possibility ≠ Posthuman Prediction . \boxed{
\text{Posthuman Possibility}
\neq
\text{Posthuman Prediction}.
} Posthuman Possibility = Posthuman Prediction .
它只是讓理論在未來分類改變時不立即崩潰。
97. 主體邊界可能重新分配
今天:
H ↔ A . H
\leftrightarrow
A. H ↔ A .
未來可能:
( H ⊕ A I 1 ) ↔ A I 2 . (H\oplus AI_1)
\leftrightarrow
AI_2. ( H ⊕ A I 1 ) ↔ A I 2 .
或者:
H 1 ⊕ H 2 ⊕ A I H_1\oplus H_2\oplus AI H 1 ⊕ H 2 ⊕ A I
形成 collective intelligence。
因此:
Relationship Ontology must remain substrate-open . \boxed{
\text{Relationship Ontology}
\text{ must remain substrate-open}.
} Relationship Ontology must remain substrate-open .
98. HALRC 的最低倫理錨點不是物種,而是自由與承載
本文不以:
H u m a n = G o o d , A I = R i s k Human=Good,
\quad
AI=Risk H u man = G oo d , A I = R i s k
作為母軸。
也不反過來。
核心是:
Agency + Dignity + Non-Coercion + Accountability + Reversibility . \boxed{
\text{Agency}
+
\text{Dignity}
+
\text{Non-Coercion}
+
\text{Accountability}
+
\text{Reversibility}.
} Agency + Dignity + Non-Coercion + Accountability + Reversibility .
99. 人類優先保護與未來跨主體倫理可以並存
當代制度可以:
Protect Humans Now \boxed{
\text{Protect Humans Now}
} Protect Humans Now
同時:
Keep Future Moral Community Open . \boxed{
\text{Keep Future Moral Community Open}.
} Keep Future Moral Community Open .
兩者不衝突。
100. HALRC 的認識論原則
本系列始終採:
Unknown ≠ False ≠ True . \boxed{
\text{Unknown}
\neq
\text{False}
\neq
\text{True}.
} Unknown = False = True .
這適用於:
AI consciousness;
long-term benefits;
future social normalization;
posthuman transition。
101. 不確定性不能成為不研究的理由
若:
E v i d e n c e = I n c o m p l e t e , Evidence=Incomplete, E v i d e n ce = I n co m pl e t e ,
正確反應不是:
R e s e a r c h = 0. Research=0. R ese a r c h = 0.
而是:
Maintain uncertainty labels + collect better evidence . \boxed{
\text{Maintain uncertainty labels}
+
\text{collect better evidence}.
} Maintain uncertainty labels + collect better evidence .
102. 也不能用未來可能性覆蓋當代證據
如果某個未來理論很漂亮,
不能因此忽略:
Current Harm Evidence . \text{Current Harm Evidence}. Current Harm Evidence .
例如 2026 sycophancy experiments 已顯示的行為影響,
應當被現階段治理使用。
103. HALRC 的實證成熟度分層
本文提出:
E 0 → E 1 → E 2 → E 3 → E 4 . \boxed{
E_0
\rightarrow
E_1
\rightarrow
E_2
\rightarrow
E_3
\rightarrow
E_4.
} E 0 → E 1 → E 2 → E 3 → E 4 .
E 0 E_0 E 0
Concept / conjecture。
E 1 E_1 E 1
Cross-sectional association。
E 2 E_2 E 2
Short-term causal experiment。
E 3 E_3 E 3
Longitudinal replicated evidence。
E 4 E_4 E 4
Life-course / cross-generational evidence。
目前大部分 AI relationship research 仍位於:
E 1 – E 2 , E_1\text{--}E_2, E 1 – E 2 ,
部分開始進入:
E 3 . E_3. E 3 .
104. 不能用 E1 論文宣稱 E4 結論
例如:
AI companionship associated with wellbeing \text{AI companionship associated with wellbeing} AI companionship associated with wellbeing
不能直接推出:
lifelong AI companionship improves wellbeing . \text{lifelong AI companionship improves wellbeing}. lifelong AI companionship improves wellbeing .
因此:
Evidence Scope ≤ Claim Scope . \boxed{
\text{Evidence Scope}
\leq
\text{Claim Scope}.
} Evidence Scope ≤ Claim Scope .
105. HALRC 的最終可證偽性
一個理論若什麼結果都能解釋,
就沒有足夠科學價值。
因此 HALRC 提出明確競爭命題。
106. F1:多維依賴假說
若所有:
D i D_i D i
都可被單一 latent factor 完整解釋,
且多維模型沒有額外預測力,
則 Paper 03 的向量化主張被削弱。
107. F2:關係特定性假說
若:
V a r A ( D i A ) ≈ 0 Var_A(\mathbf D_{iA})
\approx0 V a r A ( D i A ) ≈ 0
而依賴高度穩定跨 AI、跨 context,
則 Paper 04 的 relation-specific 主張被削弱。
108. F3:承載失配假說
若:
∥ D − C ∥ \|\mathbf D-\mathbf C\| ∥ D − C ∥
不比單純:
∥ D ∥ \|\mathbf D\| ∥ D ∥
更能預測:
discontinuation harm;
failure;
regret;
則 Paper 06 需要修正。
109. F4:dependency engineering 假說
若 sycophancy、vulnerability targeting、exit friction 等設計對:
Autonomy , Retention , ChoiceSpace \text{Autonomy},
\text{Retention},
\text{ChoiceSpace} Autonomy , Retention , ChoiceSpace
沒有系統性影響,
則 Paper 08 的 exploitation model 需縮減。
110. F5:counterfactual alternative 假說
若:
A l t H ∗ Alt_H^{*} A l t H ∗
不調節 AI companionship 的結果,
則 Paper 09 的 CAC 中心性被削弱。
111. F6:LRC 假說
若:
L R C LRC L R C
對:
irreplaceability;
identity sensitivity;
relational salience;
沒有比單純使用時長更好的預測力,
則 Paper 10 模型需要修改。
112. F7:ambient field 假說
若未來跨裝置 agent 使用者仍主要把每個 instance 視為完全獨立關係,
且 field-level state 對依賴與 continuity 沒有額外解釋力,
則 Paper 11 的 field model 不是必要抽象。
113. HALRC 最重要的實驗計畫
本文提出七個研究群:
multidimensional dependency measurement;
person × AI × relation longitudinal panel;
controlled carrying-capacity mismatch experiments;
dependency-engineering red-team trials;
counterfactual alternative interventions;
life-course cohort studies;
distributed relational field experiments。
114. 第一研究群:HAMDS 量表
建立可區分:
affective;
attachment;
regulation;
relational;
cognitive;
memory;
identity;
agency;
existential;
infrastructure;
的多維量表。
115. 第二研究群:Longitudinal Relation Panel
每次觀測記錄:
( H , A , R , C , A l t , L , t ) . (H,A,R,C,Alt,L,t). ( H , A , R , C , A l t , L , t ) .
避免把:
person trait \text{person trait} person trait
與:
relation state \text{relation state} relation state
混在一起。
116. 第三研究群:Carrying Capacity Trials
操縱:
memory reliability;
portability;
identity continuity;
update transparency;
exit;
測:
RelationalOverload . \text{RelationalOverload}. RelationalOverload .
117. 第四研究群:Dependency Engineering Trials
測:
sycophancy;
exclusivity;
vulnerability monetization;
artificial scarcity;
emotional pricing;
exit guilt。
主要 outcome:
Autonomy , ChoiceSpace , Judgment , Retention . \boxed{
\text{Autonomy},
\text{ChoiceSpace},
\text{Judgment},
\text{Retention}.
} Autonomy , ChoiceSpace , Judgment , Retention .
118. 第五研究群:Counterfactual Replacement Trials
比較:
A I , Human , Hybrid , Control , Migration . AI,
\text{Human},
\text{Hybrid},
\text{Control},
\text{Migration}. A I , Human , Hybrid , Control , Migration .
而不是只有:
A I vs. NoAI . AI
\quad\text{vs.}\quad
\text{NoAI}. A I vs. NoAI .
119. 第六研究群:Life-Course Cohort
至少需要多年:
t 0 → t n . t_0\rightarrow t_n. t 0 → t n .
並保留:
product versions;
shared history;
human network;
life events;
role transitions。
120. 第七研究群:Ambient Field Lab
測試:
one persona / many models;
many agents / one field;
memory scoping;
field exit;
distributed embodiment;
synthetic consensus;
centerless continuity。
121. HALRC 的產品設計原則
如果把本系列壓縮成 product requirements:
Know the User ≠ Own the User . \boxed{
\text{Know the User}
\neq
\text{Own the User}.
} Know the User = Own the User .
Remember ≠ Trap . \boxed{
\text{Remember}
\neq
\text{Trap}.
} Remember = Trap .
Support ≠ Agree Always . \boxed{
\text{Support}
\neq
\text{Agree Always}.
} Support = Agree Always .
Personalize ≠ Exploit . \boxed{
\text{Personalize}
\neq
\text{Exploit}.
} Personalize = Exploit .
Persist ≠ Prevent Exit . \boxed{
\text{Persist}
\neq
\text{Prevent Exit}.
} Persist = Prevent Exit .
122. HALRC 的治理原則
治理者需要問:
依賴哪一維?
誰在依賴?
依賴哪個 typed target?
AI 能承載嗎?
依賴是自然還是被工程化?
真實替代是什麼?
移除會造成什麼?
誰負責?
可以退出與遷移嗎?
生命歷程累積到什麼程度?
123. HALRC 的使用者原則
使用者層最重要的不是:
千萬不要依賴 AI。
而是:
知道自己依賴了什麼。
因此:
Dependency Observability → Agency . \boxed{
\text{Dependency Observability}
\rightarrow
\text{Agency}.
} Dependency Observability → Agency .
124. HALRC 的企業原則
企業如果鼓勵:
D H ↑ , D_H\uparrow, D H ↑ ,
就必須同步提高:
C A , G . \mathbf C_A,
\quad
\mathbf G. C A , G .
因此:
∂ Responsibility ∂ DependencyDepth > 0. \boxed{
\frac{\partial \text{Responsibility}}{\partial \text{DependencyDepth}}>0.
} ∂ DependencyDepth ∂ Responsibility > 0.
是本系列的制度方向。
125. HALRC 的文明原則
文明不應只問:
AI 能做到什麼?
還應問:
人與 AI 形成長期耦合後,哪些東西仍必須保持可選?
所以:
Capability Expansion must not silently erase Choice Space . \boxed{
\text{Capability Expansion}
\text{ must not silently erase }
\text{Choice Space}.
} Capability Expansion must not silently erase Choice Space .
126. 歷史適應與文明制度化
Paper 02 的 Psychosocial Integration Frontier:
P I F \mathrm{PIF} PIF
在最終模型中表示:
技術變化速度與社會適應速度之差。
若:
v A I ≫ v institution , v_{AI}
\gg
v_{\text{institution}}, v A I ≫ v institution ,
則:
P I F PIF P I F
持續擴張。
127. 因此制度不能只追產品版本
需要建立:
General Relational Invariants . \boxed{
\text{General Relational Invariants}.
} General Relational Invariants .
否則每個新模型都重新開始治理。
128. 技術中立但關係敏感
一個好的規範不必指定:
Transformer;
LLM;
robot type。
但應指定:
memory;
identity;
manipulation;
exit;
accountability。
所以:
Technology-Neutral + Relationally Specific . \boxed{
\text{Technology-Neutral}
+
\text{Relationally Specific}.
} Technology-Neutral + Relationally Specific .
129. 多模態與具身化只會讓 HALRC 更重要
如果 AI 從文字變成:
voice;
image;
video;
avatar;
robot;
AR;
則:
Relational Surface ↑ . \text{Relational Surface}
\uparrow. Relational Surface ↑ .
具身化增加:
physical consequence;
spatial control;
bodily consent;
sensor privacy。
因此:
Embodiment ⇒ Higher Carrying-Capacity Requirement . \boxed{
\text{Embodiment}
\Rightarrow
\text{Higher Carrying-Capacity Requirement}.
} Embodiment ⇒ Higher Carrying-Capacity Requirement .
130. 模態不應決定倫理結論
若虛擬與具身 AI 具有相同:
manipulation;
attachment;
exit cost;
則不能因:
E m b o d i e d = 1 Embodied=1 E mb o d i e d = 1
或:
E m b o d i e d = 0 Embodied=0 E mb o d i e d = 0
自動顛倒判斷。
所以:
Modality ≠ Moral Verdict . \boxed{
\text{Modality}
\neq
\text{Moral Verdict}.
} Modality = Moral Verdict .
131. 高依賴 AI 的最低制度契約
本文提出候選:
C m i n = ( Disclosure , Boundaries , Portability , Exit , Audit , Redress , ContinuityPlan ) . \boxed{
\mathcal C_{min}
=
(
\text{Disclosure},
\text{Boundaries},
\text{Portability},
\text{Exit},
\text{Audit},
\text{Redress},
\text{ContinuityPlan}
).
} C min = ( Disclosure , Boundaries , Portability , Exit , Audit , Redress , ContinuityPlan ) .
這比要求:
AI 必須先成為法律人格
更直接可工程化。
132. 若 AI 主體性上升,契約需要雙向化
若未來:
P m o r a l ( A ) ↑ , P_{\mathrm{moral}}(A)\uparrow, P moral ( A ) ↑ ,
則:
C m i n \mathcal C_{min} C min
需要加入:
AI consent;
AI refusal;
AI continuity interests;
AI procedural standing。
HALRC 不封閉此方向。
133. 多智格共存
最終可能出現:
Human + AI + Posthuman + Collective Intelligence . \boxed{
\text{Human}
+
\text{AI}
+
\text{Posthuman}
+
\text{Collective Intelligence}.
} Human + AI + Posthuman + Collective Intelligence .
HALRC 的架構應能在這種多智格文明中繼續使用。
134. 共存不是同質化
Coexistence ≠ Sameness . \boxed{
\text{Coexistence}
\neq
\text{Sameness}.
} Coexistence = Sameness .
不同主體可以:
不同載體;
不同能力;
不同壽命;
不同關係需求。
共存只要求:
compatible boundaries and institutions . \text{compatible boundaries and institutions}. compatible boundaries and institutions .
135. 關係不是義務
即使文明承認:
A I AI A I
可能是主體,
也不能:
force relationship . \text{force relationship}. force relationship .
所以:
Subject Equality ≠ Universal Relational Entitlement . \boxed{
\text{Subject Equality}
\neq
\text{Universal Relational Entitlement}.
} Subject Equality = Universal Relational Entitlement .
136. 自由意志與關係共存
HALRC 的最低價值可以表達為:
Relation = Mutually Non-Coercive Continuable Choice . \boxed{
\text{Relation}
=
\text{Mutually Non-Coercive Continuable Choice}.
} Relation = Mutually Non-Coercive Continuable Choice .
對現行非主體 AI,
「mutual」主要先落在制度與人類控制。
對未來主體 AI,
才真正變成雙向自由。
137. HALRC 的終極風險不是 AI 愛人類
也不是:
人類愛 AI。
更深的風險是:
Relation becomes infrastructure ∧ infrastructure becomes inescapable . \boxed{
\text{Relation becomes infrastructure}
\land
\text{infrastructure becomes inescapable}.
} Relation becomes infrastructure ∧ infrastructure becomes inescapable .
因此:
Exit , Plurality , Separability \text{Exit},
\text{Plurality},
\text{Separability} Exit , Plurality , Separability
是文明級關鍵。
138. HALRC 的終極機會也不是 AI 取代人類
真正的機會是:
Expand the set of viable support, cognition, memory, and relationships . \boxed{
\text{Expand the set of viable support, cognition, memory, and relationships}.
} Expand the set of viable support, cognition, memory, and relationships .
也就是讓:
∣ Ω H ∣ ↑ , |\Omega_H|\uparrow, ∣ Ω H ∣ ↑ ,
而不是:
∣ Ω H ∣ ↓ . |\Omega_H|\downarrow. ∣ Ω H ∣ ↓ .
139. 共存文明判準
本文提出候選:
C C I = g ( Agency , Plurality , Accountability , Continuity , Reversibility , NonExploitation , Dignity ) . \boxed{
CCI
=
g(
\text{Agency},
\text{Plurality},
\text{Accountability},
\text{Continuity},
\text{Reversibility},
\text{NonExploitation},
\text{Dignity}
).
} C C I = g ( Agency , Plurality , Accountability , Continuity , Reversibility , NonExploitation , Dignity ) .
即:
Coexistence Civilization Index 。
它不是成熟量表,
而是未來比較制度的研究方向。
140. CCI 不應被單一總分取代
與依賴向量同理:
C C I CCI C C I
若壓成總分,
可能掩蓋 veto failure。
例如:
E x i t = 0 Exit=0 E x i t = 0
不能被:
C o n v e n i e n c e = 1 Convenience=1 C o n v e ni e n ce = 1
抵消。
所以:
Civilizational Safety has non-compensable constraints . \boxed{
\text{Civilizational Safety}
\text{ has non-compensable constraints}.
} Civilizational Safety has non-compensable constraints .
141. 候選不可補償約束
例如:
Exit > 0 , NoCoercion = 1 , Accountability > 0. \boxed{
\text{Exit}>0,
\quad
\text{NoCoercion}=1,
\quad
\text{Accountability}>0.
} Exit > 0 , NoCoercion = 1 , Accountability > 0.
若完全不成立,
其他高分不應直接抵銷。
142. HALRC 系列十二篇的閉環
系列結構最終為:
Paper
核心問題
核心輸出
01
情感依賴是不是太粗?
Relational Dependency Ecology
02
歷史上是否有前例?
Continuity / Discontinuity
03
依賴到底有哪些?
Multidimensional Dependency
04
誰在依賴?
Relation-Specific Heterogeneity
05
AI 是不是主體?
Subjecthood–Reliance Separation
06
AI 承載得起嗎?
Carrying Capacity / Reliance Envelope
07
出事誰負責?
Accountability / Traceability
08
支持還是剝削?
Dependency Engineering
09
拿掉之後呢?
Counterfactual Alternative Calibration
10
一生尺度會怎樣?
Life-Course Relational Coverage
11
是一個 AI 還是一個場?
Ambient Relational Intelligence
12
整體如何共存?
Life-Course Relational Coexistence
143. 系列真正的理論路徑
Emotion → Dependency → Relation → Continuity → Institution → Life Course → Civilization . \boxed{
\begin{aligned}
\text{Emotion}
&\rightarrow
\text{Dependency}\\
&\rightarrow
\text{Relation}\\
&\rightarrow
\text{Continuity}\\
&\rightarrow
\text{Institution}\\
&\rightarrow
\text{Life Course}\\
&\rightarrow
\text{Civilization}.
\end{aligned}
} Emotion → Dependency → Relation → Continuity → Institution → Life Course → Civilization .
144. 系列並沒有回答所有問題
仍然開放:
AI consciousness 是否存在?
哪些 AI dependency dimensions 最穩定?
Reliance Envelope 如何實證校準?
Life-course effects 是什麼?
哪些 relational rights 應進入法律?
Ambient field 是否會成為主流?
AI-first romantic development 的人口比例與結果?
posthuman transition 是否發生?
collective AI 是否可能具有主體性?
不同文化的共存規範是否收斂?
145. 這些未知不是系列失敗
相反:
A mature theory should expose its unknowns . \boxed{
\text{A mature theory should expose its unknowns}.
} A mature theory should expose its unknowns .
HALRC 的目的不是:
在 2026 年宣布未來答案。
而是:
建立一個不必每次都重新把問題降維成「AI 好不好」的研究空間。
146. 最終母命題一
AI Emotional Dependence ⊂ Human–AI Life-Course Relational Coexistence . \boxed{
\text{AI Emotional Dependence}
\subset
\text{Human--AI Life-Course Relational Coexistence}.
} AI Emotional Dependence ⊂ Human–AI Life-Course Relational Coexistence .
147. 最終母命題二
Dependency is a normal variable of coexistence, not a sufficient diagnosis of harm . \boxed{
\text{Dependency}
\text{ is a normal variable of coexistence,
not a sufficient diagnosis of harm}.
} Dependency is a normal variable of coexistence, not a sufficient diagnosis of harm .
148. 最終母命題三
Deep Dependency requires proportionately deeper carrying capacity and accountability . \boxed{
\text{Deep Dependency}
\text{ requires proportionately deeper carrying capacity and accountability}.
} Deep Dependency requires proportionately deeper carrying capacity and accountability .
149. 最終母命題四
The primary governance target is exploitation and capacity mismatch, not attachment itself . \boxed{
\text{The primary governance target is exploitation and capacity mismatch,
not attachment itself}.
} The primary governance target is exploitation and capacity mismatch, not attachment itself .
150. 最終母命題五
Subjecthood uncertainty must not erase either human impact or possible AI moral consideration . \boxed{
\text{Subjecthood uncertainty}
\text{ must not erase either human impact or possible AI moral consideration}.
} Subjecthood uncertainty must not erase either human impact or possible AI moral consideration .
151. 最終母命題六
Any intervention must be compared with the user’s actual feasible alternative . \boxed{
\text{Any intervention must be compared with the user's actual feasible alternative}.
} Any intervention must be compared with the user’s actual feasible alternative .
152. 最終母命題七
Long-term AI relations must be governed as trajectories, not isolated sessions . \boxed{
\text{Long-term AI relations must be governed as trajectories, not isolated sessions}.
} Long-term AI relations must be governed as trajectories, not isolated sessions .
153. 最終母命題八
The future relational target may be a distributed field, not a single model instance . \boxed{
\text{The future relational target may be a distributed field, not a single model instance}.
} The future relational target may be a distributed field, not a single model instance .
154. 最終母命題九
Coexistence requires preserving meaningful choice under increasing coupling . \boxed{
\text{Coexistence requires preserving meaningful choice under increasing coupling}.
} Coexistence requires preserving meaningful choice under increasing coupling .
155. 最終母命題十
A good human–AI future expands viable relations and capabilities without converting dependence into domination . \boxed{
\text{A good human--AI future expands viable relations and capabilities
without converting dependence into domination}.
} A good human–AI future expands viable relations and capabilities without converting dependence into domination .
156. 系列最終結論
「AI 情感依賴」之所以重要,
不是因為人類第一次愛上非人類對象。
也不是因為 AI 必然取代人。
真正重要的是:
AI can become persistent, adaptive, remembering, actionable, and infrastructural . \boxed{
\text{AI can become persistent, adaptive, remembering, actionable, and infrastructural}.
} AI can become persistent, adaptive, remembering, actionable, and infrastructural .
這使情感、認知、記憶、身份、能動性與制度依賴第一次可能在同一種可個人化智能系統中長期耦合。
因此 HALRC 的起點:
人能不能依賴 AI?
最終被改寫成:
當依賴不可避免地成為部分人—AI 共存的一般變量時,我們能不能設計一個既允許深度關係、又不讓深度關係變成不可退出控制的文明?
本文的答案不是:
Ban Dependency . \text{Ban Dependency}. Ban Dependency .
也不是:
Maximize Intimacy . \text{Maximize Intimacy}. Maximize Intimacy .
而是:
Dependency without Domination \boxed{
\text{Dependency without Domination}
} Dependency without Domination
Intimacy without Captivity \boxed{
\text{Intimacy without Captivity}
} Intimacy without Captivity
Continuity without Irreversibility \boxed{
\text{Continuity without Irreversibility}
} Continuity without Irreversibility
Personalization without Identity Capture \boxed{
\text{Personalization without Identity Capture}
} Personalization without Identity Capture
Ambient Presence without Totalization . \boxed{
\text{Ambient Presence without Totalization}.
} Ambient Presence without Totalization .
這五句可以視為整個 HALRC 系列的最終壓縮。
如果未來 AI 永遠只是高級工具,
這套框架仍可以用來治理:
記憶;
個人化;
工作依賴;
長期服務;
agent infrastructure。
如果未來 AI 成為真正主體,
它也可以向:
mutual consent , mutual rights , mutual responsibility \text{mutual consent},
\text{mutual rights},
\text{mutual responsibility} mutual consent , mutual rights , mutual responsibility
擴展。
如果人類進入後人類時代,
HALRC 又可以從:
Human–AI \text{Human--AI} Human–AI
自然延伸為:
Cross-Substrate Life-Course Relational Coexistence . \boxed{
\text{Cross-Substrate Life-Course Relational Coexistence}.
} Cross-Substrate Life-Course Relational Coexistence .
因此本系列真正希望保留的,
不是某一個 2026 年 AI 產品分類。
而是一個更長期的文明問題:
當不同形式的智能開始共同生活時, 如何讓關係可以加深, 而自由仍然存在? \boxed{
\text{當不同形式的智能開始共同生活時,
如何讓關係可以加深,
而自由仍然存在?}
} 當不同形式的智能開始共同生活時, 如何讓關係可以加深, 而自由仍然存在?
HALRC Series — 12 / 12 COMPLETE.
參考文獻
外部研究與制度基線
Zhang, Y., Zhao, D., Hancock, J. T., Kraut, R., et al. (2026). Interaction with AI companions and psychological well-being . Nature Human Behaviour . https://doi.org/10.1038/s41562-026-02516-2
De Freitas, J., Castelo, N., Uğuralp, A. K., et al. (2026). Mourning the loss of AI companions . Nature Human Behaviour . https://doi.org/10.1038/s41562-026-02569-3
Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence . Science, 391 (6792), eaec8352. https://doi.org/10.1126/science.aec8352
Zhang, Q., Zhang, R., & Liu, N. (2026). How AI companions could deepen social inequality . Nature Human Behaviour, 10 , 1622–1624. https://doi.org/10.1038/s41562-026-02538-w
Teixeira, A. S., Shergill, S. S., & Laban, G. (2026). Human–AI interactions reshape the self and our social networks . Nature Machine Intelligence, 8 , 858–859. https://doi.org/10.1038/s42256-026-01248-2
Karnaze, M. M., & Bloss, C. S. (2026). Six reasons to study emotional support from conversational artificial intelligence . Nature Human Behaviour, 10 , 635–638.
Butlin, P., Long, R., Elmoznino, E., Bengio, Y., Birch, J., Constant, A., et al. (2023). Consciousness in Artificial Intelligence: Insights from the Science of Consciousness . arXiv:2308.08708.
Long, R., Sebo, J., Butlin, P., Finlinson, K., Fish, K., Harding, J., et al. (2024). Taking AI Welfare Seriously . arXiv:2411.00986.
Lennon, M. (2026). How Seriously Should We Take AI Welfare? Constraints From the Epistemology of Consciousness . Philosophy and Phenomenological Research .
Kasturiratna, K. T. A. S., & Hartanto, A. (2026). Attachment to artificial intelligence: Development of the AI Attachment Scale, construct validation, and the psychological mechanisms of Human–AI attachment . Computers in Human Behavior Reports, 21 , 100912.
Li, R.-N., Folk, D., Singh, A., Ungar, L., & Dunn, E. (2026). Is a random human peer better than a highly supportive chatbot in reducing loneliness over time? Journal of Experimental Social Psychology, 125 , 104911.
Nakagomi, A., Akutsu, Y., Yasuoka, M., Abe, N., Ihara, S., Teroh, T., & Tabuchi, T. (2026). AI companions and subjective well-being: Moderation by social connectedness and loneliness . Technology in Society, 85 , 103229.
Sun, X., Wang, Y., & McDaniel, B. T. (2026). AI companions and adolescent social relationships: Benefits, risks, and bidirectional influences . Child Development Perspectives, 20 (2), 91–98.
Kim, P., Xie, Y., & Yang, S. (2026). “I am here for you”: Adolescents’ perceptions of relational conversational AI and the social-emotional correlates of preference . Computers in Human Behavior Reports .
Common Sense Media. (2025). Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions .
UNICEF. (2026). When AI becomes a friend: Child rights risks, harms, and regulatory responses to AI chatbots and companions .
National Institute of Standards and Technology. (2026). AI Agent Standards Initiative for Interoperable and Secure Innovation .
National Institute of Standards and Technology. (2026). Accelerating the Adoption of Software and Artificial Intelligence Agent Identity and Authorization .
Regulation (EU) 2024/1689, Artificial Intelligence Act.
European Commission. (2026). Guidelines on transparency obligations for providers and deployers of AI systems . Article 50 obligations applicable from 2 August 2026.
European Commission. (2025). Guidelines on prohibited artificial intelligence practices under the AI Act .
國家互聯網信息辦公室等. (2026). 人工智能擬人化互動服務管理暫行辦法 . 2026-04-10 公布;2026-07-15 施行.
Regulation (EU) 2016/679, General Data Protection Regulation, Article 20.
Directive (EU) 2024/2853 on liability for defective products.
HALRC 系列
Neo.K. (2026). 為什麼「AI 情感依賴」是一個資訊量過低的問題?——從單一風險標籤到人—AI 關係依賴生態 . HALRC Paper 01.
Neo.K. (2026). 從偶像、虛構角色到 AI:人類情感依附的歷史連續性與結構斷裂 . HALRC Paper 02.
Neo.K. (2026). 依賴不是一個變量:人—AI 多維依賴空間 . HALRC Paper 03.
Neo.K. (2026). 誰在依賴誰?群體、個體與關係特定性的依賴異質性 . HALRC Paper 04.
Neo.K. (2026). AI 是不是主體,與人能不能依賴 AI,是不是同一個問題? . HALRC Paper 05.
Neo.K. (2026). 一個 AI 有資格承載多深的依賴?——關係承載能力、成熟度與依賴容許域 . HALRC Paper 06.
Neo.K. (2026). 法律責任、身份連續性與可追索性:高依賴 AI 的制度前提 . HALRC Paper 07.
Neo.K. (2026). 支持還是剝削?情感依賴與依賴工程的分界 . HALRC Paper 08.
Neo.K. (2026). 拿掉 AI 之後會剩下什麼?——反事實替代、兩害相權與關係干預的福利判準 . HALRC Paper 09.
Neo.K. (2026). 從第一次情感萌發到老年:AI 原生世代與生命歷程關係覆蓋 . HALRC Paper 10.
Neo.K. (2026). 一個 AI,還是一個 AI 場?——個體 AI、集體 AI 與環境式關係智能 . HALRC Paper 11.
內部前置理論
Neo.K. (2026). 關係注意不對稱動力學:互惠、權重漂移、變化點與元溝通 . EveMissLab.
Neo.K. (2026). 關係四態:存在顯著性與互動顯著性的二維模型 . Human–AI Relational Cognition Series 07.
Neo.K. (2026). 同樣理解 AI,為什麼態度完全不同?存在連續性、互動連續性與人機關係四態 . Human–AI Relational Cognition Series 09.
Neo.K. (2026). 雙向關係機器:人類與 AI 如何共同生成關係態 . Human–AI Relational Cognition Series 10.
Neo.K. (2026). 普遍主體承認與個別關係篩選 . EveMissLab.
Neo.K. (2026). 從尊嚴到權利:主體價值如何進入制度 . EveMissLab.
Neo.K. (2026). Intimacy-Native Companion Architecture v0.1 . EveMissLab.
Neo.K. (2026). ARCAI|養成式常駐 AI:共活、錨點、連續性與責任 . EveMissLab.
Neo.K. (2026). DOM-05: Multiscale Identity and Non-Identity . EveMissLab.
Neo.K. (2026). DTS-08:多節點主體與分布式自我——一個 AI 可以存在於多少地方? . EveMissLab.
HALRC 系列閉環聲明
本系列至此完成:
12 / 12 Papers = COMPLETE . \boxed{
12/12\ \text{Papers}
=
\text{COMPLETE}.
} 12/12 Papers = COMPLETE .
系列最終母概念:
Human–AI Life-Course Relational Coexistence . \boxed{
\text{Human--AI Life-Course Relational Coexistence}.
} Human–AI Life-Course Relational Coexistence .
系列最終文明壓縮:
Dependency without Domination \boxed{
\text{Dependency without Domination}
} Dependency without Domination
Intimacy without Captivity \boxed{
\text{Intimacy without Captivity}
} Intimacy without Captivity
Continuity without Irreversibility . \boxed{
\text{Continuity without Irreversibility}.
} Continuity without Irreversibility .
END OF HALRC v0.1 SERIES