The Final Extent of Language

Thirty-five chapters have repeatedly enlarged the role of the communication system. The final question is whether those enlargements justify saying that AI replaces language, extends language, or moves language to a different layer of communication. The chapter answers only after reconstructing the argument — and revises the original thesis where the manuscript fails to support its strongest form.

Four transformations, recomposed

PART I — REPRESENTATION
source → semantic intermediate → many possible representations
Language stops being identical to one fixed rendering.

PART II — ATTENTION
information world → selection, resolution, preservation, timing
Communication stops requiring the receiver to consume the whole stream.

PART III — PERSONAL MEDIATION
policy, evidence, mode, boundaries, visible model
→ recipient-conditioned interface
The same information need not take the same form for everyone.

PART IV — PROXY COMMUNICATION
prediction, delegation, speech authority, private discovery,
proxy ecosystems → communication partly between representatives
before reaching humans.

Each stage answers a different question; composition is the argument. The future stack:

HUMAN PURPOSE / SOURCE STATE → authorised semantics
→ mediation + selection → recipient / task / policy
→ representation compiler → language / diagram / table /
   interaction / notification / proxy speech / action
→ RECIPIENT

and increasingly, principal → proxy → private discovery/negotiation → counterparty proxy → recipient policy → human-facing representation → recipient. The literature’s starting premise is real, not speculative: Hancock, Naaman and Levy (JCMC 2020, 25:89–100, DOI 10.1093/jcmc/zmz022, verified) defined AI-mediated communication as agents operating on communicators’ behalf by modifying, augmenting, or generating messages — agency inside the channel, exactly this book’s Chapter-1 thesis. A 2026 scoping review of generative-AI-facilitated social interaction (user-supplied specifics: 85 sources; social outcomes underexplored relative to usability — existence confirmed, numbers fenced) supports the book’s restraint more than its reach. Neither establishes this book’s architecture. Both establish its premise.

The revised thesis: language moves up the stack

Not: AI replaces language. The manuscript supports something precise:

AI can move language from being the primary carrier of communication toward being one compiled representation of a deeper mediated communicative process. Language does not disappear. It moves up the stack.

Human symbolic systems remain; representation becomes variable; communication becomes computational (selecting, transforming, contextualising, timing between source and recipient); the receiver joins compilation (purpose, knowledge, preferences, mode, policy); communication becomes agentic under explicit authority. The compiled output is prose when prose serves — table, diagram, progressive view, warning, silence, action, private match, proxy negotiation otherwise. And the limit travels with the thesis: the inequalities below are the book’s mathematical backbone, encoding its deepest pattern, now strengthened by the hostile reviews into a broader synthesis — the architecture depends on preserving boundaries between representations, constructs, authorities, and actors, and most serious failures arise when a successful operation in one domain is treated as permission or evidence in another (decodable assumed retrievable; fast assumed learned; preferred assumed beneficial; observed assumed authorised; authorised action assumed authorised data use; private computation assumed legitimate purpose).

REPRESENTATION / COGNITION
source ≠ semantic record · record ≠ complete meaning
similarity ≠ relationship · decodability ≠ retrieval utility
effort reduction ≠ cognitive benefit
semantic preservation ≠ learning preservation

ATTENTION / PERSONALISATION
importance ≠ relevance ≠ novelty ≠ urgency
preference ≠ performance · behaviour ≠ consent
temporary state ≠ durable policy · profile ≠ policy

AUTHORITY / PRIVACY
access ≠ permission to use
action authority ≠ data-use authority ≠ speech authority
prediction ≠ authority · confidence ≠ consent
revocation ≠ reversal of prior flow
match ≠ reveal · introduction ≠ future reuse authority
one person's consent ≠ another's authorisation
cryptography ≠ purpose legitimacy

SYSTEMS / SOCIETY
emergence ≠ legitimacy · local correctness ≠ global safety
representation ≠ intention ≠ permission to communicate

Beside them, the thirteen earned laws: every transformation measures what survived; reduced representations omit without mutating; contestability never requires continuous contestation; evaluators are instruments requiring validation; inference proposes, authorised policy disposes; confidence never becomes consent; temporary never silently persists; delegation narrows; least privilege over composed workflows; correctness never cures unauthorised action; cryptographic privacy never legitimates unauthorised purpose; one person’s consent never erases another’s privacy or agency; local correctness never implies global safety.

Evidence states: programme, not victory

Four states, stated plainly because zero book experiments have run: established background (generative/adaptive/long-media/retrieval/delegation/privacy precedents; AI-MC already shifting authorship assumptions); supported mechanism precedents (progressive summarisation, conditional adaptation, editable preferences, bounded replicas, negotiation, PSI/MPC, scoped permissions); book hypotheses awaiting experiment (browser, Radar, policy runtime, gated resolution, Personal AI composition, SpeechGrant/DiscoveryGrant systems, proxy ecosystem) — with the explicit rule that architectural repair never promotes anything upward: DataUseAuthority, safe-default personalisation, and operation-separated delegation remain proposed mechanisms, not established solutions; horizon (widespread proxy societies, rich private semantic discovery, cross-domain replicas, BCI intent interfaces, prose-as-optional-view systems). The manuscript constructed an experimental programme — separable mechanisms, falsifiable claims, narrowing-capable designs — and its honesty instrument is the falsification list: dynamic representation harming understanding despite gates; task-conditioning adding nothing over fixed views; policies adding correction burden; systematic unnoticed omissions; prohibitive delegation overhead; incomprehensible proxy provenance; privacy destroying discovery utility; uncontrollable population externalities; language remaining overwhelmingly preferable. Any of these narrows or breaks the thesis, and the conclusion welcomes them — architecture manifestoes fear falsification; scientific books invite it.

EXP-36 is therefore the book-level integration protocol, not a 37th mechanism: capability levels 0–7 (fixed source through population ecosystem), each inheriting all obligations below, promotion requiring evidence, failure forcing the thesis backward — no higher extent claimed until its additional capability survives its additional failure modes — with proof obligations accumulating per level (L1 preservation evidence; L2 evaluator validity + purpose-bounded sufficiency; L3 preference/performance separation + safe default; L4 burden + operation-level data authority; L5 action/data-use/revocation enforcement; L6 predicate legitimacy + relational privacy; L7 externality/counterparty/governance evidence). And its complement: a capability gained by collapsing a boundary does not count as progress. The BCI edge gets one paragraph: constrained inner-speech decoding is real (verified Ch 29), LLM+BCI integration counts eleven heterogeneous studies per a 2026 systematic review (user-supplied, fenced) — an active frontier reinforcing representation ≠ intention ≠ permission, then left behind.

BEFORE:  mind → language → channel → language → mind
AFTER:   person → source / purpose / authority
         → AI-mediated system (select · transform · act)
         → representation → person / proxy

The final extent of language is not its disappearance. It is the point at which language ceases to be the whole communication system.

References

• Hancock, J.T. et al. (2020). JCMC 25:89–100. DOI 10.1093/jcmc/zmz022. Verified: AI-MC as modify/augment/generate on behalf of communicators. Used: premise reality.
• 2026 scoping review (S245195882600374X): user-supplied specifics (85 sources; outcomes underexplored); existence confirmed, numbers fenced. Used: restraint support.
• 2026 LLM+BCI systematic review (11 studies, heterogeneous): user-supplied, fenced. Used: frontier-activeness only.
• Ch 01–35 evidence bases: recomposed, not re-argued.

Proposed experiment EXP-36: book-level integration protocol

Status: PROPOSED. Levels 0–7 with inherited obligations, evidence-gated promotion, backward-forcing failures, and the nine falsification conditions as acceptance criteria for the central thesis.