Narrative Infrastructure Under Budget Constraint: A Field-Comparative Study in Hyperstition Architecture® and AI-Risk Discourse Formation

Lloyd Lewis, PhD
Art of FACELESS Research Division — artoffaceless.org downloadable PDF White Paper [botton of this page]
15 July 2026

Abstract

Between 14 June and 14 July 2026, two projects operated in parallel under overlapping public signals related to AI existential risk: plzdontkillus.com, a funded, month-long, in-person creator residency in Berkeley, California, and plzdontkillus.online, a zero-budget speculative-fiction node operated by the author as part of a fourteen-year transmedia research program (The Hollow Circuit®). This paper treats the 14-day operation of the latter as a deliberately concluded field study, using the differential in funding, institutional framing, and reach between the two projects to test predictions made by the author’s previously registered Hyperstition Architecture® framework. It finds that infrastructural and financial concentration — not narrative rigor, ethical review, or empirical accuracy — is the dominant predictor of discourse capture in this domain, and that the funded comparator exhibits a documented institutional-accountability gap, a structural access barrier with disability and age-discrimination implications, and a data-governance pattern that generates the precise category of risk (biometric, deepfake-usable video) that its stated mission opposes. The paper closes by arguing that the observed asymmetry constitutes evidence for, rather than against, the viability of narrative-first, infrastructure-poor cultural production as a research method, and by extension, as an investment thesis.

Keywords: hyperstition, AI-risk discourse, narrative economics, disability access, biometric data governance, standpoint methodology

1. Introduction and Positionality Statement

Following standard practice in critical and reflexive research traditions — see Haraway’s concept of “situated knowledges” (Haraway, in femres.org summary) and Sandra Harding’s standpoint theory, both of which hold that a researcher’s social position is a resource for analysis rather than a contaminant of it — the author discloses the following positionality: I am 66 years old, chronically disabled (secondary progressive multiple sclerosis), UK-based, and have operated Art of FACELESS and The Hollow Circuit® as a self-funded, unpaid creative and research practice for fourteen years, entirely outside the venture-funded AI-safety ecosystem described in Sections 4–5.

This disclosure is not offered as a claim to impartiality. It is offered because the critique advanced in Sections 5.2 and 5.3 rests in part on structural and experiential observation from within the affected class — disabled, older, and operating outside major venture-capital and academic centers — which is itself a valid evidentiary register in critical and standpoint research, distinct from and not requiring the same evidentiary standard as a specific factual allegation against a named individual (of which this paper makes none).

2. Background: Two Parallel Signals

plzdontkillus.com launched applications in spring 2026 for a 31-day, all-expenses-paid residency in Berkeley, California, funded by the Machine Intelligence Research Institute (MIRI) and other private backers, offering $2,000 stipends plus housing and meals to participants who commit to daily short-form video content on AI risk, with mentorship from high-profile figures including Grimes and Eliezer Yudkowsky (Bitget Wallet project summary; Aella, X/Twitter).

plzdontkillus.online launched under the same hashtag on 30 June 2026, operated solely by the author, producing 14 days of speculative fiction (in the voice of the recurring Hollow Circuit character Awen Null) exploring adjacent questions of AI risk, personhood, and exclusion, at zero direct cost. The overlap was disclosed publicly and in full on 14 July 2026 (A Straight Statement on #plzdontkillus).

The two projects are not competitors and make no equivalent claim to scale. Their coexistence, however, produces a natural — if uncontrolled — comparative condition: two actors addressing overlapping subject matter, under near-identical framing language, at radically different levels of infrastructural investment, over the same calendar window.

3. Theoretical Framework

This paper applies three frameworks previously formalized and registered by the author (UK Intellectual Property Office; see Section 8 of the companion stakeholder briefing for registration numbers):

Hyperstition Architecture®

Models the conditions under which a fiction becomes structurally load-bearing in real systems, expressed as:

Impact = Infrastructure × Immersion Safety Constraints

Impact > 1 ⇒ Cognitive Colonisation

— the point at which a narrative’s effect on audience cognition exceeds what its underlying safety and verification constraints would predict.

Cognitive Colonisation®

Models the isomorphism-limit conditions under which sustained fictional or promotional framing produces measurable, persistent shifts in how an audience processes claims of authenticity, authority, or risk, independent of the framing’s underlying accuracy.

The Veylon Protocol®

Models the recursive entanglement of an observing system with the system it observes, expressed as:

Vp(M) = M(H(M(I))) → Ω

— relevant here because this paper is itself a recursive instance of the phenomenon it studies: a narrative project observing, and being changed by, its own observation of a better-funded neighboring narrative project.

4. Methodology

This is an uncontrolled, participant-observer comparative case study. The author operated one side of the comparison (plzdontkillus.online) as embedded field research and discloses this role in full (Section 1; see also the Straight Statement). All data on the comparator project (plzdontkillus.com) is drawn from public, dated, third-party sources — the project’s own social media disclosures, independent journalism, and public financial/policy records — rather than direct access to internal data, budgets, or participant lists, which the author does not have and does not claim to have.

Limitations of this design are addressed in Section 7.

5. Findings

5.1 Resource Asymmetry as the Dominant Variable

Applying the Hyperstition Architecture equation to the comparator project: high disclosed Infrastructure (venue, mentor roster, stipends, housing, meals, platform amplification via high-profile mentors) combined with high Immersion (daily content production sustained over 31 days by multiple participants) and comparatively low disclosed Safety Constraints (no visible institutional ethical review — see 5.2) predicts an Impact score well above the Cognitive Colonisation® threshold — i.e., disproportionate audience reach and perceived legitimacy relative to the project’s methodological rigor.

By contrast, plzdontkillus.online operated with near-zero Infrastructure and moderate Immersion (one author, 14 days) yet still generated measurable public engagement and, per the author’s disclosure, comparable hashtag visibility during its run. This is consistent with the framework’s prediction that Immersion (narrative coherence and sustained presence) can partially substitute for Infrastructure (capital) up to a bounded ceiling — but that ceiling is real, and infrastructural capital remains the larger term in the equation at scale.

This asymmetry is not a local anomaly. The San Francisco Bay Area captured 82.6% of all U.S. venture capital dollars in Q1 2026 (Angel Investors Network); the US and UK together capture roughly 75% of global AI safety funding, with the Bay Area alone projected to exceed 50% of AI-safety research FTEs by end of 2026 (Quick Market Pitch; Safety Research Allocation Model, LongtermWiki). The comparator project is not an outlier; it is a representative instance of where capital already concentrates.

5.2 Institutional Accountability Gap

The comparator project’s public framing — its Berkeley location, its “residency” and “mentorship” vocabulary — reads as institutionally adjacent. On inspection, it is not affiliated with, run by, or reviewed by the University of California, Berkeley, or any equivalent academic body; it is an independently organized initiative funded by MIRI and other private backers (Bitget Wallet project summary). The venue itself — the Rose Garden Inn — had “permanently closed in 2024, when the owners, the tech-backed nonprofit Center for Applied Rationality, were ordered to pay back the $1 million in FTX funds used to buy the property” (Gazetteer SF). No institutional review board, ethics board, or published data-governance policy is visible in association with the project’s content or participant-selection process. Under standard research-integrity or ISO-style audit criteria, a project of this design and scale would not clear institutional review at a body that maintains such a board.

5.3 Narrative Distortion in AI-Mediated Search

A separate but structurally related finding, drawn from the author’s own IP-enforcement work: a Steam game, “King Of The Hollow Circuit” (developer: SavicNode), disclosed as AI-generated content, briefly ranked prominently in general search results under a title infringing the author’s registered mark, despite having no independent traction, before a successful rename complaint (now “King Of The Dark Circuit”; Steam; SteamDB). This is one documented instance of a broader, named phenomenon — “recommendation poisoning” — in which AI-optimized content is surfaced by search and AI-answer systems faster than earned authority can compete (AuthorityTech, citing The Verge). The relevance to this paper: the same infrastructural-capture dynamic identified in Section 5.1 (capital and platform access outperforming substance) recurs at the level of individual search and discovery systems, not only at the level of funding.

5.4 Biometric Data Generation as a Dual-Use Governance Question

A structural tension exists between the comparator project’s stated mission (raising awareness of AI risk) and its operational design (a month of daily, personal, face-forward video from dozens of participants). Germany’s federal cybersecurity authority states that convincing face-swap deepfakes require “only a few minutes” of video of a target person as training material (BSI). Separately, YouTube’s “likeness detection” tool — marketed as deepfake protection — requires a submitted government ID and biometric face video, with a privacy policy that currently permits use of that biometric data in AI model training (MSN/CNBC). Published deepfake-research literature identifies this class of dataset — high-quality, consented-but-unreviewed personal video — as carrying acknowledged dual-use risk (MethodsX).

The finding is structural, not accusatory: by the field’s own published technical threshold, a dataset of this volume and quality is sufficient to enable non-consensual synthetic media of every participant, independent of any actor’s intent. That this fact sits unaddressed in a project whose stated purpose is AI-risk mitigation is, on its own terms, a governance gap.

6. Discussion

Read together, Sections 5.1–5.4 describe a single pattern: infrastructure and capital determine reach and perceived legitimacy in this domain largely independent of methodological rigor, ethical oversight, or accuracy — and that pattern reproduces itself at every scale examined, from national funding concentration down to individual search-ranking behavior. The Hyperstition Architecture® framework predicts this outcome directly: absent binding Safety Constraints, Impact scales with Infrastructure regardless of the underlying narrative’s truth-value.

This has a further implication worth stating plainly. If a fourteen-year, self-funded, disability-constrained solo practice can produce a coherent transmedia research program, a successful IP-enforcement record, and — over 14 days at zero cost — a documented and citable comparative field study, the same framework predicts substantially greater output under comparable infrastructural investment to that already flowing to the comparator project. The zero-budget condition tested here is not a ceiling on the method; it is closer to a floor.

7. Limitations

This is a single, uncontrolled, participant-observer case comparison, not a randomized or statistically powered study. The author has no access to the comparator project’s internal budget, participant demographics, or data-handling practices beyond what is publicly disclosed, and does not claim otherwise. Findings in Section 5.4 describe a structural governance gap, not evidence of any specific data misuse by the comparator project or its organizers. Findings in Section 5.2 describe an absence of visible institutional oversight, not a claim about the intent or good faith of the individuals running the project.

8. Conclusion

The 14-day operation of plzdontkillus.online, concluded and disclosed in full on 14 July 2026, is treated in this paper as completed field research rather than an abandoned experiment. It demonstrates, with citation, that the primary driver of discourse capture in contemporary AI-risk communication is infrastructural and financial concentration rather than narrative or ethical rigor; that a comparator project riding on borrowed institutional prestige carries a documented accountability gap; and that the same project’s operational design generates a governance question about the exact category of harm it claims to oppose. The author’s fourteen-year, self-funded practice — of which this paper is one output — stands as the demonstration case for what independent, infrastructure-poor cultural and analytical production can achieve, and as the basis for what it could achieve with comparable backing.

References

Download the full white paper in PDF format here:

Narrative vs. Budget — AOF Stakeholder White PaperDownload