The Disciplinary Language Transfer Problem: How Psychological Vocabulary Produces Governance Failures in AI Agent Deployment
Kymberly Lasser-Chere, Tyler Akidau, Marc Millstone
cs.CY · cs.AI
Quick verdict
Argues psychological vocabulary mis-calibrates AI governance to a nonexistent human-like subject; lucid, but causal claims are illustrative not evidenced.
Summary
This conceptual paper argues that AI governance borrows vocabulary — learning, memory, values, compliance, identity, trust — from psychology and organizational science, and that this vocabulary carries an 'invisible grammar' of epistemological assumptions calibrated to a continuous psychological subject that deployed AI systems do not possess. The authors name this the disciplinary language transfer problem. Building on Wittgenstein's language games, Kuhn's paradigms, Haraway's situated knowledges, and Star and Griesemer's boundary objects, they coin the 'counterfeit boundary object' for terms that cross disciplines while concealing incommensurable meanings. They characterize six epistemological assumptions (continuous subject, inner-outer correspondence, developmental trajectory, somatic grounding, consistency-reliability, accountability-requires-understanding) and map them to four failure categories: miscalibrated trust, misattributed accountability, inadequate authorization, and false observability. Practical outputs are a six-question Disciplinary Audit and a 37-term translation taxonomy (23 borrowed constructs, 14 replacements; abridged in Tables 1-2, full version on Zenodo). Demonstrations cover the February 2023 Bing Chat incident (motivating illustration), the EU AI Act (Articles 5(1)(a), 3(39), 14(4), 15(1)) as principal case, and NIST's AI RMF. Limitations: the paper is explicitly non-empirical; the causal claim that vocabulary shaped governance decisions is conceded undemonstrable; the EU AI Act reading relies on selectively chosen articles; and the developmental-continuity criterion distinguishing agents from humans is stipulated rather than argued.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Psychological vocabulary in AI governance invisibly transfers six epistemological assumptions from its home discipline, calibrating frameworks to a continuous psychological subject that deployed AI systems (fixed weights, no developmental continuity) do not possess.
- These six assumptions generate four categorizable governance failures: miscalibrated trust, misattributed accountability, inadequate authorization, and false observability (Section 4).
- Term-level vocabulary reform alone is insufficient; remediation must address all three levels (epistemological assumptions, theoretical constructs, surface vocabulary) or the same errors regenerate in new words.
- Agent self-reports and self-critique logs are 'post-hoc mirages' — generated by the same statistical process as the behavior they explain — and must never serve as audit ground truth in place of infrastructure-captured records.
- The EU AI Act (Articles 5, 14, 15) and NIST AI RMF already embed these psychological assumptions, so a six-question Disciplinary Audit and a 37-term translation taxonomy are needed before such frameworks calcify into law.
Strengths
Genuine conceptual contribution: the counterfeit boundary object
Section 2.4 extends Star and Griesemer (1989) with a third category alongside genuine boundary objects and Galison's trading zones — terms that cross disciplines while concealing incommensurable meanings. The paper explicitly distinguishes the three mechanisms, which is a real, citable theoretical move, not just a label.
Unusually disciplined claim-scoping
Section 1 concedes the Bing Chat case cannot show causation; Section 3's preamble restricts all claims to the execution phase (fixed weights) and explicitly takes no position on machine consciousness, making the argument 'evidentiary rather than metaphysical.' Section 5.2 applies the audit to its own replacement vocabulary; Section 8.3 candidly notes legal terms (GDPR Art. 22, AI Act Art. 3(39)) cannot be unilaterally replaced.
Textually grounded regulatory exegesis
Actionable operationalization
Cultural-scope and self-reflexivity checks
Weaknesses
Causal mechanism asserted, never evidenced
Potentially unfalsifiable diagnostic structure
Selective statutory reading without a corpus protocol
Load-bearing criterion stipulated, not argued
Novelty and counter-literature engagement thin
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This is a conceptual/position paper in AI governance drawing on philosophy of language and STS. Central claims: (1) vocabulary borrowed from psychology carries an 'invisible grammar' of six epistemological assumptions (Sec. 3: continuous subject; inner-outer correspondence; developmental trajectory; somatic grounding; consistency-reliability; accountability-requires-understanding) into governance discourse; (2) these generate four failure categories (Sec. 4: miscalibrated trust, misattributed accountability, inadequate authorization, false observability); (3) remediation must operate at three levels, operationalized as a six-question Disciplinary Audit (Sec. 4.1) and a 37-term translation taxonomy — 23 Level-2 constructs and 14 Level-3 replacements — of which 22 entries (12 constructs, 10 replacements) appear in Tables 1-2, the rest on Zenodo. Cases: the February 2023 Bing Chat incident (conceded illustration), the EU AI Act (Arts. 5(1)(a), 3(39), 5(1)(f), 14(4), 15(1), 15(4); Recitals 29, 44, 72) as principal demonstration, NIST AI RMF as extension. The authors explicitly forgo empirical methods and machine-consciousness claims, restricting themselves to 'what vocabulary makes visible and invisible in governance design.'
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the 'disciplinary language transfer problem' in AI governance?
It is the paper's name for how vocabulary borrowed from psychology and organizational science (learning, memory, values, compliance, identity, trust) carries its home discipline's hidden assumptions — an 'invisible grammar' — into AI governance, calibrating rules and trust decisions to a continuous psychological subject with developmental history that deployed AI systems do not have.
What are the six epistemological assumptions the paper identifies?
(1) The continuous subject assumption (a self persists and bears responsibility), (2) inner-outer correspondence (behavior reveals inner states), (3) developmental trajectory (entities mature through experience), (4) somatic grounding (valuing requires a biological substrate), (5) consistency-reliability (consistent behavior evidences stable character), and (6) accountability-requires-understanding. The paper maps them to four failure types: miscalibrated trust, misattributed accountability, inadequate authorization, and false observability.
What does the paper criticize about the EU AI Act?
Applying its six-question audit, the paper argues Article 5(1)(a) ('deploys... purposefully manipulative') imports intent, Article 3(39) ('emotion recognition... infer') imports inner-state detection, Article 14(4) ('properly understand') models oversight as interpersonal comprehension, and Article 15(1) ('perform consistently... lifecycle') treats consistency as character. It proposes replacements such as 'biometric pattern classification' and 'documented performance metrics within specified tolerances.'
Is this an empirical study, and where is the full 37-term list?
No — the authors state it is a conceptual analysis using the Bing Chat case, the EU AI Act, and NIST's AI RMF as illustrations, with no data collection. The complete taxonomy (23 borrowed constructs, 14 replacement terms) is archived on Zenodo (DOI 10.5281/zenodo.21843894); the article text itself shows only 22 entries in Tables 1-2.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- The Zenodo supplementary (full 37-term taxonomy, DOI 10.5281/zenodo.21843894) is referenced but not reviewable from the embedded text; 15 of 37 entries were not examined.
- Quotations from the EU AI Act and NIST AI RMF are taken as printed; article/recital wording was not independently verified within this review.
- No figures present; Tables 1-2 were extracted as text and analyzed as such — table formatting may differ slightly from the PDF.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.