Publish and Perish: How AI-Accelerated Writing Without Proportional Verification Investment Degrades Scientific Knowledge

Abstract: Artificial intelligence tools are accelerating manuscript production far faster than peer review capacity can expand. Applying the theory of constraints from manufacturing science, we formalize this asymmetry through a minimal two-variable ordinary differential equation model coupling review queue evolution and verification quality degradation via an endogenous, queue-pressure-driven review AI adoption mechanism. The causal chain is: writing AI adoption increases submissions, growing the review queue, which drives reviewer AI adoption under pressure, degrading verification quality and reducing net knowledge output. Under empirically informed parameters (writing acceleration γ = 2.0, review acceleration δ = 0.5), the model predicts a deceptive honeymoon where knowledge output peaks at 1.10K0 (circa 2026), followed by paradox onset at t = 6 years (2028) and long-term degradation to 0.68K0 (32% loss), approaching a steady state of 0.60K0 (40% loss). The critical condition for net benefit is δ > γ; the current operating point lies deep in the paradox regime. Empirical validation against NeurIPS, ICLR, arXiv, and bioRxiv submission data shows qualitative consistency with observed post-ChatGPT acceleration patterns. Policy analysis reveals that only combined interventions such as review infrastructure investment paired with institutional quality standards can restore positive knowledge production.

• The paper presents a model demonstrating that rapid AI-driven manuscript submissions without matched verification capacity lead to long-term degradation in net scientific knowledge.
• It employs a two-variable ODE framework and the theory of constraints to analyze dynamics between submission surges and declining verification quality.
• The study suggests that enhancing review AI capacity and enforcing quality floors are vital policy levers to counteract the observed 'knowledge paradox.'

Asymmetric AI Acceleration in Scientific Publishing: Model, Evidence, and Policy Implications

Overview

The paper "Publish and Perish: How AI-Accelerated Writing Without Proportional Verification Investment Degrades Scientific Knowledge" (2604.05714) presents a formal systems-level analysis of the impact of unbalanced AI acceleration in scientific manuscript production versus peer review and verification. Leveraging the theory of constraints (TOC) and a minimal two-variable ODE model, the study demonstrates that rapid AI-driven gains in manuscript submission rates, without commensurate advances in verification capacity, induce long-term degradation in net scientific knowledge output. The analysis delineates the causal links between AI adoption, submission and review queue dynamics, reviewer AI adoption, verification quality decline, and the resultant "knowledge paradox."

Theoretical Model and Core Dynamics

The authors construct an analytically tractable framework wherein two state variables—review queue length ($Q$) and verification quality ($q$)—capture the system's evolution. Writing AI adoption ($w(t)$) triggers an exogenous surge in paper submissions, while reviewer AI adoption ($q_r(t)$) is modeled as an endogenous, Michaelis-Menten-like response to queue pressure, saturating as workloads increase. Critically, reviewer AI provides only superficial throughput enhancement without substituting for deep domain verification.

The model is governed by:

• Submission Acceleration: $S(t) = S_0 (1 + \gamma w(t))$
• Review Throughput: $R(t) = R_{\max} (1 + \delta q_r(t))$
• Verification Quality Evolution: $$\dot{q} = -\lambda q_r(t) (q-q_{\min}) + \mu (1-n q_r(t))(1-q)$$

With parameter values empirically informed by recent data from NeurIPS, ICLR, arXiv, and bioRxiv, and key acceleration factors $\gamma = 2.0$ (writing), $\delta = 0.5$ (review), the analysis explores the parameter space relevant to the current publishing ecosystem.

Numerical and Analytical Results

The model predicts two distinct phases:

• Transient Honeymoon (2022-2028): Rapid growth in knowledge output due to initial throughput benefits, peaking at $1.10 K_0$ (relative to pre-AI baseline) by 2026.
• Onset of Knowledge Paradox (post-2028): A crossover occurs whereby incremental throughput is outpaced by verification quality erosion, resulting in sustained, monotonic knowledge loss. By 2042 ($q$0 years), net knowledge output falls to $q$1 (32% loss), converging analytically to a steady state of $q$2 (40% loss) under the current $q$3 regime.

The critical regime for sustainable knowledge enhancement is found where review acceleration exceeds writing acceleration ($q$4). Present data place the system firmly in the paradox regime, with empirical validation from recent conference and preprint server statistics.

The model robustly demonstrates that only under $q$5 can systems escape knowledge degradation, and that temporary peaks are misleading due to latent verification debt.

Empirical Validation

Comparison with real-world submission data shows strong consistency in both magnitude and qualitative trends. NeurIPS and ICLR, as AI-centric venues, show post-2022 CAGRs (27% and 55%, respectively) consistent with writing AI-driven $q$6. Simultaneously, observed reviewer AI adoption rates at ICLR (16% in 2024, 20% in 2025) align with model predictions for queue-driven adoption. In contrast, biology venues (bioRxiv) experience decelerated submission growth, reflecting lower AI penetration.

Policy Levers and Sensitivity Analysis

The analysis identifies two interdependent policy axes:

• Review Capacity Enhancement ($q$7): Investment in AI tools that augment human judgment, especially for deep methodological and reproducibility checks, is necessary but technologically constrained. Even doubling review AI capacity only partially mitigates the paradox unless it overtakes writing acceleration.
• Quality Floor Enforcement ($q$8): Editorial and institutional mandates—such as code/data availability, reproducibility standards, and enforced human-in-the-loop verification—can elevate the lower bound on verification quality and substantially buffer against knowledge loss, even under suboptimal $q$9.

Sensitivity analysis reveals that the human displacement fraction ($w(t)$0) is the most critical control parameter. Prioritizing AI tools that assist (rather than replace) human reviewers is shown to be more effective than focusing solely on throughput gains.

Theoretical and Practical Implications

The study advances the science-of-science literature by quantifying how verification constraints, long recognized informally, become a bottleneck aggravated by asymmetric AI adoption. The findings have several implications:

• Metrics Reform: Conventional bibliometrics (publication count, h-index) become actively misleading; evaluation systems must shift toward reproducibility, review depth, and correction rates.
• Incentive Realignment: Institutions, publishers, and funders must redesign incentives toward quality rather than throughput, including reviewer credit systems and limits on publication quotas.
• Algorithmic Development: There is a critical need for next-generation reviewer AI systems focused on substantive verification and flagging, not merely formal checks.
• Community-Wide Governance: The paradox is not self-correcting; collective action is necessary to avoid a tragedy of the commons scenario.

Limitations and Directions for Future Work

The model abstracts over heterogeneities in field, geography, and career stage, aggregates multiple facets of verification quality into a scalar, and does not capture adversarial or strategic dynamics among authors, reviewers, and publishers. Future extensions should incorporate stratified agent-based models, direct metrics of verification quality, and empirical measurement of latent verification debt (e.g., via reproducibility and correction statistics).

Conclusion

This paper demonstrates, via a formal dynamical framework, that disproportionate acceleration of manuscript generation relative to verification creates a systemic knowledge degradation risk. Only proportional investments in review infrastructure (technological and institutional) and enforcement of rigorous quality floors can avert long-term decline in reliable scientific knowledge. The analytical criterion—review acceleration must exceed writing acceleration—is actionable, emphasizing that infrastructural and policy response must keep pace with technological advances in manuscript production for the integrity of the scientific record.