Operationalising Extended Cognition: Formal Metrics for Corporate Knowledge and Legal Accountability

TL;DR

The paper addresses how to attribute corporate knowledge in an era of AI augmented cognition by reframing knowledge as a measurable, dynamic capability. It develops a formal framework with a continuous organisational knowledge metric $S_S(\varphi)$, a thresholded predicate $K_S(\varphi;\theta_C)$, and a firm-wide epistemic capacity index $\mathcal{K}_{S,t}$, linking epistemic state to legal standards. The approach combines a Search-Generation-Verification model, statistical validation, calibration, and generalisation to produce auditable validation certificates that support legal imputation of knowledge. It demonstrates how retrieval and verification pipelines, such as RAG, can expand the epistemic frontier and generate concrete accountability artefacts, enabling more robust governance and adjudication in the algorithmic age.

Abstract

Corporate responsibility turns on notions of corporate \textit{mens rea}, traditionally imputed from human agents. Yet these assumptions are under challenge as generative AI increasingly mediates enterprise decision-making. Building on the theory of extended cognition, we argue that in response corporate knowledge may be redefined as a dynamic capability, measurable by the efficiency of its information-access procedures and the validated reliability of their outputs. We develop a formal model that captures epistemic states of corporations deploying sophisticated AI or information systems, introducing a continuous organisational knowledge metric $S_S(\varphi)$ which integrates a pipeline's computational cost and its statistically validated error rate. We derive a thresholded knowledge predicate $\mathsf{K}_S$ to impute knowledge and a firm-wide epistemic capacity index $\mathcal{K}_{S,t}$ to measure overall capability. We then operationally map these quantitative metrics onto the legal standards of actual knowledge, constructive knowledge, wilful blindness, and recklessness. Our work provides a pathway towards creating measurable and justiciable audit artefacts, that render the corporate mind tractable and accountable in the algorithmic age.

Figures (3)

• Figure 1: Distribution of pipeline scores ($s_{\pi}$) across 15 simulation runs for each legal doctrine. Each box represents the interquartile range of observed scores for a given doctrine under LegacyCorp (pink) and ModernCorp (blue) pipelines. The red dashed horizontal line marks the knowledge threshold ($\theta_C = 0.7$), which defines the minimum score required for a system to satisfy the formal knowledge predicate $\mathsf{K}_S(\varphi; \theta_C)=1$. Across all doctrines, ModernCorp consistently exceeds the threshold, indicating sufficient epistemic reliability and efficiency, whereas LegacyCorp remains below the threshold, reflecting inadequate epistemic capacity. The stability of the boxplots demonstrates that these differences persist across repeated trials, highlighting a structural epistemic gap between the two corporate information systems.
• Figure 2: Epistemic scalability: comparison of time-to-knowledge ($\tau$) growth rates for LegacyCorp and ModernCorp as corporate corpus size increases. LegacyCorp exhibits linear scaling ($O(n)$), with time-to-knowledge rising steeply from approximately six seconds for 60 documents to over 100 seconds for 1,000. ModernCorp, by contrast, achieves logarithmic-time scaling ($O(\log n)$), maintaining nearly constant retrieval time even as the dataset expands. This widening gap illustrates how algorithmic efficiency compounds epistemic advantage: as data volumes grow, efficient architectures expand the feasible set of knowable facts. Legally, this implies that failure to adopt scalable search architectures may render a firm's ignorance increasingly unreasonable, reflecting an emergent “epistemic–culpability gap” aligned with its technological lag.
• Figure 3: The price of unreliability: degradation of the pipeline score ($s_{\pi}$) as the verifier error rate ($\varepsilon_{\mathrm{ver}}$) increases. The blue line traces how the expected knowledge score declines with rising verification error, holding retrieval efficiency constant. The red dashed line marks the knowledge threshold ($\theta_C = 0.7$) required for epistemic sufficiency, while the shaded region indicates the "unreliable zone," where verifier error rates above 15% drive the system below the legal knowledge threshold. The plot demonstrates that even highly efficient systems lose epistemic adequacy once reliability falls below this critical point, emphasizing that validation and calibration of AI verifiers are indispensable for legally cognizable knowledge.