Price the harm: charge firms for how much personal data reduces uncertainty about intersectional identities using mutual information, creating a model-agnostic Pigouvian surcharge that internalizes privacy externalities and can be calibrated to social values.

Main Finding

The paper proposes a Pigouvian-style surcharge on the extraction of intersectional personal data that prices the social harm of inferential power. Concretely, the surcharge charges platforms an amount proportional to the mutual information I(X; S) between a collected datum X and a multivariate protected identity S = (S1,...,Sm). The socially relevant price of a datum is V(X) = cp + λ I(X; S), where cp is private marginal cost and λ is a policy parameter expressing society’s valuation of intersectional privacy. This rule is model-agnostic (independent of the inference algorithm) and operationalizes intersectional privacy harm as measurable entropy reduction.

Key Points

• Intersectional harm: Privacy damage is not additive across single attributes; harms arise from joint configurations of protected attributes (race, gender, disability, etc.), which data-driven inference can reconstruct.
• Information-theoretic measure: Mutual information I(X; S) captures the expected reduction in uncertainty about the full intersectional profile caused by observing X and thereby aligns naturally with intersectional exposure.
• Pigouvian interpretation: λ I(X; S) is a corrective surcharge that internalizes the social cost of intersectional inference, aligning private extraction incentives with social welfare.
• Normative requirements: A defensible price must (a) depend only on intersectional inference, (b) be zero when no inference is possible, (c) increase monotonically with informational exposure, and (d) be additive across independent disclosures—criteria under which mutual information gives a unique solution.
• Model-agnostic and implementable: Because mutual information is defined over joint distributions, the valuation is independent of whether inference is parametric, nonparametric, or machine-learned; practical approximation via discretization of joint distributions is possible.
• Distributive effect: The surcharge does more than correct an externality — it redistributes by making platforms internalize harms disproportionately borne by multiply-marginalized groups and creates the possibility of restitution or public funding.
• Political parameter: λ is inherently political/normative; its calibration determines trade-offs between profiling-enabled benefits and protection of intersectional privacy.
• Limits and enforcement risks: Effectiveness depends on regulatory capacity, ability to estimate I(X; S) under opacity, and market structure (dominant firms may absorb or pass on costs). Pricing complements but does not replace broader institutional change.

Data & Methods

• Conceptual/theoretical framework: The paper develops a political–economic model linking data valuation to information theory rather than presenting novel empirical estimation.
• Formal representation:
  • Protected intersectional identity S modeled as a multivariate random variable S = (S1,...,Sm).
  • A collected datum X reduces uncertainty about S; mutual information I(X; S) quantifies expected entropy reduction.
  • Pricing rule: V(X) = cp + λ I(X; S).
• Normative axioms: The valuation rule is derived from minimal Pigouvian-style requirements (dependence on inference, nullity when no inference, monotonicity, additivity).
• Implementation notes:
  • The rule is model-agnostic and applies regardless of the statistical/inference technique.
  • Estimation can proceed via discretization of joint distributions or with parametric/nonparametric/ML estimators of mutual information, subject to data access and auditability.
  • Policy design requires calibration of λ and institutional choices about revenue allocation (individual compensation, collective redress, public goods).
• Recognized practical challenges: estimating I(X; S) under platform opacity, potential strategic manipulation of reported leakage, and enforcement in concentrated markets.

Implications for AI Economics

• Incentives and platform behavior: Attaching an explicit monetary cost to intersectional inferential gains changes the marginal calculus of platforms — reducing extraction where social cost exceeds private benefit and encouraging privacy-preserving designs where leakage is costly.
• Data market pricing: Introduces a principled, operational price component tied to social harm (mutual information) rather than willingness-to-pay or marginal predictive value, shifting the basis for data valuation in markets and procurement contracts.
• Redistribution and compensation: Converts diffuse informational harms into fiscal receipts that can be used for restitution, funding social protection, or financing regulatory enforcement—embedding redistribution into market regulation.
• Complement to fairness/DP: The surcharge is regulatory/economic rather than purely technical; it complements algorithmic fairness or differential privacy by making the harms of intersectional inference economically salient and regulable.
• Model-agnostic regulatory tool: Because the rule does not depend on the internal model used by platforms, it is robust to model changes and obviates the need to audit every algorithmic pipeline—focus is on statistical leakage measured at the distributional level.
• Political economy and market structure: The mechanism highlights that λ is a political choice; in concentrated markets, effectiveness may be limited if dominant firms can absorb costs, implying the surcharge should be combined with competition, transparency, and audit policies.
• Measurability and enforcement demands: Realizing this approach requires regulatory capacity to estimate mutual information, audit platform claims, and prevent strategic reporting/manipulation; it therefore increases demand for empirical and institutional investments in measurement, standardization, and oversight.
• Limits to substitution: Pricing can discipline accumulation of informational capital but cannot alone transform underlying relations of power; it is a partial corrective within a broader policy toolkit (e.g., prohibitions, data governance, platform regulation).

Overall, the paper provides a compact, theoretically grounded proposal to operationalize intersectional privacy harms in market terms via mutual information and a Pigouvian surcharge. It bridges information theory and political economy to yield a model-agnostic instrument that makes inferential power traceable and regulable, while highlighting political calibration and enforcement as central practical challenges.