💻 computer science

Beyond Distributive Justice: Hermeneutical Fairness in Ad Delivery

This paper proposes a novel framework for online ad delivery that integrates Miranda Fricker's concept of hermeneutical injustice with traditional distributive justice to prevent the systematic withholding or distortion of interpretative resources for protected groups, demonstrating through simulations that balancing these epistemic concerns with economic utility can mitigate harm without significant economic loss.

Original authors: Camilla Quaresmini, Valentina Breschi, Jessica Leoni, Viola Schiaffonati, Mara Tanelli, Giulia De Pasquale

Published 2026-05-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Camilla Quaresmini, Valentina Breschi, Jessica Leoni, Viola Schiaffonati, Mara Tanelli, Giulia De Pasquale

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the internet as a massive, bustling town square. In this square, there are thousands of billboards (ads) trying to grab your attention. For a long time, experts have worried about Distributive Justice in this square. They ask: "Is everyone getting a fair number of billboards?" If Group A gets 100 billboards and Group B gets only 10, that's unfair. We've built tools to fix this, ensuring everyone gets an equal number of signs.

But this paper argues that just having the same number of signs isn't enough. There is a deeper, more subtle problem the authors call Hermeneutical Fairness.

To understand this, let's use a new metaphor: The Library of Meaning.

The Problem: Not Just "What" You See, But "How" You Understand It

The authors suggest that ads aren't just pictures; they are tools for understanding the world. They carry "interpretative resources"—concepts, labels, and ways of thinking that help us make sense of our lives.

The paper identifies two ways the town square can be unfair, even if everyone gets the same number of billboards:

Hermeneutical Deprivation (The Empty Shelf):
Imagine a group of people is systematically ignored by the billboard designers. They never see ads about a specific health risk or a new opportunity. It's not that they are being treated poorly; it's that the "library" of concepts they need to understand their situation is missing a whole section. They are left without the vocabulary to even realize they have a problem.
- Analogy: It's like trying to read a book in a language you don't know. Even if the book is handed to you, you can't make sense of it because the necessary words were never taught to you.
Hermeneutical Distortion (The Broken Mirror):
Now imagine another group is bombarded with ads, but the ads are weird, sensationalized, or oversimplified. They see the same message over and over, but it's twisted. They start to believe a distorted version of reality.
- Analogy: It's like looking at a funhouse mirror that makes you look huge or tiny. You see yourself, but the image is so warped that you can't recognize who you really are.

The Evidence: A Look Back at History

To prove this isn't just a theory, the authors dug into old data from 1986–1987 regarding AIDS advertising. They didn't use modern algorithms; they looked at how people reacted to real-world campaigns.

They found two patterns:

The Gap: Some groups (specifically heterosexuals in their data) were much less likely to see the ads than others. This meant they were missing out on the "concepts" needed to understand the risk.
The Confusion: For those who did see the ads, the reaction wasn't always positive. Sometimes, seeing too many sensationalized ads made people feel less worried or confused about the actual risks. It was as if the message was so loud and strange that it stopped making sense.

The Solution: A New Rulebook for Billboards

The authors propose a new way to design ad systems. Instead of just counting how many ads each group sees, we need to count how well the ads help people understand.

They introduce a "cost" to the system:

If you don't show an ad to someone who needs to understand a concept, the system pays a "deprivation cost."
If you show an ad that is so confusing or sensational that people misunderstand it, the system pays a "distortion cost."

They ran computer simulations to see what happens when you add these rules to the mix.

What They Found

The "Profit-Only" Approach is Dangerous: If a computer program only tries to make money (showing ads to people most likely to click), it naturally ignores the groups that are harder to reach or understand. This leads to Deprivation—the disadvantaged groups get left in the dark.
Fairness Costs a Little, But Saves a Lot: When the authors forced the system to be "Hermeneutically Fair" (ensuring everyone gets the right tools to understand), the system lost a tiny bit of profit (about 1-2%). However, this small cost prevented the massive unfairness of leaving groups without the knowledge they needed.
Equal Numbers Aren't Enough: Simply forcing the system to show the same number of ads to everyone (Distributive Justice) didn't fix the problem. You could show the same number of ads, but if the ads were confusing for one group and clear for another, the unfairness remained.

The Bottom Line

The paper concludes that we need to look at online advertising through two lenses:

Distributive Justice: Are the ads being shared equally?
Hermeneutical Fairness: Are the ads helping everyone understand the world, or are they leaving some people confused and others in the dark?

The authors argue that to truly be fair, we must ensure that the "tools for understanding" (the ads) are distributed in a way that helps everyone make sense of their lives, not just in a way that looks equal on a spreadsheet.

Technical Summary: Beyond Distributive Justice: Hermeneutical Fairness in Ad Delivery

Problem Statement
Current approaches to fairness in online advertising predominantly frame the issue as a problem of distributive justice, focusing on the equitable allocation of impressions, opportunities, or outcomes across protected groups. While this exposure-focused view addresses disparities in who sees an ad, the authors argue it is insufficient for capturing epistemic harms. Even when statistical parity is achieved, advertising can systematically disadvantage groups through the content of the ads and the interpretative frames they employ. This leads to two specific forms of harm:

Hermeneutical Deprivation: Systematic under-exposure where relevant interpretative concepts are withheld from specific audiences, preventing them from making sense of their experiences.
Hermeneutical Distortion: Systematic over-exposure to narrow, sensationalized, or skewed framings that saturate attention and mislead interpretation.

The paper posits that online ad delivery acts as a mechanism distributing interpretative resources (shared concepts and meaning-making tools). When these resources are unevenly distributed or distorted, it constitutes a form of hermeneutical injustice (drawing on Miranda Fricker's framework), which existing distributive fairness metrics fail to detect.

Methodology
The authors develop a formal framework to integrate hermeneutical considerations into ad allocation strategies, proceeding through three main stages:

Exploratory Empirical Analysis:
The authors analyze the AIDS Advertising Evaluation surveys (1986–1987) as a non-algorithmic baseline to identify patterns consistent with hermeneutical harms. They examine correlations between:
- Exposure Asymmetry: Differences in ad reach between heterosexual and homosexual respondents.
- Behavioral Uptake: The non-monotonic relationship between exposure and behavioral change (suggesting saturation effects).
- Framing Sensitivity: How different phrasings of risk (e.g., "people who need blood transfusions" vs. "hemophiliacs") affect awareness levels across groups, indicating that content framing, not just exposure, drives interpretative gaps.
Formal Modeling:
Building on the utility-maximization framework of Baumann et al. (2024), the authors introduce a hermeneutically-aware utility function.
- Hermeneutical Cost ($HI$): Defined for each user $x$ $x$ based on whether they receive the ad ( $d_x=1$ $d_{x} = 1$ ) and their probability of understanding it ( $\rho_x$ $ρ_{x}$ ).
  - If $d_x=0$ : A fixed exclusion penalty $\xi$ (deprivation).
  - If $d_x=1$ : A reward $\theta$ for successful uptake and a penalty $\omega$ for failed uptake (distortion).
- Objective: Maximize $U_H(d) = U(d) - \gamma HI(d)$ , where $U(d)$ is economic utility and $\gamma$ is a trade-off parameter.
- Constraints: The model enforces classical distributive constraints (Statistical Parity, Equality of Opportunity) and introduces a new Equality of Hermeneutical Opportunity constraint, ensuring that the share of users who understand the ad is equal across groups.
Simulation Study:
The authors conduct numerical simulations to explore trade-offs between economic utility, distributive fairness, and hermeneutical costs. They vary parameters representing:
- Economic asymmetries (opportunity costs of targeting).
- Asymmetric benefits of understanding.
- Risks of hermeneutical distortion (penalties for failed uptake).
- Costs of exclusion.
  They compare four allocation rules: Unconstrained, Parity of Exposure, Equality of Opportunity, and Equality of Hermeneutical Opportunity.

Key Contributions

Conceptual Shift: The paper reframes ad delivery not merely as the allocation of attention, but as the allocation of interpretative resources. It identifies "epistemic exclusion" and "epistemic bombing" as mechanisms leading to hermeneutical deprivation and distortion.
Formal Operationalization: The authors provide the first formal definitions for Hermeneutical Deprivation and Hermeneutical Distortion in the context of algorithmic ad delivery, translating philosophical concepts into quantifiable costs ( $\xi, \omega, \theta$ ) and constraints.
New Fairness Metric: Introduction of Equality of Hermeneutical Opportunity, a constraint that ensures groups have equal capacity to make sense of ad content, distinct from mere exposure parity.
Integrated Framework: A utility-driven allocation strategy that simultaneously optimizes for economic gain, distributive fairness, and hermeneutical fairness, allowing for the analysis of trade-offs between these objectives.

Results

Utility vs. Fairness Trade-offs: Imposing fairness constraints (distributive or hermeneutical) results in a modest reduction in total utility (typically less than 2% in the simulations). The "price" of hermeneutical fairness is low.
Limitations of Unconstrained Optimization: Without explicit constraints, utility-maximizing policies systematically under-deliver ads to disadvantaged groups (hermeneutical deprivation), particularly when economic asymmetries exist (e.g., higher opportunity costs for targeting a specific group).
Effectiveness of Constraints:
- Distributive Constraints: Enforcing statistical parity or equality of opportunity significantly reduces exposure disparities (bringing the difference close to zero) and prevents the structural sidelining of disadvantaged groups.
- Hermeneutical Constraints: While classical constraints reduce exposure gaps, they do not guarantee equal understanding. The study shows that policies satisfying exposure parity can still produce hermeneutical deprivation or distortion if the content itself is skewed.
Interaction of Costs: When the stakes of withholding ads are high (high $\xi$ ), distributive constraints effectively reduce hermeneutical costs with minimal utility loss. However, optimizing solely for hermeneutical cost without distributive constraints can lead to policies that over-concentrate ads on disadvantaged groups, potentially causing saturation.

Significance and Claims
The paper claims that fairness analyses in online advertising must expand beyond classical distributive notions to account for the epistemic conditions of interpretation and uptake.

Beyond Exposure Parity: The authors argue that ensuring equal exposure is insufficient; systems must also ensure that the interpretative resources provided by ads are accessible and intelligible to all groups.
Preventing Systemic Harms: By modeling ad delivery as a distribution of interpretative resources, the framework offers a principled way to detect and mitigate exclusionary dynamics that persist even under exposure parity.
Modest Scope: The authors explicitly state that their empirical analysis relies on offline survey data as a baseline to motivate the framework, not as a direct proxy for modern online systems. They acknowledge limitations in parameter identification (e.g., $\rho_x$ is latent) and call for future controlled experiments to validate uptake probabilities and calibrate epistemic parameters. The work is presented as a conceptual and formal foundation for future auditing and platform design, rather than a fully deployable solution for current ad systems.