Breaking the Illusion of Artificial Consensus: Clone-Robust Weighting for Arbitrary Metric Spaces

Damien Berriaud, Roger Wattenhofer

TL;DR

This paper introduces a mathematical framework for "clone-robust" weighting, ensuring that coordinated campaigns of near-duplicate messages (astroturfing) cannot create a false sense of consensus. By treating message similarity as a continuum in a metric space rather than a binary "same or different," the authors provide a provably robust way to aggregate opinions that is resistant to AI-amplified replication.

What Problem Does This Paper Address?

In the age of generative AI, "Coordinated Inauthentic Behavior" (CIB) has evolved. It’s no longer just about bot farms posting the exact same sentence 10,000 times—which is easy to detect and filter. Instead, an attacker can use LLMs to generate thousands of variations of a single claim, each differing slightly in wording, tone, or framing.

To a human observer or a simple algorithm, these look like independent voices, creating an "illusion of consensus." Current systems struggle because they often treat similarity as binary. This paper addresses the "popularity bias" where frequency is mistaken for credibility. The goal is to design a weighting system where a cluster of 1,000 "near-clones" carries no more weight than a single, unique, independent opinion, regardless of the underlying data structure.

Methods

The authors move away from specific geometric assumptions (like Euclidean space) and propose a framework that works for arbitrary metric spaces.

  1. Metric Embedding: Messages are embedded in a space where "distance" represents dissimilarity.
  2. Neighborhood Graphs: For any given radius $r$, they construct a graph where messages are connected if their distance is $\le r$. Within these graphs, they identify "equivalence classes" of nodes that share the same neighbors (clones).
  3. Graph Weighting Functions: They define rules for assigning weights to nodes in these graphs based on two core principles: Symmetry (isomorphic nodes get equal weight) and Locality (adding a node only affects its neighbors).
  4. Integration over Radii: The "secret sauce" of the paper is Theorem 1. To make the weighting smooth and continuous, they don't just pick one radius; they integrate the graph weights over a range of radii $[0, \alpha]$. This transforms discrete graph properties into a continuous, Lipschitz-stable weighting function.
  5. Information-Theoretic Approach: They explore a "Maximum Entropy Principle" for graphs, selecting weight distributions that are as "uninformative" as possible under neighborhood-preserving coarsenings, which acts as a robustness criterion against adversarial manipulation.

Datasets

This paper is primarily theoretical and axiomatic. It does not evaluate against a specific real-world dataset (like a Twitter dump). Instead, it provides the mathematical foundation and an efficient algorithm (Algorithm 1) that can be applied to any dataset that can be represented in a metric space (e.g., text embeddings from a model like BERT or Ada).

Key Insights and Ideas

  • The Continuity Requirement: Earlier work on clone-robustness was "brittle"—a tiny change in a message's position could cause a massive jump in its weight. This paper uses integration to ensure that weights vary smoothly, making the system harder to "game."
  • The Sharing Coefficient: The authors introduce a new analytical tool to explain why a post has a certain weight. It quantifies how much influence is "shared" between two posts. This is a major step forward for the interpretability of these algorithms.
  • The "Spoiler Effect": A fascinating insight is the "tension" between symmetry and locality. They show that in simple models, adding a neighbor to an isolated node can sometimes decrease that node's weight—a counter-intuitive "spoiler effect" that they attempt to solve using clique-based constructions.
  • Topological Invariance: Unlike previous methods that only worked in flat, Euclidean spaces, this approach works on any metric, making it applicable to complex data structures like social graphs or hierarchical taxonomies.

Key Results

  • Theorem 1 (General Construction): The authors prove that any graph weighting function satisfying symmetry and locality can be turned into a full-blown clone-robust weighting function for any metric space.
  • Computational Efficiency: They provide an algorithm that computes these robust weights in $O(|S|^2 \cdot (\text{GraphEval} + \text{MetricEval}))$ time, making it practical for moderately sized sets of messages.
  • Axiomatic Strengthening: They provide a more rigorous definition of clone-robustness using Lipschitz constants, providing a "quantitative" guarantee of fairness rather than just a qualitative one.

Limitations and Future Work

  • Computational Complexity of Entropy: While the general framework is efficient, the "Maximum Entropy" version—which is the most robust against certain "spoiler effects"—is likely NP-hard to compute exactly, as it relates to the clique-partition problem.
  • The "Perfect" Function: The authors conclude with two conjectures, suggesting that no weighting function based on "maximal-clique-covers" can perfectly satisfy all their desired sharing axioms. This points to a fundamental structural tension in the geometry of influence.
  • Real-world Dynamics: The paper assumes a static set of messages. Future work would need to address how these weights evolve in real-time as a "conversation" unfolds on a platform.

Summary updated

Abstract

Independent media are central to democratic decision-making, yet recent technological developments, such as social media, pseudonymous identities, and generative AI, have made them more vulnerable to coordinated influence campaigns--usually referred to as Coordinated Inauthentic Behavior. By automatically generating large numbers of similar messages and news reports, such campaigns create an illusion of widespread support, and exploit the tendency of human observers and aggregation mechanisms alike to treat frequency as evidence of credibility or consensus. Clone-robust weighting functions offer a solution to this problem by assigning influence in a way that is insensitive to arbitrary duplication or near-duplication, as measured by a metric. This axiomatic framework rests on three principles: symmetry (equivalent elements are treated equally), continuity (weights vary smoothly under perturbations), and clone-robustness (adding duplicates or near-duplicates does not distort the overall distribution). We provide a general construction of clone-robust weighting functions that applies to arbitrary metric spaces, is entirely independent of the underlying topology, and admits efficient computation. Our approach identifies radius graphs as a natural invariant under cloning, and builds on graph weighting functions that satisfy a basic locality condition. We explore the resulting design space, starting with a simple family that satisfies the core axioms, and then identify explainability as a guiding criterion for navigating this design space. To this end, we introduce sharing coefficients that enable meaningful comparison and interpretation of different constructions, but require additional axiomatic principles. We then consider alternative constructions based on clique-covers, and unveil approaches using clique-partitions that are grounded in information-theoretic principles.
BibTeX
@misc{berriaud2026breaking,
  author = {Damien Berriaud and Roger Wattenhofer},
  title = {{Breaking the Illusion of Artificial Consensus: Clone-Robust Weighting for Arbitrary Metric Spaces}},
  year = {2026},
  eprint = {2602.24024},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2602.24024},
}
Figure