Consensus in Opinion Formation Processes in Fully Evolving Environments

Vincenzo Auletta, Angelo Fanelli, Diodato FerraioliAAAI 2019 · DOI

TL;DR

This paper introduces a mathematical model for opinion formation where everything—social relationships, individual stubbornness, and even "innate" beliefs—evolves simultaneously over time. The authors prove that as long as the social network is "ergodic" (influence eventually flows between everyone) and individuals have a "finite recall" of the past, the entire system will eventually reach a consensus.

What Problem Does This Paper Address?

Most classical models of social influence, such as the DeGroot (DG) or Friedkin-Johnsen (FJ) models, assume a relatively static world. In these models, you might update your opinion based on your friends, but your "innate belief" and the list of friends you trust usually stay fixed.

In reality, humans are more dynamic. We stop talking to certain people, we become more or less stubborn based on social pressure, and our core beliefs often shift slowly based on the opinions we’ve expressed in the past. The authors address the lack of a unified framework that accounts for this "fully evolving environment." They seek to answer: under what conditions can a society still reach a consensus when the very foundations of their interactions are constantly shifting?

Methods

The authors propose a model based on epochs. Within an epoch (a sequence of rounds), social ties and stubbornness levels are fixed, but expressed opinions update round-by-round. At the start of a new epoch, three things can change:

  1. Social Ties: The weights agents give to each other's opinions.
  2. Stubbornness: How much an agent values their own innate belief versus their neighbors' opinions.
  3. Innate Beliefs: Crucially, the model assumes an agent's "innate" belief for the current epoch is a weighted average (convex combination) of the opinions they expressed in previous epochs.

To analyze this, the authors represent the opinion update process using stochastic matrices. They define a recursive matrix $T^{(\ell, j)}$ that maps the initial opinions of all agents to the opinions expressed at any future time. The mathematical heavy lifting involves proving that this matrix eventually converges to a state where every row is identical, signifying that every agent holds the same opinion.

Datasets

Not discussed in the paper. This is a theoretical and mathematical research paper; it focuses on formal proofs and model definitions rather than empirical data or simulations.

Key Insights and Ideas

  • The Power of Ergodicity: The paper shows that the social network doesn't need to be connected at every single moment. Instead, it needs to be ergodic—meaning that over a long enough period, influence from any agent $x$ has a path to reach any agent $y$. This is a much more realistic requirement for shifting social landscapes.
  • Finite Recall is Essential: A major finding is the necessity of "finite recall." If an agent remembers and is influenced by an opinion they held infinitely far in the past, a single outlier from the beginning of time could prevent the system from ever settling. By assuming agents eventually "forget" or stop weighing very old opinions, consensus becomes mathematically guaranteed.
  • Generalization of Classics: The model is robust enough to "swallow" previous models. By setting the epoch length to infinity, you get the FJ model; by setting it to one round and making beliefs update instantly, you get the DG model.
  • Consensus vs. Stability: The authors distinguish between the system settling into a fixed (but differing) set of opinions and reaching a true consensus. They show that without "positiveness" in the influence matrices, the society might split into distinct, non-communicating communities (recurrent sets) that each reach their own internal consensus but disagree with each other.

Key Results

  • Convergence Theorem: The primary result (Theorem 2) proves that if the sequence of social graphs is ergodic and every agent has finite recall, the profile of opinions $z(t)$ converges to a consensus value $z^*$ as time goes to infinity.
  • Fixed Graph Simplification: For environments where the social graph is fixed, the authors show that simple strong connectivity and aperiodicity (standard properties in Markov chain theory) are sufficient to guarantee consensus, even with evolving beliefs.
  • Mathematical Framework: The authors successfully derived a closed-form recursive definition for the state of the system at any time $t$, providing a tool for others to analyze complex, co-evolving social dynamics.

Limitations and Future Work

  • Convergence Rate: While the paper proves that consensus will happen, it doesn't specify how long it takes. In a rapidly changing world, a consensus that takes a thousand years to reach is functionally useless.
  • Quality of Consensus: The paper doesn't explore whether the final consensus is "good" or "correct" (e.g., in a truth-seeking context). It only proves that everyone eventually agrees.
  • Experimental Validation: The authors mention preliminary experiments in the conclusion but suggest that extensive experimental activity is needed to see how different parameters (like the length of epochs) affect the speed of convergence.
  • Information Aggregation: A suggested future path is studying this model in "social learning" scenarios where agents start with noisy signals of a "ground truth" to see if the evolving environment helps or hinders the group from finding the right answer.

Summary updated

BibTeX
@inproceedings{auletta2019consensus,
  author = {Vincenzo Auletta and Angelo Fanelli and Diodato Ferraioli},
  title = {{Consensus in Opinion Formation Processes in Fully Evolving Environments}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2019},
  doi = {10.1609/aaai.v33i01.33016022},
}

Loading related content...

Figure