[AAMAS] Manufacturing Consent: How Strategic Edge Addition Can Guarantee Election Outcomes

TL;DR

Researchers from the University of Salerno have developed an algorithm that can guarantee the victory of any chosen candidate in a social network election. By strategically adding a handful of "friendship" links (edge addition), they can manipulate the information flow to strategic voters, ensuring the system reaches a stable equilibrium where the sponsored candidate wins—every single time.

The Fragility of Digital Democracy

In a world where Facebook "Friend Suggestions" or LinkedIn "People You May Know" shape our social circles, we often assume our voting choices remain our own. However, Social Choice Theory suggests otherwise. In an Iterative Voting setting, voters aren't just expressing a preference; they are playing a game. If they see their favorite candidate has no chance, they might strategically vote for a "lesser evil."

The core problem identified in this paper is the Instability Cascade. Prior work attempted to manipulate networks by adding edges, but these new links often caused a chain reaction: Voter A changes their mind, which makes Voter B change theirs, and suddenly the entire election flips in an unpredictable direction. Previous heuristics failed 30% of the time because they couldn't "freeze" these non-target voters.

The Insight: Stabilizers and Blockers

The authors' breakthrough lies in their mathematical approach to Node Stabilization. Instead of just overwhelming a voter with new information, they identify two critical sets of candidates for every voter:

1. Stabilizers ($S{t}_i$): Candidates whose supporters, if connected to the voter, will force that voter to revert to their "truthful" favorite candidate.
2. Blockers ($B{l}_i$): Candidates who might seem like stabilizers in Round 1 but would actually trigger a strategic shift in Round 2, making the voter unstable again.

By specifically avoiding "Blockers" and connecting voters to precisely 1 or 2 "Stabilizers," the algorithm ensures that once a voter chooses their favorite, they have no incentive to ever change it again, regardless of what the rest of the network does.

Methodology: The Two-Phase Attack

The algorithm operates in two sophisticated stages:

1. The Influence Phase

The "Attacker" identifies voters who can be swayed to vote for the sponsored candidate $w$. By adding edges between these voters and $w$-supporters, they create a "false majority" in the voter's local view, inducing a strategic shift.

2. The Stabilization Phase (The Secret Sauce)

To prevent the rest of the network from reacting negatively to the Influence Phase, the algorithm partitions the remaining voters:

• Superseeds & Seeds: A tiny core of voters connected strategically to act as "anchors."
• The Rest ($B$): Connected to seeds to ensure they stay truthful.

Figure 1: The stabilization logic ensures that even if local views change, the strategic equilibrium remains fixed.

Experimental Battleground: Facebook MHRW

The researchers tested their algorithm against the "Facebook MHRW" dataset (a sample of 900,000 real social connections). They compared their performance against the previous industry standard (Sina et al., 2015).

The Results:

• Success Rate: The new algorithm achieved 100% success, whereas the baseline failed in 3 out of every 10 elections.
• Efficiency: Despite the absolute guarantee of winning, the number of added edges remains in the same order of magnitude as the less-reliable heuristic.
• Convergence: The election process is guaranteed to end in just 2 rounds, preventing long-term volatility.

Figure 2: Success rate and edge addition count comparison between the proposed algorithm and prior heuristics.

Critical Insight: The Power of the Controller

This research is a "white hat" exposure of a "black hat" reality. It proves that a social media platform doesn't need to delete your posts or shadowban your favorite candidate to rig an election. By simply suggesting the "right" friends to the "right" people, they can create a local information environment where your most rational, strategic move is exactly what the controller wants you to do.

Limitations & Future Work

While mathematically robust, the model assumes the controller knows the full preference rankings of all voters—a feat that requires massive data harvesting. However, with modern AI and "Big Data," this assumption is becoming uncomfortably realistic. Future research will likely look at how these manipulations hold up when voters have "limited rationality" or when the information they receive is "blurry" or noisy.

Conclusion

This paper serves as a stark warning: the structure of our networks is just as important as the content within them. To protect democracy, we must look beyond content moderation and start scrutinizing the algorithms of connection.