A theoretical paper by Nadav Kunievsky examined how increasingly inexpensive, precise AI persuasion could change the incentives of political elites seeking majority support. The model treats the public distribution of policy preferences as something leaders can deliberately influence, rather than assuming that opinion is fixed before political competition begins.
The study starts from a democratic constraint: major decisions generally require a majority or broader agreement, so an elite must obtain public support to govern. Historically, the paper says, institutions such as schools and mass media offered relatively limited ways to shape that support. AI-driven persuasion could reduce the cost and increase the precision of intervention, making more targeted changes to public preferences feasible.
Kunievsky builds a dynamic model in which an elite chooses how much to alter the preference distribution while facing both persuasion costs and majority rule. In the single-elite setting, the optimal intervention tends to move society toward a more polarized profile. The paper calls that tendency a polarization pull. As persuasion technology improves and intervention becomes cheaper, the modeled drift accelerates.
The result changes when two opposing elites alternate in office. Each side must consider not only the policy it can secure now but also how easily its rival could reverse the public alignment later. Under some conditions, the model gives an incumbent an incentive to move opinion into what the paper calls a semi-lock region: a more cohesive configuration that is difficult for an opponent to overturn. Cheaper persuasion can therefore increase or reduce polarization depending on the competitive environment.
That conditional finding is central. The paper does not claim that better AI persuasion will always split the public, nor that cohesion is necessarily benign. Instead, it portrays the shape of public opinion as a strategic variable. A concentrated distribution may be valuable to one elite because it raises the cost of a rival's future campaign, while a polarized distribution may make a current majority easier to construct in another setting.
The work is an economic model, not an empirical measurement of an election or a test of a deployed persuasion system. Its conclusions follow from the structure and assumptions described by the author. The supplied record does not establish how closely real political actors, voters or AI tools match those assumptions.
Submitted to arXiv on December 3, the paper is categorized under general economics, artificial intelligence, and computers and society. Its contribution is to frame polarization as potentially designed through strategic persuasion, with technology changing the price of that strategy. The analysis suggests that assessing AI's democratic effects requires asking who controls persuasion, whether power alternates and what opinion pattern best protects an incumbent's objectives.
