In democracies, major policy decisions typically require some form of majority or consensus, so elites must secure mass support to govern. Historically, elites could shape support only through limited instruments like schooling and mass media; advances in AI-driven persuasion sharply reduce the cost and increase the precision of shaping public opinion, making the distribution of preferences itself an object of deliberate design. We develop a dynamic model in which elites choose how much to reshape the distribution of policy preferences, subject to persuasion costs and a majority rule constraint. With a single elite, any optimal intervention tends to push society toward more polarized opinion profiles-a "polarization pull"-and improvements in persuasion technology accelerate this drift. When two opposed elites alternate in power, the same technology also creates incentives to park society in "semi-lock" regions where opinions are more cohesive and harder for a rival to overturn, so advances in persuasion can either heighten or dampen polarization depending on the environment. Taken together, cheaper persuasion technologies recast polarization as a strategic instrument of governance rather than a purely emergent social byproduct, with important implications for democratic stability as AI capabilities advance.
Deep Dive into AI 설득 기술이 민주주의 엘리트의 정책 편향과 양극화 전략을 재구성한다.
In democracies, major policy decisions typically require some form of majority or consensus, so elites must secure mass support to govern. Historically, elites could shape support only through limited instruments like schooling and mass media; advances in AI-driven persuasion sharply reduce the cost and increase the precision of shaping public opinion, making the distribution of preferences itself an object of deliberate design. We develop a dynamic model in which elites choose how much to reshape the distribution of policy preferences, subject to persuasion costs and a majority rule constraint. With a single elite, any optimal intervention tends to push society toward more polarized opinion profiles-a “polarization pull”-and improvements in persuasion technology accelerate this drift. When two opposed elites alternate in power, the same technology also creates incentives to park society in “semi-lock” regions where opinions are more cohesive and harder for a rival to overturn, so adva
Democratic policymaking is mediated by public support. Constitutions, electoral rules, and informal norms all embed some version of a majority or consensus constraint: policy can be implemented only if enough citizens, or their representatives, are willing to endorse it. For elites who care about which policies are chosen, this makes the distribution of public preferences a central strategic object. To govern, they must either accept the beliefs they inherit or invest resources in reshaping them.
Historically, the tools available to shape mass preferences have been blunt, dull, and slow. States and parties have relied on school curricula, public broadcasting, subsidized media, patronage networks, and propaganda campaigns to tilt opinion in their favor. 1 These instruments operate at coarse levels of targeting, require long lead times, and are costly to adapt. As a result, even powerful governments have typically faced high and relatively rigid costs of persuasion.
The rapid diffusion of modern AI is changing this constraint. Generative models, agentic systems, and platform-level tools make it possible to generate, test, and personalize persuasive content at scale, in real time, and at very low marginal cost. 2 A “nation of geniuses on the cloud” (Korinek [2024]) can now draft tailored messages, probe millions of individuals, and adapt narratives as feedback arrives. As these technologies lower and reshape persuasion costs, public preferences cease to be a fixed constraint and become something closer to a choice variable for elites. This raises a natural question: if elites can cheaply engineer the distribution of opinions, what distributions will they find optimal?
In this paper, we develop a dynamic model in which, in each period, an elite faces a binary policy choice, must secure a majority to enact its preferred policy, and can, at some cost, shift public support. The model’s key mechanism is the uncertainty elites face about future ideal policies, which may change over time. Because public preferences are sticky and costly to alter, shaping opinion today influences an elite’s ability to enact its preferred policy 1 See, among many others, DellaVigna and Kaplan [2007], Adena et al. [2015], Cantoni et al. [2017], and DellaVigna and Gentzkow [2010].
2 For recent evidence on LLM-based persuasion, see Salvi et al. [2025], Schoenegger et al. [2025], Tappin et al. [2023], Argyle et al. [2025], Bai et al. [2025], Simchon et al. [2024], and Hackenburg and Margetts [2024].
in the future.
Our analysis focuses on the degree of polarization that would arise if elites could shape public preferences. We define polarization as the distance from unanimous agreement: in a binary setting, society is maximally polarized when support is evenly split and minimally polarized when near consensus. The central object of choice is not the policy itself but the distribution of mass preferences that governs future policy contests.
Our central departure from the existing polarization literature is conceptual. A large body of work studies polarization as the outcome of deeper forces such as changes in income distributions, identity cleavages, media markets, party strategies, or social networks-and then analyzes its consequences for turnout, policy, and welfare. 3 We instead treat polarization as the result of a choice: a policy instrument in the hands of elites who face majority or consensus constraints and have the technology to reshape public opinion. In our framework, elites endogenously decide whether to maintain a cohesive society or to manufacture a more divided one, given their expectations about future states of the world and the technology of persuasion.
We begin by analyzing the case of a single, forward-looking elite. In each period, the state of the world determines which of two policies is ex post desirable. The elite observes this state and can adjust public support before the majority rule determines policy. With costly persuasion technology, we show a robust polarization pull: whenever the elite chooses to influence public opinion, society moves weakly toward maximal polarization. Intuitively, a distribution of opinions clustered near the majority threshold provides insurance against future uncertainty, allowing the elite to respond quickly and cheaply to changing priorities. If the state flips in the next period, the cost of shifting public support across the threshold is minimal. Polarization is therefore not intrinsically valuable to the elite but strategically attractive because a divided society is easier and cheaper to steer in the face of shocks. As persuasion costs decline-reflecting, for example, increasingly capable AI-the model predicts faster convergence toward highly polarized opinion profiles.
We then introduce a second elite with diametrically opposed preferences and alternating control. When today’s leader anticipates that tomorrow’s rival will also be able to reshape public opinion, the probl
…(Full text truncated)…
This content is AI-processed based on ArXiv data.
