The collapse of an authoritarian regime does not begin with tanks in the streets. It starts with whispers among people watching from the shadows. Over the last week, thousands of Iranians streamed out of Tehran. Some fled Israeli airstrikes. But others, judging by emerging patterns using sentiment analysis, are voting with their feet and their words. There are growing signs that segments of the population no longer trust the Islamic Republic to protect them. While missiles fall, a quieter signal rises across the society: a shift in public language, a withdrawal from social participation, and a reframing of blame from “the regime” to the supreme leader himself.
Predicting whether these weak signals will cascade into widespread unrest and a threat to the Islamic Republic during Israel’s air campaign is an impossible task for intelligence frameworks built around traditional strategic warning and structural analytical frameworks. The changes are too subtle and nonlinear, confounding the use of traditional forecasting methods. But viewed through the lens of complex systems theory, they matter. The collapse of any authoritarian regime is rarely linear; it’s emergent—triggered by feedback loops, cascading grievances, and changes in how people think and talk about power.
This case calls for thinking about new methods to help analyze the resilience of authoritarian regimes and how best to support civil society organizations and democracy promotion. It calls for embracing techniques that analyze language and even the arts to see how the mood of a society adjusts to shocks and changing political events.
This commentary serves as both an assessment of how to study the prospect of unrest in authoritarian states like Iran as well as a call to arms for new approaches to analyzing social trends and promoting democracy. As the Department of State reorganizes, it should build an analytical capacity for analyzing sentiment across even the most repressed societies, tasks that used to reside in a suboptimal scattering of programs across The United States Agency for International Development (USAID) and the Global Engagement Center. Outside of government, philanthropic actors should explore how they can fund new digital tools for helping activists better organize and challenge authoritarian states on the ground. This process must start with the ability to track the type of localized interactions and feedback loops prone to sudden sparks of discontent.
From Global to Local Models of Discontent
Historically, governments and social scientists have long tried to predict unrest using macro-level indicators like regime type, GDP per capita, military expenditure, and demographic pressure. One of the earliest and most controversial efforts was Project Camelot, a 1964 U.S. Army–funded research program designed to develop a scientific model for predicting revolution. Later efforts became more rigorous but were still largely linear and based on forecasting derived from statistical correlations. In the early 1990s, at the request of then–Vice President Al Gore, the CIA created the State Failure Task Force to study when and why modern states fail. The resulting models pointed toward a global model of instability that saw increased risk of unrest and collapse in countries with certain regime types (i.e., anoncracy), high infant mortality rates, economic decline, and factionalized elites.
Yet even the best models struggled with timing and underestimated nonlinear triggers, especially how grievances coalesced into collective action. The models were good at predicting increasing risk windows but not at the types of local dynamics and contentious politics that would signal a new wave of mobilization. Researchers building global models struggled with everyday acts of resistance and collective action frames that served as better reference points for understanding sentiment.
Now, technology makes it possible to study these trends in the aggregate. Natural language processing makes it possible to analyze how local language changes over time, from underlying sentiment to emerging frames for collective action. Past research has shown that topic modeling and text classification can reveal underlying political preferences before they manifest in elections or protests. Daily speech is laden with emotions, beliefs, and politics, even when groups need to conceal their preferences from censors, creating a framework to analyze and understand societies. Algorithms can now even interpret how online communities use sarcasm as a subtle form of discontent. And this logic extends beyond language: Application programming interfaces (APIs) in platforms such as Spotify make it possible to analyze mood based on music preferences.
A New Approach to Early Warning and Democracy Promotion
Even the most authoritarian state struggles to repress all language. As a result, shifts in tone in chat-room banter may be the best starting point for analyzing growing discontent in digital authoritarian states where monitoring is constant. How people describe their daily lives in online forums is dataset rich with meaning that can be mined to better understand emerging trends and when it makes sense to openly challenge a dictator.
This perspective means it is time to rethink some of the key tasks required to promote democracy in the twenty-first century. There is still bipartisan consensus that authoritarian regimes such as the Chinese Communist Party are a threat to democratic values and free societies. What is missing is a suite of tools that help diplomats and activists better analyze social trends. Imagine an online toolkit secured by a virtual private network (VPN) that civil society leaders could use to analyze social trends, balancing their observations with larger patterns that emerge from sentiment analysis. The same tool could be used by diplomats to gain a deeper understanding of local dynamics often opaque from the confines of a capital city. Such a tool would need to be protected from malicious monitoring and malware, and constantly updated to ensure it has the right data and computational power to support use by a mix of U.S. government actors and civil society organizations.
Like ecosystems on the edge of collapse, authoritarian regimes appear stable until one small shock pushes the system across a tipping point. Diplomats and civil society leaders need a new approach to early warning that embraces complexity and helps them understand society as a living organism as opposed to a static model.
Benjamin Jensen is director of the Futures Lab and a senior fellow for the Defense and Security Department at the Center for Strategic and International Studies (CSIS) in Washington, D.C.
