AI for Good? AI Finds Lasting Peace in Unexpected Places

Source: Photo by Alexandra Mirgheș on Unsplash

Written by Larry S. Liebovitch and Peter T. Coleman

The quest to understand and promote world peace has long relied on traditional metrics like GDP, crime rates, and expert surveys. But new research suggests we may be able to measure peace in a completely different way: by analyzing the language used in a society’s news media.

In a series of studies led by researchers at Columbia University, machine learning algorithms were able to distinguish between higher and lower-peace countries based solely on the words central to their news coverage. Even more remarkably, this AI-powered analysis revealed linguistic patterns that challenge our assumptions about what peace looks like on the ground.

The traditional view of peace often focuses on the absence of conflict—no war, no violence, no strife. But when expert researchers initially hypothesized which words would be most common in peaceful versus less peaceful nations, they were largely wrong. The terms they expected to find, like “harmony,” “cooperation,” and “unity” in peaceful countries or “violence,” “conflict,” and “aggression” in less peaceful ones, turned out not to be the key differentiators at all.

Instead, the AI discovered something far more subtle and profound. News media in more peaceful countries tends to focus on the fabric of daily life—stories about work, family, games, creativity, and community. The language is more diverse and informal, suggesting a society comfortable with multiple viewpoints and ways of being. In contrast, news from less peaceful nations is dominated by words related to government, politics, control, and formal power structures, with less linguistic diversity overall.

This finding hints at a deeper truth about peace: It may be less about grand declarations of harmony and more about the small stuff of everyday life being allowed to take center stage. When a society is truly at peace, perhaps the most notable thing is that people can focus on living rather than surviving, on creating rather than controlling.

The research team’s methodology is unorthodox. Rather than starting with human assumptions about what matters, they let the AI examine millions of words across more than 700,000 articles from nine high-peace vs. nine low-peace countries to find meaningful patterns. This “data-driven” rather than “hypothesis-driven” approach allowed them to discover insights that might have remained hidden if they had stuck to testing predetermined theories.

The results weren’t just academic—they were remarkably practical. The machine learning model, trained on this linguistic data, was able to create a “peace index” that strongly correlated with traditional measures of peace that require extensive surveys and expert analysis. But unlike those standard indices, which are updated annually at best, this AI system could potentially monitor peace in real time by analyzing changes in ongoing news coverage.

This has profound implications. Imagine being able to detect early warning signs of deteriorating peace by spotting shifts in media language or being able to measure the impact of peace-building initiatives through changes in how a society talks about itself. The researchers are already exploring how their system might predict future changes in peace levels.

Perhaps most intriguingly, the studies found that the U.S. was an outlier—scoring high on traditional peace measures but showing linguistic patterns more typical of less peaceful nations in terms of political focus and limited positive intergroup dynamics. This raises important questions about the relationship between institutional stability and genuine societal peace.

Artificial Intelligence Essential Reads

Are the Systems We Have Created Now Controlling Us?

The Doctor Will See You—Again and Again

The research also points to a broader revolution in how we study society. Traditional social science starts with theories and tests them against data. But new AI tools allow us to reverse this process—letting patterns emerge from vast amounts of data that human researchers might never have thought to look for.

This isn’t to say that AI can replace human insight—indeed, it took local experts from the nations studied to make sense of the linguistic patterns the AI discovered and connect them to broader theories about how peaceful societies function. But it suggests that AI can be an invaluable partner in helping us see past our preconceptions and recognize how peace manifests in the real world.

At a time when many worry that AI will mainly be used to promote conflict and division, this research offers an inspiring counter-example. Here is AI being used not to manipulate or control but to help us better understand the delicate social dynamics that allow human beings to live together in peace.

The research reminds us that peace isn’t just the absence of war—it’s the presence of a society functioning well enough that people can focus on living, learning, playing, and creating together. And while we may not always be able to see these patterns with our own eyes, AI is helping us understand what peace looks like in the wild. Now, the challenge is to use these insights to help more societies achieve it.

Larry S. Liebovitch is a physicist and Adjunct Senior Research Scholar at the Advanced Consortium on Cooperation, Conflict and Complexity in the Climate School at Columbia University. He has studied complex systems with many strongly interacting pieces in the physical, biological, and social sciences, and most recently used machine learning and artificial intelligence to study peace.

Peter T. Coleman is a professor of psychology and education at Columbia University and a renowned expert on constructive conflict resolution, intractable conflict, and sustaining peace. His most recent book, The Way Out: How to Overcome Toxic Polarization (2021), was published by Columbia University Press.