For the past few decades, scientists have tried to build models that mimic human societies. These models have used what’s known as agent-based modelling (ABM), where virtual agents follow basic, hand-coded rules to simulate human behaviour. However, these simulations always fell short. Human lives are complex, unpredictable, and messy, and simple rules couldn’t capture that chaos. Until now.
A groundbreaking study from Stanford University is shaking up how we think about artificial intelligence and its potential to reflect human behaviour. The work, led by AI researcher Joon Sung Park, has produced a new kind of virtual agent—one that mirrors the thoughts, behaviours, and values of real people in an astonishingly accurate way. And it’s not just for show. This could change how we understand and study human society in the future.
The research began with a simple concept: interview real people, capture their beliefs, values, and everyday thoughts, and feed that information into a large language model (LLM). Over 1,000 participants were involved in the study, each taking part in a two-hour interview covering everything from daily routines to political opinions. These interviews weren’t just for data collection. They were designed to ensure the LLM could learn the nuanced and deeply human aspects of how people think and feel.
Once the data was processed, the team had created over 1,000 AI agents—each one a digital twin of a real human participant. Then came the real test: could these AI agents reflect their human counterparts’ responses to social and cultural issues? To find out, they used the General Social Survey, a tool often employed to measure public opinion on various societal matters. The results were striking. Each AI agent predicted the opinions of their human counterpart with 85% accuracy.
This breakthrough is significant not just for AI, but for how we view human behaviour in general. Agent-based models have always struggled with accuracy because they were based on simplified assumptions about how people act. They lacked the rich, messy details of human life, often reducing complex behaviours to a series of overly neat, mechanical rules. But by training AI models on the real, organic complexities of human life, the team at Stanford has opened up a whole new way to simulate and predict human behaviour.
The implications of this research are vast and go far beyond academic interest. AI agents that can simulate human behaviour with this level of accuracy could be applied in countless areas. For instance, corporations could use these models for market research, predicting consumer responses to new products or services. Governments might apply them to test out policies before implementing them on a larger scale. And social scientists could use them for experiments, examining how different factors affect social dynamics.
Perhaps the most exciting part of this development is the potential for virtual environments that mimic real human societies. In 2023, Park’s team created a virtual town populated by 20 AI agents tasked with organising a Valentine’s Day party. At first glance, this might sound trivial, but the results were nothing short of fascinating. The AI agents started interacting with each other, brainstorming ideas, making shopping lists, and even creating invitations—none of it was pre-programmed. The behaviours that emerged were organic and unpredictable, just as they would be in the real world.
While the initial experiment was small in scale, the potential for expansion is huge. Imagine scaling this from a town of 20 agents to a full-scale society of millions. Add in increasingly powerful computing capabilities, and what you get is a digital mirror of humanity, a rich and diverse simulation where virtual people live, interact, and evolve.
In the future, such simulations could become a new platform for understanding everything from human behaviour to social structures. A “mirror world,” as some are calling it, could be used in ways we haven’t even begun to imagine. With billions of virtual people in complex, living digital environments, we could explore how societies evolve, how cultural norms develop, and how individuals make decisions in a variety of contexts. It’s a platform that could reshape everything from the way we study sociology to how we create virtual economies.
Governments could use these simulations to test the impact of policy changes. Want to know what might happen if a new tax law is passed or if a city’s infrastructure is changed? A simulation with millions of virtual people could provide an answer before a single real-world dollar is spent. For businesses, such platforms could revolutionise market research by offering unprecedented insights into consumer behaviour. Rather than relying on limited surveys or focus groups, companies could watch how virtual people interact with products, advertisements, and services in a fully realised environment. The possibilities are endless.
And it’s not just about big corporations or governments. The new era of social simulation could provide benefits to individuals as well. Imagine having access to a platform where you can watch how your decisions might affect your community or see how your personal beliefs align with the broader population. As these simulations become more advanced, they could provide a window into the complexities of human society, offering us new insights into our behaviours, preferences, and values.
However, as with any new technology, there are ethical considerations to keep in mind. The ability to simulate human behaviour so accurately raises important questions about privacy, consent, and the potential misuse of such models. What happens if these digital twins are used to manipulate public opinion or sway elections? How do we ensure that these simulations remain transparent and accountable? These are questions that will need to be addressed as the technology develops.
For now, though, the future seems clear. The era of digital humans is upon us. And while we’re still in the early stages, the implications are already massive. From transforming how we understand human societies to reshaping how businesses and governments operate, this new generation of AI-powered simulations has the potential to change everything. We are, in essence, creating a digital world that reflects the intricacies of our own, allowing us to explore and understand human behaviour on a level we never thought possible.
In the coming years, as these AI agents continue to evolve, we might see a new form of social experimentation that could be just as impactful as the internet itself. A platform not just for entertainment, but for learning, for understanding, and for advancing our collective knowledge of what it means to be human. The digital mirror is getting clearer every day, and the future is more human than we ever imagined.
The only question left is: are we ready for it?
