Could AI simulations soon replace traditional surveys and interviews?
A groundbreaking study by researchers from Fudan University suggests that it might be possible. The team has introduced SocioVerse, an innovative world model driven by large language model (LLM) agents, designed to simulate public opinions, behaviors, and reactions — all without relying on conventional survey methods.
What Is SocioVerse?
SocioVerse is a sophisticated framework built around four major components, designed to mirror the real-world social environment:
- Social Environment: Continuously integrates external, real-world information into the simulation.
- User Engine: Provides realistic user-profiles and contexts.
- Scenario Engine: Ensures that the simulated scenarios align closely with real-world events.
- Behavior Engine: Reproduces authentic human behaviors among AI agents.
The system is further enriched by a virtual user pool composed of profiles based on 10 million real individuals. According to the researchers, this multi-faceted design allows SocioVerse to create highly realistic simulations of human social dynamics.
Why Replace Traditional Surveys?
Traditional research methods, such as surveys, interviews, and observational studies, come with notable drawbacks — high operational costs, small or biased sample sizes, and ethical concerns regarding participant privacy and consent.
In contrast, SocioVerse offers a scalable, cost-effective alternative that can potentially overcome these limitations by using autonomous AI agents to simulate human decision-making at scale.
By deploying AI agents that can mimic human behavior, researchers can now observe micro-level decisions and predict macro-level social trends with greater efficiency.
Testing SocioVerse: Real-World Simulations
To validate the capabilities of SocioVerse, the Fudan research team conducted large-scale experiments across multiple domains, including politics, news, and economics. They employed several leading AI models, such as:
- Llama-3-70b-Instruct
- 5-72b-Instruct
- DeepSeek-V3
- GPT-4o mini
- GPT-4o
- DeepSeek-R1-671b
1. Predicting the U.S. Presidential Election
In one test, AI agents were tasked with forecasting state-level outcomes for the U.S. presidential election.
GPT-4o-mini and Qwen2.5-72b delivered highly competitive results, correctly predicting over 90% of state voting outcomes. In comparison, DeepSeek-R1-671b showed less accuracy, as researchers observed it tended to “overthink” the prediction scenarios, leading to deviations from actual results.
2. Simulating Public Reaction to Breaking News
In another experiment, researchers simulated public responses to a major news event: the release of ChatGPT.
Here, GPT-4o and Qwen2.5-72b were more closely aligned with real-world public sentiment, outperforming other models in terms of realism and consistency.
3. Analyzing Economic Behavior in China
Finally, the team simulated behaviors corresponding to a national economic survey in China. In this case, Llama-3-70b outshined its counterparts, accurately replicating the spending habits of individuals based on survey patterns.
Researchers Highlight Promising, Yet Imperfect Results
While the SocioVerse framework demonstrates remarkable potential, the team acknowledges that it is not yet flawless.
In their research paper, they noted, “Our findings indicate that state-of-the-art LLMs demonstrate a notable ability to simulate human responses in complex social contexts, although some gaps still remain between the simulated response and observed real-world outcomes.”
This acknowledgment highlights that while LLM-driven simulations can complement traditional methods, they may not fully replace them — at least not yet.
What’s Next for SocioVerse?
Looking ahead, the researchers aim to expand SocioVerse’s capabilities by exploring a broader array of real-world scenarios. By enhancing the realism and flexibility of AI-driven social simulations, they hope to create a tool that could revolutionize social science research, policymaking, marketing analysis, and beyond.
Read more: Germany’s Top 10 Women Entrepreneurs to Watch in 2025
