The video explores Stanford researcher Jun Park’s new project, Simile, which uses AI-powered digital twins to simulate entire societies and predict human behavior in response to real-world events, attracting major industry backing and enterprise clients. It suggests this technology could shift decision-making from analyzing big data to running large-scale simulations, enabling organizations to anticipate outcomes and rare reactions more accurately and efficiently.
The video discusses a groundbreaking project led by Stanford researcher Jun Park, who previously created the “Smallville” experiment, where a simulated village was populated with AI agents powered by large language models. The original experiment aimed to see if digital agents could realistically mimic human social interactions and behaviors. Each AI character had its own backstory, personality, and daily routine, and the researchers tested how information, like the idea for a Valentine’s Day party, spread organically through the community. The results showed that the AI agents behaved in surprisingly human-like ways, with information diffusing through social networks and individuals responding differently based on their personalities and relationships.
Building on this, Jun Park has launched a new, much larger project called Simile, which aims to create entire digital societies using AI-powered “digital twins” based on real-world data such as transcripts, transaction logs, and scientific studies. The goal is to simulate complex social phenomena and answer questions about how populations might react to changes like new policies, marketing campaigns, or breaking news. The project has attracted significant attention and investment, with backing from prominent figures in the AI industry, including Andrej Karpathy (OpenAI co-founder), Fei-Fei Li (Stanford’s Human-Centered AI Institute), Adam D’Angelo (Quora CEO and OpenAI board member), Guillermo Rauch (Vercel CEO), and Scott Belsky (Adobe executive and Behance founder).
Simile’s technology is already being used by major enterprise clients such as CVS Health and T-Mobile, who are interested in using these simulations for market research, product testing, and user interface feedback. The simulations have proven to be highly accurate, with the AI agents correctly predicting 8 out of 10 analyst questions during simulated earnings calls. This level of predictive power could be extremely valuable for companies preparing for real-world events, allowing them to anticipate reactions and optimize their strategies accordingly.
The video highlights a potential paradigm shift from “big data” to “big simulation.” Traditionally, companies have relied on collecting and analyzing vast amounts of real-world data to make decisions. However, if simulations become accurate enough, it may become more efficient to generate insights by running large-scale digital experiments instead. This could dramatically reduce the “innovation tax”—the cost and risk associated with being the first to try something new—by allowing organizations to test thousands of scenarios virtually before committing resources in the real world.
Finally, the speaker notes that simulations like Simile could capture not just average responses, but also rare or extreme reactions that can have outsized effects in society—something traditional statistical analysis often misses. This could be transformative for fields like social science, economics, and even stock market analysis, where understanding the full range of possible human behaviors is crucial. The video ends with a philosophical reflection on the nature of reality, suggesting that as simulations become more sophisticated, it raises questions about whether our own world could be a simulation.