Joon Sung Park on Scaling Simulations to 8 Billion People

Key Takeaways

• Simile AI is mapping early scaling laws in agent simulation: pouring in more behavioral data and compute produces predictable, repeatable accuracy gains.
• Joon Sung Park targets a digital twin engine for all 8 billion people to model multi-agent "wicked problems" like climate agreements and economic instability.
• Simile blends deep interviews, transaction logs, and randomized controlled trials to reach 85% fidelity when modeling human choices.
• Park expects future societal simulation runs to cost tens of millions of dollars, matching the compute budgets of frontier base models.

The Emergence of Simulation Scaling Laws

When Joon Sung Park published the Stanford "Smallville" Generative Agents paper, it looked like a delightful toy: 25 generative agents throwing Valentine's Day parties in a pixelated town. Today, as CEO of Simile AI, Park is pursuing a much larger compute thesis. Simile builds behavioral digital twins from deep interview transcripts, transactional data, and randomized controlled trials, reaching 85% fidelity against real human decisions.

Park sees the same dynamic playing out in social physics that OpenAI found in language. “The thing that we're actually seeing is the early glimpse of scaling law in simulations,”▶ 41:57 Park explains. “The more data about humans and more compute you ingest, you actually start to get predictive and predictable gains of the model performance in simulating and predicting people.”▶ 42:03

Standard foundation models predict the next token based on internet text. They fail at specific behavioral quirks because general web crawls do not capture personal incentives or private trade-offs. By capturing high-density behavioral inputs, Simile creates agents that react like specific human cohorts under economic and social stress.

Solving the Wicked Coordination Trap

Park's long-term target is planetary scale. “Can we create a simulation of 8 billion people living on earth?”▶ 42:51 Park asks. “I think that's quite interesting and that really is the vision and once you get to that kind of state the kind of questions that you can help answer for the society also start to change from my perspective and for me it's questions like can we help solve climate change.”▶ Play

Climate policy fails in the real world because of competing game-theoretic incentives across billions of independent actors. Social scientists classify these as wicked problems: systems where every intervention creates unpredictable ripples and trial-and-error in production is fatal.

“If you look at climate change as a problem space, this is what we like social scientists would often call the wicked problems problem where you have many actors with competing incentives who are trying to make a very complex decision,”▶ 43:26 Park notes. “In my hunch here is I do think in the next some number of years we will start creating simulations that will actually cost as much as training a foundation model but perhaps it's going to be so valuable to the society that it would be a no-brainer.”▶ 49:56

Park sees simulation and artificial intelligence as paired pillars of advanced civilization. If an agent system can test 10,000 policy variations before passing a law or launching a market mechanism, governments and enterprises can eliminate coordination failure before writing code or spending capital.

What to Do With This

Audit your next product or pricing change before running live A/B tests on users. Build a testbed of 50 agent personas generated from your last 20 customer discovery transcripts, supply each with their real budget constraints, and simulate their reaction to a 30% price hike. Compare the simulated churn rate against your historical cohort drop-off before pushing changes to your production billing system.