[Future Wave] From AI Future Prediction to Experimenting with the Future

"MiroFish" AI Agents Gain Widespread Popularity
Beyond Predictions Based on Past Data
A Revolutionary Shift in Policy Design Paradigm

Recently, an artificial intelligence (AI) system developed by a Chinese university student in just 10 days has garnered global attention. This project, called "MiroFish," topped the popularity rankings on GitHub immediately after its release and even attracted investment worth billions of won. On the surface, it seems like another fascinating success story born in the era of generative AI. However, as someone who studies the future, I believe this event carries significance that goes beyond mere technical achievement.

Future prediction has long been understood, in part, as a "data problem." The common approach is to collect data from the past and present, identify patterns within that data, and then estimate what the future will hold. While machine learning has made this approach more sophisticated, the fundamental premise—envisioning the future as an extension of the present—remains largely unchanged.

Data inherently comes with a fundamental limitation: it is a product of current environmental structures. The data we possess is essentially a trace of a world that has already passed. If the future were to maintain the same structure as the present, data-driven predictions could be extremely powerful tools. In reality, however, this is not the case. The structures of technology, economy, society, and politics are constantly evolving and, at times, undergo dramatic transformations.

In such circumstances, excessive reliance on data can actually pose the risk of excluding a wide range of future possibilities. While data excels at capturing stable patterns, it often fails to detect new pathways and nonlinear changes when the situation and structure shift.

There is yet another critical layer of uncertainty. The most unpredictable variable in the future is people themselves. Human perception and reaction, individual choices, and collective behavior continually distort and reshape predictions. Furthermore, the very act of prediction changes the future; when a prediction is made public and people start to believe and act on it, those actions themselves alter what the future becomes.

This is precisely where the approach demonstrated by MiroFish is noteworthy. The system does not stop at data analysis; it creates thousands to hundreds of thousands of AI agents and forms a virtual society. Each agent interacts based on different dispositions, memories, and interests, and through these interactions, public opinion and collective behavior emerge. This is not a prediction of a single, definitive outcome, but rather an experiment with a wide range of possibilities.

This approach offers new possibilities for policy design in a conflict-ridden society like Korea. Until now, policies have often revealed their effects and side effects only after implementation. However, by utilizing multi-agent-based simulations, it becomes possible to test diverse social responses before introducing a policy. Policymakers can preemptively explore whether a specific policy will spark conflict between groups, how such conflict might spread, and under what conditions it may be alleviated. This can reduce the cost of policy failure and provide a foundation for managing social conflict more precisely.

We are now entering an "era of experimenting with the future." The future is a space filled with diverse possibilities, and we are beginning to acquire tools to explore them. What matters is not the accuracy of prediction, but how many possibilities we can explore within ever-changing structures, and how we can make better choices as a result.

Professor Yongseok Seo, Graduate School of Future Strategy, KAIST

This content was produced with the assistance of AI translation services.