Modeling the Planet with Millions of Events and Causal AI

Todd Moses
Nov. 19, 2023
10 min read

In 1958, McDonnell Douglas created the F-4 Phantom II for the U.S. Navy. A multirole military aircraft that later proved popular for the Air Force, Marines, and 15 other nations. This aircraft is incredible because it can fly from 150 to 1,500 knots with an angle of attack of up to 30 degrees. All without any flight computer, something modern aircraft require to perform similar maneuvers.


In 2022, Austin Meyer simulated the aircraft in his company's X-Plane software. Meyer tested his work with an actual Israeli Air Force F-4 Phantom flight instructor. The complexities of the F-4 made it perfect for ensuring the X-Plane flight model was correct. Any flaws in the simulated aircraft meant a more serious flaw in the flight simulation software.


The first attempt by Meyer resulted in an aircraft limited to an angle of attack of 20 degrees, the same as most modern airplanes [Meyer 2022]. However, the actual F-4 is much more dangerous in its performance envelope. To simulate the F-4, changes had to be made to the underlying flight simulation software. The reason for this is an error in classical wing theory. Meyer discovered that the F-4 wings function differently than modern aircraft. In a typical airplane, vortices on the leading edge cause a stall, but on the F-4, the air rises over the wing and improves performance.


Physical Simulation


Contemporary science involves breaking systems into their underlying parts and then analyzing them as much as possible to understand them [Wolfram 2019]. This is how computer simulations like X-plane and many video games are created. The issue is that it can be impossible to understand how the system operates by analyzing just its parts. Something Meyer discovered while adding the F-4 to X-Plane.


Mathematical simulation is usually limited to those systems that can be described using arithmetic or geometry, such as classical wing theory or ballistics in first-person shooting games. However, more complex systems, such as those where human intervention occurs, cannot be modeled with math. Probabilities can tell us what a person is most likely to do, but they cannot describe what a specific person will do in a particular situation.


Physicist and founder of Mathematica software, Stephen Wolfram, discovered the Principle of Computational Equivalence to deal with simulating phenomena more complex than standard math can describe. He declares, "Whenever one sees behavior that is not obviously simple - in essentially any system - it can be thought of as corresponding to a computation of equivalent sophistication." Physicists are searching for a theory that summarizes the whole universe. Something that cannot be done with traditional approaches [Wolfram 2019].


The Rise of AI


Geoff Hinton created a company that had no plan to make a product. At 64 years old, Hinton, an academic by nature, had soon gotten the attention of many large companies for his infant company. The first valuation was $12 million. The reason for such excitement around a company with nothing was that Hinton and his students, now co-founders, had changed how machines saw the world [Metz 2021].


This was the father of deep learning, taking his research to the commercial market. In 2012, Hinton and his co-founders proved that a neural network could recognize everyday objects more accurately than any other technology. Fellow A.I. researcher Kai Yu, a Chinese employee of Baidu who grew up under communism, recently moved to Silicon Valley. Explaining the new technology to his bosses in China resulted in the first offer to buy the new company.


Alpha Go made deep learning famous by beating human masters in the game of Go. It is an ancient game considered more complicated than chess. They did this by training the network on two pixels, one white and the other black. The machine then played millions of games to determine the best strategies. A method very similar to those Wolfram uses in his computer experiments of complex systems.


Predicting Volatility


Unlike modeling the mathematics of a wing through air, the process of simulating volatility is much more complicated. For example, the first recorded time global markets moved in sync was the 1987 stock market crash [Rhoads 2020]. Unlike previous financial calamities, there was no single explanation. It occurred on a Monday morning in New York and echoed worldwide markets as the day progressed.


Another volatility event that is impossible to realize was the September 11, 2001 Terrorist Attack on the World Trade Center. This event crippled trading in the U.S. for that day, and the ones to follow as the target were the financial offices of many large asset management companies. Mathematical simulations are not enough to forecast a coming problem for this, and so many like it.


Cause and Effect


Turing Award winner Judea Pearl has a different approach to simulating complexity. His Causal Inference is a set of mathematical tools for dealing with cause and effect. By knowing the cause of an event, one can change that cause in simulation to better understand how such an event can form and what factors contribute most to its impact.


Unlike the deep learning of Hinton, which dominates machine learning, Causal AI goes beyond correlation and curve fitting. Causal A.I. aims to answer the question of why in a manner similar to how humans do. Pearl describes, "While probabilities encode our beliefs about a static world, causality tells us whether and how probabilities change when the world changes."


Millions of Events


The world as we know it is a series of events. These events represent a model of reality as groupings of cause and effect. A perfect data source for Causal A.I. These events are not necessarily significant, as many never make the news. Instead, according to Chaos theory, it is the tiny events that alter the world. For example, hurricanes come from the heating of ocean water by a few degrees.


Perhaps most important is that each event has a cause and an effect, like a line of dominos falling. Unlike the domino effect, such a chain of events usually gets more significant as it continues. Together, this constant series of events makes up our reality. The times of peace, war, drought, and floods are all controlled by an ever-increasing line of events. Understanding this cause and effect is critical for knowing what will happen due to what is occurring now.


Estimand uses these events to model the globe in real time. Our Causal A.I. takes in over 500 million real-time global events, categorizes them, and trains itself on them. The result is a system that shows where you are now and where you will be, including the ability to change actual events to observe the future impacts of such actions.




- Metz, C. (2021) Genius Makers - The Mavericks Who Brought AI To Google, Facebook, and The World. Dutton

- Meyer, A. (2022) Modeling the F-4 Phantom – a technical treatise. Key.Aero.

- Pearl, J. (2018) The Book of Why: The New Science of Cause and Effect. Basic Books

- Rhoads, R. (2020) The VIX Traders Handbook. Harriman House

- Wolfram, S. (2019) A New Kind of Science. Wolfram Media, Inc.


How to Predict a Hurricane
Predicting Financial Crisis with Causal AI