What if We Could Simulate Everything
Figuring out how to perfectly simulate large systems is the solution to all problems. We could simulate everything on a computer rather than setting up an experiment for it.
I focus on longevity because I believe it’s the most important problem to solve. Hypothetically if we learned to simulate large systems, all longevity problems become trivial. Longevity is a part of biology. Biology is a part of chemistry. And chemistry is a part of physics. This means if we make progress in physics, it has a knock-on effect for the rest. We could simulate a human body and experiment virtually with what drugs cause what effects. Eventually, we would find the miracle drug that stops aging.
The people currently doing longevity research are amazing. And the real longevity breakthroughs will probably comes from here. But what if we did entertain the simulation idea?
How would we go about solving the simulation problem? I have a few ideas. And I think the point is not to take the first-principles breakdown too seriously, but to use it for inspiration.
One idea is looking at deep learning applied to fluid simulations. Basically, fluid simulations are accelerated using reduced-order models, while respecting the convergence constraints provided by higher quality models. You can get a realistic simulation even while dropping most information. A good way to think of it is that if an RNN predicts the next word in a sentence by looking at the previous words, then a physics RNN could do the same thing and predict the simulation’s next frame. Getting training data is easy because you can just generate it with higher-quality models.
Another approach I’m looking at is quantum computing applied to simulations.
Necessary for all this exploration is self-learning the math required to really understand the physics. I started The Feynman Mafia to help me with that.
I think it’s fun for me to contribute by exploring this alternative approach to longevity using simulations.