According to an article in Discover, deep learning AI-powered systems are positioned to simulate real world phenomena, from weather forecasting to drug design.
In a white paper for the DC-based Computing Research Association, Professor Ian Foster and his team at the University of Chicago argue that AI-driven simulations are set to have a dramatic impact on the way we predict the future.
Could AI like this be a modern-day Oracle at Delphi?
The answer according to the team behind the paper seems to be yes.
- Per the team, “the new opportunity presented through AI driven simulators is to learn from data, to accelerate simulation through prediction, and to augment physics-based simulation with predictive models of social and economic phenomena.”That sounds like some Foundation-level Hari Seldon
Three areas where deep learning/AI driven simulations could have big impact, according to the white paper:
1. The Three-Body Problem.
According to the Encyclopedia Britannica (electronic version, folks, wish we had a hard copy set though), the Three-body problem in astronomy is “the problem of determining the motion of three celestial bodies moving under no influence other than that of their mutual gravitation. No general solution of this problem (or the more general problem involving more than three bodies) is possible.”
As a refresher, last year, Philip Breen at the University of Edinburgh and a few colleagues trained a neural network to calculate solutions to the problem up to 100 million times faster than state-of-the-art, conventional means. Per the article, AI learns how the motion of such bodies evolves rather than trying to calculate the motions by numerical brute force.
This research could be applied to other complex systems such as climate forecasts, earthquake aftershocks and traffic flow patterns, according to Professor Foster.
2. Systems determined by human behavior. Think economies, stock markets and crowds. Bottom-up models and irrational actions of individuals make predictions hard as we all know. But AI systems work on trying to learn overall outcomes.
- Foster rhetorically asks: “Analogous to the dramatic advances in modeling human language, can there be dramatic advances in modeling human behavior?”
3. Optimizing decision-making. Google’s Deep Mind researchers have had a lot of well-publicized success in beating humans at games such as Chess, Go and Star Craft. Foster posits this could be expanded to real world problems more and more. The Deep Mind team has already done this with the “protein folding problem.” You can check out The Scroll’s recent article on that here.
Link to the full paper: The Rise of AI-Driven Simulators: Building a New Crystal Ball: arxiv.org/abs/2012.06049