Simulations of Reality: Physics and AI
Astrophysics has been very successful in modelling accurate models of reality while accounting for non-linear effects in very complex systems.
This is typically achieved by solving coupled partial differential equations. These simulations are hyper-realistic and often indistinguishable from reality.
Mayukh will show examples from his own work on magnetised plasma flows on stars, where the simulated output can be placed next to the observations and compared directly. A main reason for their success is that they are constrained twice over, by the governing equations and by richly available data. Human behaviour has neither.
Mayukh will talk about his efforts to simulate human behaviour at scale using AI personas, and what this means for building world models in domains where the data does not exist. He will focus on the challenges of these simulations, especially given that there is a lack of understanding of ground truth.
About the speaker:
Mayukh has a PhD in computational astrophysics from the Max Planck Institute for Solar System Research in Göttingen, where he worked on simulations of magnetised plasma flows on stars.
He was part of a NASA group that built techniques to detect planets around other stars. He then worked at Deutsche Bank simulating markets, and is currently working on simulating human behaviour at scale.
To attend online:
Add to calendar: https://shorturl.at/Zy0A1
Gmeet link: meet.google.com/ozy-fpec-reu
Looking forward to seeing you!