About Me

Why I am a control theorist

lftuft 

The age of AI. We live in the formative age of automation and artificial intelligence. Like a growing root system, control algorithms already appear, and will continue to appear more physically and more pervasively in all underlying aspects of our society. What these systems do, the choices we allow them to make, and how we design them today will set the stage for humanity's relationship with automation for years to come.

But increasingly, many of our systems are so sophisticated that engineers have hit a complexity roadblock: nobody can keep track of it all. For example, I am deeply worried about unsafe AI, created without formal guarantees about its effects and interactions with the world. Hidden oversights, engineering mistakes, wrong abstractions, and plain old complexity can cause destruction and unforeseen consequences — having written many types of automation software, I know how easy even simple control algorithms are to fool, and how innocent bugs can wreak havoc.

We can do better. I believe we can engineer provably safe, reliable, resilient, and useful automation software. We should not need to worry about the systems we create, because we should know everything about the systems we create. I dream of a world where control and AI algorithms are automatically generated and correct by construction, because unlike humans, computers do not make mistakes.

But even if this is just a dream, we can get past the complexity barrier if we implement good engineering philosophies from the start. I believe that control theory, modeling, and formal methods, properly and broadly applied, are the best ideas we have to provide the assurance that autonomous systems — including physical ones like robots, self-driving cars, and airplanes, as well as virtual ones like operations research, scheduling, financial, and logistics systems — function as intended. These ideas are not simply nice to have in an academic context divorced from application. Their development is essential for continued safe technological advancement in the modern world.

My role. Because I believe in the good that is possible when we slow down, be careful, and do it right from the start, I spend my days advancing the state of the art in theory, tools, and education related to safe and reliable automation.

Bio

Currently, I am a postdoctoral fellow at the Institute for Computational Engineering and Sciences at the University of Texas at Austin, where I explore and invent new formal methods for autonomy and verification of machine learning systems. I completed my PhD in Control and Dynamical Systems at the California Institute of Technology (Caltech), funded by a National Defense Science and Engineering Graduate Fellowship (NDSEG) and by Boeing. My PhD research, supervised by Richard Murray in the Networked Control Systems Lab, was on robustness, adaptation, and learning in optimal control — in particular, I applied convex optimization methods and formal verification techniques to aerospace systems. I received my BS and MS degrees, both in Electrical Engineering, from Stanford University in 2011.