DeepMind’s AlphaZero Engineering Use Cases

The possibilities go from an electric motor, to a gasoline engine, to an entire car, to perhaps somewhere far beyond our own imaginations.

Using the parameters of how something is manufactured, AlphaZero should be able to optimize, to near perfection, almost any design. It could start simply: for example, AlphaZero could optimize the way the copper wire is spun around the central rotating rod of a motor. It could do this all virtually by simulating it in a computer program.

It would start out by randomly moving the arm that directs the wire as the copper is spun around the central rod. The speed of moving the directing rod, its direction, and the timing of any changes would all be randomized. Then AlphaZero would get a baseline performance for that initial random design by powering up the virtual motor with a simulator to determine its power and efficiency. From here it would repeat randomly many times, with many variations, as AlphaZero does so well. As it goes along, it incorporates the design changes that help it improve over time, automatically. I hope you can see how, in theory, this process could be scaled up to optimize any design conceivable by man, and perhaps in time devices not conceivable by man. Perhaps the hardest part of this process, since AlphaZero is already so powerful, fast, and efficient, is accurately and efficiently simulating the electric motor, with all of its electromagnetic physics interactions.

Imagine AlphaZero optimizing all the parts of a car, then being given all the parts for a car (with many variants) to teach itself how to build a car. We would give AlphaZero the constraints of certain features to optimize for such as the ability to move, power, reasonable cabin shape and size, and efficiency. AlphaZero may be able to improve the car, but using this method AlphaZero may come up with a concept that does not remotely resemble a modern car. Perhaps it would come up with a legged or a winged creation, or something that forces air around itself for propulsion. Who knows?

I am so excited for DeepMind to grow and for them to keep letting us know the many use cases that they dream up and implement for AlphaZero, as well as the use cases of any new AI engines that they might create.

DeepMind’s AlphaZero Engineering Use Cases was originally published in Hacker Noon on Medium, where people are continuing the conversation by highlighting and responding to this story.