Machine Learning, Physics, and Order - Understanding Generative AI

The subject of the last Meetup was applying Machine Learning to mtDNA, and in this event, I'll show you how to apply Machine Learning to a wider set of problems in the sciences, including physics, and reconstructing order, when no time information is available.

Though the subject is very different from mtDNA, the ideas are relatively similar, in that Machine Learning is much more than prediction, and is instead a fundamental tool of the scientific method.

This will allow us to understand how people are creating "generative A.I." videos and images, but we'll be taking a rigorous, mathematical approach, to what is a very serious topic, rather than trying to generate memes.

In fact, we'll prove mathematically that a Turing Machine (i.e., a computer), cannot generate information, and can only manipulate a constant amount of information, or destroy information, forcing the question of whether generative A.I. is by definition infringement.

I'll also discuss actual Artificial Intelligence, specifically, optimization, and how this differs from Machine Learning. I'll introduce a generalized optimization algorithm I developed that can balance weights, sort a list, or interpolate a polynomial, all without any specialization, suggesting the possibility of universally intelligent machines.

Slides will be coming soon, for now see Section 2.4 of my paper, "Vectorized Machine Learning" for the core physics ML algorithms.