Mathematical Modelers Understanding and Using Artificial Intelligence
0Presented by:
Frank Wattenberg, Emeritus Professor of Mathematics, US Military Academy, West Point NY USA
https://qubeshub.org/community/groups/simiode/expo
Abstract: ChatGPT is the most recent and currently most in the news flavor of artificial intelligence. In this talk we will distinguish GPT (generative pre-trained transformer) and LLM (large language models) AI from algorithmic AI and from other forms of AI based on extremely large data sets. This talk focuses equally on two things: How we as modelers, mathematicians and mathematics educators at the ODE level can understand how GPT works and how we can use GPT to do and to teach mathematics and modeling.
The most visible aspect of ChatGPT is an illusion — talking with ChatCPT feels like talking with a flesh-and-blood human being. The extent to which this is an illusion depends as much on the nature of flesh-and-blood intelligence as on the nature of GPT intelligence. We are learning more about both. The work of Daniel Kahneman and his book Thinking, Fast and Slow is particularly helpful.
Like all scientific advances, GPT builds on other work. One important example is "word vectors." The Ars Technica article "A jargon-free explanation of how AI large language models work" is particularly helpful. Further, the recent article Forbes article "What is the best way to control today's AI?" is a wonderful article. As modelers, we know that how we represent the features of the real world in our models is crucial and representing words as vectors is particularly brilliant. We are all here today because of our passion for modeling and as educators at the ODE level. But, ODEs are only one modeling paradigm and this talk will argue that for most students the traditional ODE class should be replaced by linear algebra and more broadly inclusive modeling.
The purpose of modeling is building understanding of the world in which we live and how we can change our world for the better. GPT builds in part on extremely large scale data analysis and like its predecessors and in stark contrast to algorithmic AI provides little or no understanding. Cathy O'Neil's popular book Weapons of Math Destruction is particularly helpful.
In short, this talk has three goals: Increasing our understanding of GPT, helping us use GPT effectively, and changing our mathematics courses at the undergraduate ODE level — both for math majors and for other majors. Words are just one form of modeling and writing-across-the-curriculum needs a new partner modeling-across-the-curriculum.
This is entirely optional but if you happen to have a pair of red-cyan 3D glasses please have them handy. These glasses are often packaged with 3D books.