Melanie Mitchell       . . . is a renowned computer scientist. One of the world’s leading practitioners of and authority on genetic algorithms used in the study of complex systems, Dr. Mitchell is currently Associate Professor of Computer Science and Engineering at Oregon Health and Science University and and a member of the External Faculty of the Santa Fe Intstitute.. She and others scientists there are trying to evolve computer programs that will recognize images from satellites. Formerly a visiting faculty member of the Santa Fe Institute, Dr. Mitchell earned her PhD at the University of Michigan, working with intellectual giants John Holland and Douglas Hofstadter.

1:18  4:22

Evolving Solutions

What is "intelligence"? Right now, no one knows. But we will. The work is being done right now by people like computer scientist, Dr. Melanie Mitchell. She's convinced that when we understand pattern recognition, we will have cracked the problem. Understanding analogies will be a good start.

Scientists have learned the limitations of an "expert systems" approach to understanding complex systems, including intelligence. They're now exploring with new tools, including neural networks (inspired by how brains work) and genetic algorithms* (inspired by Darwinian evolution). Dr. Mitchell is recognized as being at the forefront of scientists applying genetic algorithms. She displays the excitement and the discomfort of a person going into unknown territory.

Since the very first days of the computer revolution when "computers" were women doing pen and paper calculations of missile trajectories at Los Alamos, we've learned that what's hard for computers is easy for people and vice versa. People can recognize a face or walk through a room without bumping into things. Impossible for a computer. Chess has a new Grand Master, but that doesn't mean the computer is intelligent. The difference? The computer uses a "brute force" solution: "Try all and every combination." The person uses analogy: "I've seen a similar situation before (I recognize the pattern), I should play X."

Science is full of Big Questions so new many are only now being formed. Take genetic algorithms alone. What can they do? Where are they useless? How does a scientist decide what parameters are appropriate in creating one? Genetic algorithms have already proven useful in aerodynamics and some drug design but not others. How will they drive technology? And what IS "intelligence"? Stay tuned.

[*Here's how genetic algorithms work: A genetic algorithm is a computer program, a piece of software inspired by biology, with which scientists try to evolve solutions to some problem. The computer has a bunch of different programs in its memory, devised and picked at random. Most of them don’t do anything. But occasionally, one of them will, by chance, do something semi-useful. That one program, because it’s better than the rest at doing the thing you want, gets to have children, to make copies of itself, with slight changes (mutations). Then the "child" tries to solve the same problem. The "child" that does the best gets to reproduce, gets to have more children, to make more copies of itself. As in Darwinian evolution, over time, the "children" get better and better at solving the problem. It doesn’t always happen, but it happens. The result is a computer program that no person wrote, doing the thing you want done.]

[This Program was recorded January 24, 2000, in Los Alamos, New Mexico, US.]

Conversation 1

Melanie Mitchell tells Paula Gordon and Bill Russell why one can reasonably locate the beginning of both the computer age and the nuclear age in Los Alamos, New Mexico. She reminds us that “computers” were originally women sitting at desks, doing pen and paper calculations of missile trajectories, then brings us into the present filled with ubiquitous computers.

 7:09

Conversation 2

Dr. Mitchell tells us what computers are doing for science these days, using weather as an example of the extremely complicated systems being modeled. She recalls how weather inspired computer pioneer John von Neumann, who had to go to Los Alamos to pursue his interest. Understanding biological systems was an early part of the computer revolution, she explains, bringing us current with a description of Artificial Intelligence and the new field of science called Artificial Life. Dr. Mitchell offers examples of how the hardest things for computers are the easiest things for people and vice versa. She describes computers as specialists, people as generalists, with implications. She tells us why she believes recognizing patterns is what general intelligence is all about.

 9:42

Conversation 3

Dr. Mitchell describes alternatives to approaching general intelligence from a “top-down”/expert systems perspective: neural networks (which attempt to mimic some aspects of the brain) and genetic algorithms (which use ideas from evolution to try to evolve programs that can recognize patterns.) She describes how a software engineer uses genetic algorithms. She offers her view of the science of complexity, posing fundamental questions, describing possible approaches, applying it to everyday activities. Dr. Mitchell describes the kind of computer systems she wants to develop. She explains why the ability to adapt is of central importance -- being able to change to do something differently when circumstances call for it is, she maintains, the hallmark of life. She describes how the science of complex adaptive systems attempts to find general underlying principals.

 11:23

Conversation 4

Dr. Mitchell poses the question of whether scientists can come up with a kind of system that will explain all complex systems, from life itself to culture to economics. She compares the answers a physicist and a biologist seek in fundamental research. How do brains work and how do humans recognize faces? she wonders. Dr. Mitchell describes how a scientist approaches vast fields she does not yet understand and the excitement that accompanies the discomfort. She defines “genetic algorithms,” with examples. She explains the critical nature of defining parameters, with examples from airplane wings to the stock market.

 11:57

Conversation 5

Describing the work she and her colleagues do as in some ways “more art than science,” Dr. Mitchell describes big research questions, including: On what problems will genetic algorithms work well? How does one sets various parameters? She describes various approaches psychology is taking to discerning how brains work. She describes her work, with colleagues, on “Copycat” and “Metacat,” attempts to give computers the ability to make analogies. She explains why she and others believe the ability to make analogies is central to human thought.

 11:33

Conversation 6

Dr. Mitchell compares the computational nature of genetic algorithms to the paintings of Georgia O’Keefe. Dr. Mitchell describes her wonderful experience in the Southwest and at the Santa Fe Institute. She describes the kinds of things she and fellow scientists are trying to understand, from biology and physics to social sciences and culture. She reiterates that if a scientist could “crack” pattern recognition, the problem of intelligence would be solved.

 3:42

Acknowledgements

Related Links:

An Introduction to Genetic Algorithms is available from MIT Press.
Dr. Mitchell’s website containts references and links to her work.