---

Professor Judy Franklin Asks, 'Can Humans and Computers Make Beautiful Music Together?'

Professor Judy Franklin plans to trade jazz riffs with ... her computer!

In an ordinary office on the fourth floor of Clapp Hall there is a computer--not noticeably unlike the computers in most other offices on campus--and a professor with a flute.

And, if Professor Judy Franklin is successful in her work this year, soon her computer will be able to do what few, if any, throughout the world can do. Her computer will learn to play, to improvise, jazz music with the flute-playing Franklin as a human collaborator.

This fall Franklin received a grant through the National Science Foundation's Professional Opportunities for Women in Research and Education Program to carry out her study, which Franklin feels may open a new avenue of theoretical inquiry in the field of machine learning.

"I've been in computer science a long time and I've gotten concerned about how the human species and computer species will manage to get along, or not," Franklin said. "I want to see some positive, almost interdependent relationships emerge so I can study them ... At some level music is communication, so machines and humans are entering a discourse that they enjoy, if it is possible for machines to enjoy."

To describe her work in oversimplified terms: in Franklin's study the computer will learn the rules of jazz composition and improvisation--rules that are in themselves complex because they are made to be broken. Franklin and the computer will then trade four-measure passages of music.

In this process, the machine will assess its performance through a number of mechanisms: through what Franklin plays in response to the machine's solos and, as the experiment progresses, through Franklin's facial reactions and posture, or body language, in response to what she hears from her mechanical collaborator. (The machine will eventually be able to see Franklin react through a video mechanism.) This kind of machine learning is called "reinforcement learning": if Franklin grooves, the machine is swinging; if she winces, the machine has flopped.

Franklin, in her second year at Mount Holyoke, already has substantial experience in this field. For the past nine years she worked at GTE's research labs in Waltham on machine learning as applied to manufacturing and telecommunications.

Now, however, Franklin is putting her mind to more theoretical and artistic endeavors. Playing music together is one of the most fundamental and enjoyable human activities. Franklin is using her work to begin to look at fundamental questions about computers: Can machines enjoy playing music? Can playing duets be a first step in forging a more harmonious collaboration between humans and machines?