swift@reed.UUCP (Theodore Swift) (05/14/85)
> Not to get involved in a long discussion of AI and philosophy, but... > > Your assumption is that the computer would generate music by the machine > equivalent of a throw of dice. I think most people would agree that this > method is unlikely to produce great music. However, suppose we built a > machine with all the memories and music knowledge of Beethoven, and then > let this machine loose to create music by the same techniques of inspiration > and experimentation that human composers use. It might be able to create > a great piece of music AND tell you why. It's a wonderful and interesting idea, but how do you implement "experimentation", "inspiration", and "create", not to mention "let loose". I suspect the result of the first pass at it would give you something that would sound "like" Beethoven, but you'd have real problems getting it to write a Bach cantata or fugue, or a Mozart minuet. If you just put all the "knowledge" (how do we define that, in this situation??) into your data base, you'd probably confuse the hell out of the program, as well as getting some really strange output. The "memories" and "musical knowledge" of a composer would be dead to the program, like looking through someone else's family album. You'd have to consider other things in designing the programs, like style, the musical fashions of the composers time, performance technique; all sorts of things. One really cool result of a program to do this successfully would be to "get the computer to" write a fantasia, or medley, slowly changing the style of the composition from one period of history to another. You could start with a historical (western) piece which begins with, say, a gregorian chant, slowly blending into multiple-voice vocal work, then bring in some lutes, then pipe organs, harpsichords, violins.... It could be very interesting, and it could be really bad. To do it well, though, would take some real skill.