I've been enjoying people's discussion of this. I ponder a bit myself below.
i'm not sure the word "improvise" can really be
> applied to a computer... this is where the toplap "human" approach is
> much different - it sure as heck makes a difference that we're
> "outputting" live...
Cognitive models of improvisation can be explored on computer in a live
setting as participants. This gives some very interesting possibilities.
Because the models are bound to be very complicated, there would be no hope
of programming them live, preparation beforehand (including possible
supervised training) is necessary. They may also contain unsupervised
learning capabilities for the performance, and I would hope interact with
human musicians in a sensitive way.
Their liveness makes sense in terms of their models of human cognition of
music and the fact they interact with humans. The devil's advocate would
run them against each other in a band.
The future TOPLAPPER might devise an AI which live codes, but I fear
fundamental issues of consciousness and human intelligence might delay this
work...as they delay models of traditional music making too of course.
Alex's point that human programming effort goes into their construction is
of course valid, and the same point is made I think by Doug Hofstader about
computer chess programs- it's amazing we can theorise and construct about
such things, there is no shame or threat in their autonomy.
> (nick please correct me if i've misunderstood either the interests of
> the group or how the systems work.)
I don't know the group properly yet, but I'll try to attend their first
meeting in December.
The more cognitive science I discover, the more inspired I am. I don't know
how deeply the members of LAN feel this; perhaps some feel it deeply as an
artistic field for exploration, others just enter trendy areas like
connectionism to grab research results compatible with existing biases in
the academic field.
I am torn because I like live coding as an interfacing method for
electronic music improvisation; but I am also developing automated systems
with perceptually motivated signal processing. I get round this right now
by controlling the parameters and the uses of the output of the complex
algorithms (those which have to be done in advance) through live code.
Either approach definitely involves humans, and I still feel algorithmic
composition is human. The question is just how effectively a model
understands human perception and production of music.
Received on Wed Nov 10 2004 - 15:05:02 GMT