Information Technology Reference
In-Depth Information
One concomitant implication for a Live Algorithm would be an embodiment that
involves a means to play a physical instrument by movement rather than a synthetic
production of sound by electronics. The task of learning to play a physical device
would involve the development of a better potential to listen to the type of sounds
the instrument could make. Hence sensor and actuator components would develop
together in a feedback loop. Ultimately the expectation would be that the Live Algo-
rithm would have a greater ability to make and hear complex sounds, an important
aspect of human improvisation. Robots move in real space, and have some goal,
even if only not to fall over. The analogy for our purposes would be movement in a
sonic field. It remains to be seen what goals would be pertinent; perhaps to navigate
between two timbres, or to find an interpolative rhythm.
6.6.2 Learning
It is without doubt a feature of improvisation that practitioners improve with time. It
would be unreasonable to deny our Live Algorithm the chance to reflect on its own
performance and find ways to improve. Consequently, the algorithm must be able
to make mistakes. The definition of what a mistake might be for a Live Algorithm
raises many fundamental issues.
There are many machine learning techniques that can be imported into the field,
but they would all require the existence of some kind of performance metric, as
discussed in Sect. 6.3.4 . Some kind of objective evaluation of an improvised per-
formance is needed. Such a measure could be developed by the analysis of human
group performance. Unfortunately results in this area are lacking. An information-
theoretic approach might be fruitful: organisation (and dis-organisation) can be
computed over various timescales using entropy and complexity measures. The
analysis would have to be then checked against human evaluation.
Ultimately we would require that the Live Algorithm becomes its own critic;
should the algorithm feel shame as well as satisfaction?
6.6.3 Anticipated Criticisms
In human-machine dialogue, the human input to the machine is already a source of
considerable organisation and information. Many algorithm designers exploit this
information either intentionally or by accident. The algorithm ultimately feeds on
the inherent musical organisation of the input stream.
In order to guard against this, tests could be set up involving groups of Live
Algorithms (i.e. without human performers). If such a group could spontaneously
generate new structures there would be more confidence in the ability of the algo-
rithm to create its own patterns within the context of a machine-human dialogue.
Interestingly, Miranda ( 2008 ) has demonstrated the emergence of songs from in-
teracting robots; giving Live Algorithms the chance to interact with other artificial
musicians might provoke growth in unexpected directions.
Search WWH ::




Custom Search