[ml] Systems Theory

[Kai Chang]

This thread is to gauge interest in a talk/discussion on systems theory (in
its broadest sense) in relation to machine learning. Please respond if
you're interested, with topics, theories you would be interested in
discussing.

Modern machine learning is based on a set of algorithms that have proven
very successful for solving broad classes of problems based on empirical
data. ML grew out of artificial intelligence. AI takes much of its
inspiration from human intelligence, however attempts to replicate HI have
experienced many frustrations and fundamental setbacks. There is currently
not a theoretical framework to study general intelligence (strong AI,
self-awareness, creativity, etc). In biology and neuroscience, there are few
widely accepted theories of how the material substrate of the brain
generates conscious experience or solves many complex intellectual tasks.

Systems theory attempts to provide a framework within which generally
intelligent beings and other sources of complexity can be reasoned about. To
use an ML example, consider financial markets. There are trading algorithms
that capitalize on market inefficiencies on short timescales. The existence
of the algorithms is known to market participants, and algorithm-makers can
create new algorithms based on their experience with existing algorithms.
Firms, governments and people can adjust their behavior based on
observations of the algorithms' effect on the market. In this way, the
system is highly
reflexive<http://en.wikipedia.org/wiki/Reflexivity_(social_theory)>
(economic
systems are already, algorithmic trading adds another layer of reflexivity).
For instance, the May 2010 Flash Crash brought widespread media attention.
Media attention beings new dynamics and social forces into play. The flash
crash itself could be an example of weak emergence, where the combined
properties of individual algorithms create systemic instability.

The above situation can be considered a complex adaptive
system<http://en.wikipedia.org/wiki/Complex_adaptive_systems>
.

More broadly, I'm interested in systems theory because it creates a
framework of discussion where expert knowledge from disparate fields,
philosophical, linguistic and intuitive concepts can be considered as
systems with relationships to each other. Everyone has a unique perspective
on the topic, insofar as we are all generally intelligent beings composed of
internally complex systems, interacting with external complexity, and
communicating with other generally intelligent beings.

Hopefully we would generate inspiration for new machine learning projects,
and gain a broader appreciation for machine learning's impact and potential
on systems familiar to us.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.noisebridge.net/pipermail/ml/attachments/20110719/9ad936de/attachment-0002.html>