[aipy] statistics vs logic

[Rick Thomas]

Some Python, please ...

Here's two projects to compare: one is a basis for a statistical
approach to AI and the other a logic approach. Both are applied
projects coming from the life sciences. I also note that neither uses
the term "AI".


pebl: Python Environment For Bayesian Learning
http://code.google.com/p/pebl-project/
This is coming from a dynamic field (gene protein expression), looks
like a modern Python project (Sphinx), has some cool features (grid
computing), and seems like it might be a good basis for an AI
platform.

Contrast that with python-dlp ... framework for DLP-based reasoning
over RDF/N3 content
http://code.google.com/p/python-dlp/
This is an impressive implementation of logic, a much more complex
topic (to me) than bayesian reasoning. Applied to clinical
cardiology.


I contrast those pure approaches, statistics vs logic, to raise the
topic of hybrids.

In the spirit of controversy here's some comments about Google's
recent advocacy of statistics over logic.
Unreasonable Hypocrisy
http://www.betaversion.org/~stefano/linotype/news/275/
Short version: Google uses a hybrid so they are being evil-ish when
they argue against logic.


This paper gives a tutorial on a hybrid approach. It's accessible and
fun.
Multi-entity bayesian networks without multi-tears
http://ite.gmu.edu/~klaskey/papers/Costa_Laskey_MEBN_wo_Tears.pdf
(Imagine an intelligence that keeps Starship Noah in top operating
condition.)

The exercise coach could use both techniques. So we might consider how
the two projects could be used together?





[Off-topic] Python humor:
http://oui.com.br/blog/nando-en/post/python-can-cater-to-java


Rick