I sometimes feel like some kind of cave man programmer. Frozen in ice sometime after the 6502 assembly era, thawed out in the post-OO LAMP age. There's lots of new stuff. Some of it good. Why am I so damned cranky?
Some aspects of the modern world delight. I discovered Applied Cryptography with glee; like a box containing a lighter, sharp knife, flashlight, mirror, binoculars, and a compass, the usefulness of the tools in that book immediately leapt out at me. Far beyond the security domain, knowing how to do protocol analysis, use MD5/SHA, decent RNGs, salts, Diffie-Hellman, stream ciphers and the like seem like essential tools. Does everyone learn this stuff as core CS in school now? I certainly hope so.
I search for an analogous "Applied AI" to no avail. Some algorithms seem promising, but instead of sharp knives and binoculars there are only plastic toys. Useless Bayesian 85% A/B classifiers that require tons of training data, only good for writing papers, but not actual code.
Entire chattering research volumes of nonsense, tautologically proving nothing very interesting, because if the books knew how to do what their titles suggested, we'd all be a lot further along with this stuff.
The damned book I want hasn't been written yet. I should have stayed frozen longer.
Comments (4)
So write it yourself!
Posted by Paul Montgomery | January 17, 2007 4:25 AM
Posted on January 17, 2007 04:25
:-)
Posted by Rich Skrenta | January 17, 2007 5:54 AM
Posted on January 17, 2007 05:54
"Applied AI" would be nice but, yes, most of the textbooks and research work in AI seems to focus on toy problems, not the big data problems we love.
There are a couple notable exceptions that I know about. First, I thought "Foundations of Statistical Natural Language Processing" did a remarkable job talking about big data for NLP. While NLP is only one part of AI, it still may be of interest. I have some thoughts on the book here:
http://glinden.blogspot.com/2006/06/foundations-of-statistical-nlp.html
Second, if you consider search to be AI, then the field is wide open with books and papers talking about search and big data. My favorite has been "Managing Gigabytes", but I suppose that book may be getting dated by now.
Finally, many data mining texts focus on statistics and building models over big data. Again, a subfield of AI, but it may be of interest.
By the way, a big problem here may be that researchers at university often do not have access to the big data sets necessary to do this kind of work. I have lamented that in some of my previous posts on my weblog:
http://glinden.blogspot.com/2006/01/recommender-systems-and-toy-problems.html
http://glinden.blogspot.com/2006/08/chance-to-play-with-big-data.html
Posted by Greg Linden | January 17, 2007 9:11 AM
Posted on January 17, 2007 09:11
Machine learning is just a part of AI, checkout the Weka toolkit from the University of Waikato (http://www.cs.waikato.ac.nz/ml/weka/). Implementations of tons of ML algos and a neat little environment to prep your data & run experiments.
Sadly it's all written in Java, not PERL, but I'm sure you'll survive.
Posted by Craig Pfeifer | January 17, 2007 8:25 PM
Posted on January 17, 2007 20:25