Neural Nets In The News, Again

I sneezed and split my lip open this morning. Even the application of Burt's Bees lip balm can't keep my lips moist enough. I've resorted to using skin lotion, mainly because I've gotten really tired of looking at my scaly lizard skin. The thumb cuticles are constantly threatening to crack open.

If this is what Mars is like? (And it is). With the cold and extremely dry air? You can fucking have Mars. As it is, I don't know about you, but I'm definitely done with Winter.

Speaking of which, are we done yet? Stephen Hawking, Bill Gates, Elon Musk are all got an existential fear over AI. As I've said before, BFD, but.. maybe they are on to something. AI is in the news. Neural nets. Or, as they are called now, Deep Learning.

I actually played around with neural nets back in the late 80s. Got a book from John Wiley on, learned all about Hebbian networks with feedback and forward propagation. Even did a art piece called "The Chinese Room" that used a lot of neural net diagrams in it (and the I Ching, and Red Army propaganda, and blast it if I didn't end up understanding Chinese in the end). (No I didn't).

Not only did I play with it, I wrote a program I called an artificial retina to process data for a company. And the problem went like this: We had contracts that normally, on an old mainframe, would be batch processed overnight into an active status, all linked up to the rest of the database and inventoried and documented and such. The company would have preferred to have contracts processed as they came through the daily paperwork, but for managerial and practical reasons, the batch program could be run during the day. A trigger program was tried, but it got into other problems, as some of these contracts contained thousands of inventory items, processing of which would slow the daytime online system to a crawl. (Bad architecture and design? Yes).

So, I wrote a little neural net program that would 1) look for changes in the contract system, and 2) learned what was and was not an acceptable volume of inventory items that big huge contracts, and 3) learned what was and was not considered high volume traffic that would tax our system.

That's all like baby stuff now. In fact, that program is, along with the Old Iron it ran on, fallen into the data Ginnungagap to which all things eventually get dumped in, existing perhaps as stray static charges on anchor chains and angle iron in China.

But like I say, retarded dwarf baby stuff, compared to what they are doing now.