Wednesday, December 13, 2006

Amazing story illustrating a simple principle: if you know exactly what you have to do, you do not need to employ a neural network when automating the process. Just hard-wire the algorithm in.

And a couple of conclusions one can derive from the story:

1. (The notorious one). Some people are convinced that the more expensive the solution, the more reliable it is.

2. (The pessimistic one). May be the only reason we want to create Neural Networks and stuff is because we do not understand how things work, and relay on the unconscious to do the job. Which implies a lot of wasted resources because it is, more or less, a 100-monkeys-behind-the-typewriters approach plus the reward for the monkeys who managed to accidentally come up with the first sentence from "The War and Peace". Is it really the way to go?

No comments: