From: gt0228b%prism@gatech.edu (FALCO,VINNIE)
Subject: Re: Implementing a virtual world
Date: 4 Jun 90 07:56:31 GMT
Organization: Georgia Institute of Technology



A proposal :

I believe that I have heard of a program called NetTalk, that takes in
regular english text (a 27 letter alphabet, plus punctation maybe)
and converts it into the correct phonetic form (approx. 55 sounds)
by utilizing a neural net that has been trained over and over again
by comparing the produced output with the desired output, and then
adjusting the weights.

I have also seen a crude device for the Apple II where electrodes are
hooked up to the scalp, and heart muscles (and whatever, I am not a
doctor :). There is some software that comes with it that puts up a
graph of the relevant data.

What are the possibilities of using a neural network, where the inputs
are the biodata from a user (i.e. alpha wave activity, heart rate, whatever)
and the output is some useful function, like dragging a file into the trash,
opening a window, running an application, or selecting a menu item (definite
Macintosh bias here...)?

The network could be trained over and over again, while the user thinks about
whatever (s)he wants to associate the desired operation with.

Of course, depending on how hard the user concentrates on the proper ideas,
the network would have a certain amount of 'slop', so this might be a
problem, because if your mind wandered, you might erase some file or
quit an application by accident!

eh?