Google, today, unveiled yet another neural network thing for you to play with. It gives you a word to draw, and as you draw it, it says aloud some nouns.. 'guessing' what it is you're drawing.. until it 'gets it right'.
It appears to be 'trained' over time by using the keywords and drawings done by visitors to the site.
Here is an example of drawings done by redguts for which the thing 'guessed' correctly 5 times..
After drawing with it for a little while, I remembered the last time an innocent-seeming interactive 'neural network' plaything was debuted online, it became subverted almost immediately and wondered if something like that would happen to this one, too.
Amusingly, the same clever actors responsible for bringing down Tay are 'training' this google thing in ways it wasn't intended to be trained. Below is a sampling of fun-times AI subversiveness:
then.. as one person put it.. they began to break the conditioning:
it seems as though this is already memeing into existence at the moment.. here are the last two threads that have hundreds more examples.