Zero Mind

DeepDream is still the greatest thing to come out of artificial intelligence neural nets.

Today is a twofer. Not only has it occured to us that AI needs ethics training, but it turns out that in order to do some things, it needs no training from us at all.

Alphabet's DeepMind forms ethics unit for artificial intelligence
Oct 2017, phys.org

First of all, I don't know about you, but I need Human Subjects Research (HSR) training before I can conduct experiments involving humans. Regardless, the AI region of the Google empire now has a way to question and guide the ethical implications of its human-like thinking machines. (Wait a minute, doesn't that name belong to IBM?)

This is probably a good thing, since they already control the stock market (high frequency trading), and your social life (no explanation necessary). I just hope they have a diverse staff there on the ethics panel, because well, does anyone remember the Google gorilla fail?

How about the one where hundreds of bots were released on twitter in a competition to see who could make the most convincing human-analog, but then some of the algorithms were so good that some people started to flirt with them, and the creators really had to ask themselves when or how they should break it to the poor souls? Those were the early days of experimenting on people via the digital world (2011).

In fact, the head of those experiments, Tim Hwang was very recently named director of the Ethics and Governance of AI Fund, which does research on ethics and AI.

Having covered that, it comes out simultaneously that Google's same DeepMind (the one that did DeepDream and AlphaGo) has now taken unsupervised learning to the final frontier. Instead of teaching the program what to do or even how to learn, they let the thing figure out for itself how to play the game of Go, and it still beats the human. Done. They call it AlphaGo Zero, because it starts with nothing.

Google DeepMind: AI becomes more alien
Oct 2017, BBC


Post Script
Social Bots
Network Address, 2012