Google’s Working on Dueling AI That Can Teach Each Other Without Humans Needed
SKYNET. FREAKING SKYNET.
Sorry, every time I see one of these articles, images of the Terminator run through my head as I’m sure they do a lot of yours. I digress, though.
Regardless of whether the push into AI is one day going to kill us all, it is an incredible technology. The basic concept that is getting a lot of attention right now, is machine learning. In a nutshell, this is the process at by which a human can teach a program to do something through example, instead of coding hard if/then statements that the machine then has to abide by.
For example, machine learning with self-driving cars allows someone to essentially turn it on and drive the car over and over and over and as they drive the machine learns what to do and what not to do. The more info it has to work with, the better it gets.
The “issue” here is that there is still the need for a human to teach the robot, but what if a robot could teach the robot?
Ian Goodfellow, who works at Google Brain (Google’s AI department) has been working on this very thing. The idea behind his tests is that if you teach one robot to nitpick the other and then tell it what it did wrong, the first robot can then learn to improve and the loop continues.
He gives the example of an AI that paints and one that judges paintings as fake or real. The one paints, the other tells it why it’s fake, and it uses the data to try again. And again. And… you get it.
Eventually, the painting robot gets to the point where he can paint copies that the second can’t distinguish. Clever, creepy? Yes to both.
Either way, it’s still super interesting to see how fast this machine learning (and these new techniques of it) are evolving at a rapid rate. Besides the inevitable Skynet concept that movies have put in our head, there is a tremendous amount of good these programs could achieve at a much faster rate than we ever could. Let’s just never teach them how to be self-aware, cool?
For the entire story, check out the source link below. Thoughts on these things, guys? And does anyone want me to do something explaining any of this in more detail (i.e. What is Machine Learning?)? If you liked this article, please share it it’s greatly appreciate it! Thanks!
