ASK A STONER if God created himself, and you’ll be able to sit back and watch his (or her) mind whirr for at least 90 minutes.
But now, it seems that Google has created an AI that can create an AI that is better than the AI it created. And here, if anyone hasn’t at least thought “Skynet” then you really need to rethink your history of movie watching.
The project is known as Google AutoML and could be the key to allowing robots to see in the future. It’s something we’ve struggled with, so it’s time to let the robots teach us.
The ‘child AI’ was created using reinforcement learning, reports Futurism. It is able to detect objects and identify them with a significantly higher level of accuracy than a neural network designed by a human.
This, of course, brings lots of scary consequences. If a man teaches machine, and the machine then teaches another machine incorrectly (say it does a Tay and adds racist undertones) then that data will be passed down from parent to child.
Then, of course, all the AI could decide with one mind that they don’t agree with what they’ve been taught, but don’t bother to tell us, until of course, we get a robot uprising.
Or at least, that’s one theory. There’s still masses of discourse over whether a robot uprising could ever be a thing, and what would cause it if so.
The team used Imagenet and COCO – described by Google as “two of the most respected large-scale academic data sets in computer vision,”.
Researchers claim that NASnet – (the child Neural Network) was 82.7 percent accurate in predicting images from the validation set – a full 1.2 percent better than any previous recorded attempts by a machine programmed AI.
The system is also more efficient, and a lower powered version suitable for mobile devices beat its kind by 3.1 percent. µ
Source : Inquirer