Google’s artificial intelligence has ‘learned to replicate itself’
Artificial intelligences “teaching themselves” are the stuff of science fiction nightmares – like when Skynet begins “learning geometrically” in Terminator 2.
But Google is already attempting a rather similar feat: teaching machine-learning software to write more machine-learning software.
The AutoML project is already coming up with software which is better than the best human researchers can create.
In a test where machine-made systems had to mark the location of objects in an image, the machine-made system scored 43 per cent and the best human-made alternative 39 per cent, Wired reports.
But the end goal isn’t for computers to take over the Earth (obviously): it’s to make it cheaper to produce machine-learning software, which could have an impact in spheres such as healthcare as well as tech.
Google CEO Sundar Pichai talked about the software last week.
“Today these are handcrafted by machine learning scientists and literally only a few thousands of scientists around the world can do this,” he said.
“We want to enable hundreds of thousands of developers to be able to do it.”
