GOOGLE'S DEEPMIND machine learning artificial intelligence (AI) computer has made a significant step towards ‘thinking like a human', says the company.
And no, that doesn't mean it's learned to burp, fart and turn on repeats of Mock The Week on Dave because it can't think of anything good to do. Rather, it can learn to do tasks one after the other using transferable skills from the old tasks.
The key here is that DeepMind has been taught not to forget useful information and instead store it away to draw on. Not as well as human beings, but it's a start.
"If we're going to have computer programs that are more intelligent and more useful, then they will have to have this ability to learn sequentially," James Kirkpatrick at DeepMind told The Guardian.
The problem comes from the fact that computers learn to do something, and then when they've mastered it, they get taught something else and forget the first thing.
Now, AI researchers have found a way to cure this so-called ‘catastrophic forgetting', instead putting it to good use. And it's a good thing too because without that skill, there can never be a general purpose artificially intelligent being.
Or on the other hand, it's a terrible thing because it makes Skynet all the more possible.
"Humans and animals learn things one after the other and it's a crucial factor which allows them to learn continually and to build upon their previous knowledge," said Kirkpatrick.
The secret is in getting the neural network to evaluate what it's learned in a task, and "hard-wire" the information that was most useful. The results were noticeable.
When given 10 Atari games to learn (you know, Space Invaders and such), the new technique allowed DeepMind's machine to perfect seven of 10 games. Without the new processes, it had barely learned one in the same period.
But just like humans, AI computers don't know what they don't know and quite often miss the information that they needed to remember. Just like humans, they don't always learn their lesson.
"We know that sequential learning is important, but we haven't got to the next stage yet, which is to demonstrate the kind of learning that humans and animals can do. That is still a way off. But we know that one thing that was considered to be a big block is not insurmountable," Kirkpatrick said.
But although this is a big step in the right direction, we're not exactly at automation level yet, so you can stand down the Anderson Shelters. µ
'Ah - yes - we're ignoring your wishes for a reason there, leave it alone'
And, er, not much else
To serve, protect, and get incredibly hot and dusty
Symantec links attack to prolific Lazarus hacking group