I came across this article on Yahoo today, sounded interesting so thought I would share it.

Elon Musk is predicting strong AI in 5 years, max 10,

https://ca.news.yahoo.com/elon-musk-robots-could-start-103600968.html

Anyone’s project 5 years from going Skynet on us?

He mentions Deepmind, anyone know anything about them?  Their website is beyond sparse.  Seems they were acquired by Google for an outrageous some of money for building a learning system that can play old Atari video games.

My project will go Skynet on you today.
http://www.Skynet-AI.com

Large Scale Deep Learning is the latest thing in AI.
http://static.googleusercontent.com/media/research.google.com/ja//people/jeff/CIKM-keynote-Nov2014.pdf

Deepmind is a Google AI research project, and (I’m guessing) part of Google’s search algorithms(?) There’s a brief description here.

Here’s your killer robot: http://www.youtube.com/watch?v=-HD_Rd_U0NQ

I pretty much stopped reading anything these pundits are fantasizing.
I have seen a video of Deep Mind, I presume, where they left what seemed to be a neural net mucking about with the controls of an Atari Space Invaders game, and by the next morning it had learned the optimal way of shooting “alien invaders”. So yeah, hilariously similar to military application, except I think it would be a lot harder to capture human movements in patterns than those of 8-bit sprites, and I’m not sure they’d have enough live practice targets.

Teaching a robot to shoot anything that moves however, is very easy to do, right now, with very little AI. Seeing as I just programmed some basic vision recognition in two weeks, all I’m missing is a laser rifle and impervious armour.

It seems to me if we are stupid enough to give robots guns we deserve to be shot by them.

It is moronic on an epic scale to arm robots or put them in charge of system that could kill us all.

“Strong AI has be coming for 5 years” I have heard that for at least 20 years so far.

Dan.

“James Sutherland, Elon Musk is predicting strong AI in 5 years, max 10

https://ca.news.yahoo.com/elon-musk-robots-could-start-103600968.html”

He mentions Deepmind, anyone know anything about them?  Their website is beyond sparse.  Seems they were acquired by Google for an outrageous some of money for building a learning system that can play old Atari video games.
“

Now you’ve done it. I’ve actually glanced at the text.

If its [function] is just something like getting rid of e-mail spam and it determines the best way of getting rid of spam is getting rid of humans ...

There is a more famous paperclip-machine version of this that suggests human tissue makes good material for paperclips. It’s an amusing exercise in the way that my fictional scenarios can beat your fictional scenarios any day.
Speaking of which: Spambots produce 90% of all spam, so it would leave the humans alone, maybe kill all computers, and in doing so, isolate itself.

But minutes after he posted the comment, it was deleted.

And I’m not surprised. I saw some football playing robots at the Loebner Prize. This is how close we are to killer robots:
http://youtu.be/-CT3ZIrSFjw

The vision systems can just about make out simple shapes. They have no chance of distinguishing between a crazed terrorist and a child holding an umbrella.

The NAO robots are really cute. Wonders of engineering, really. But their AI looks like:

Interesting you should mention vision systems. This morning I read two news items: One that announces an improvement in vision accuracy (from 6% successfully recognising cat faces and/or street activities to 12%),
http://www.nytimes.com/2014/11/18/science/researchers-announce-breakthrough-in-content-recognition-software.html?partner=rssnyt&emc=rss
and one about security guard robots that can supposedly, but not reportedly, distinguish between burglars and some teens just hanging out:
http://cleanbayarea.com/electronics/here-come-the-autonomous-robot-security-guards-what-could-possibly-go-wrong/
They also mentioned that one fell down a curb and couldn’t get up anymore

I’ve recently had a go at computer vision myself, and I have to say it’s very hard to get stable and accurate meaning from a bunch of pixels. There’s a reason most vision systems still draw rectangles around round objects.

However, I read this article recently:
http://www.nytimes.com/2014/11/18/science/researchers-announce-breakthrough-in-content-recognition-software.html

I’ve not seen or heard anything else about it apart from this one article though.

Probably because they’re announcing the breakthrough before it actually happens. Both companies have been using practically the same technology, so combining the two doesn’t automatically result in twice the accuracy.
I have seen impressive videos of this cutting edge with an AI describing events on a live video as “man puts something in bag, moves to the left”, but it is also worth saying that what I saw was a carefully selected best case, and with a stationary camera.

I think killer AI would have a very hard time getting at us in the physical world.