US Air Force buying IBM 64 million neuron computer

[vienna]

It's getting scary how quickly AI development is progressing:

https://www.nextbigfuture.com/2017/0...-computer.html

IBM may not be the biggest name in business and consumer products anymore, but its advances in AI are impressive. I wonder what will happen if they somehow meld this neuron system with IBM's WATSON?...




<O>

[Gargamel]

I'm really not believing this, unless it's only being purchased as a tiny tiny research platform for DARPA and USAF to work on. I don't see this as being that news worthy, given the specs, as neural networks have been under development for decades.

The article reads more like a brochure ad than a real news article.

But it lists it at 7" high that fits on a standard server rack. That's really small. And it says it only has 64m neurons. The human brain has 100 billion. It's a simple little test project, nothing more.

[vienna]

Originally Posted by Gargamel

I'm really not believing this, unless it's only being purchased as a tiny tiny research platform for DARPA and USAF to work on. I don't see this as being that news worthy, given the specs, as neural networks have been under development for decades.

The article reads more like a brochure ad than a real news article.

But it lists it at 7" high that fits on a standard server rack. That's really small. And it says it only has 64m neurons. The human brain has 100 billion. It's a simple little test project, nothing more.

Well, the article did say the purchaser was the Air Force Research Lab, so you're right about that...

However, there is a bigger picture: while the described module is only 64M neurons, remember that an awful lot of the technology we use today and take for granted started out on a scale much, much smaller than in use now and at speed, capability, and power consumption specs unimagined as practical when those technologies first appeared. The idea you can power a device of 64M neurons on just 19 watts is a very impressive accomplishment. Much as how WATSON was a game-changer for cognitive data processing, the capabilities of TrueNorth open the door to abstract cognitive processing. True, its only 64M neurons now, but I can easily remember when a RAM chip of 1MB was a really big deal; and, much like RAM, if you take roughly 1600 of the TrueNorth units as they stand now and find a way to network them, say through a WATSON-type system, numerically you have your 100B neurons...

Here are a couple of links about TrueNorth:

From IBM Research --

http://www.research.ibm.com/articles/brain-chip.shtml

..and if you really want to go deep into TrueNorth, here's a detailed article from the Proceedings of the National Academy Of Sciences of the United States of America --

http://www.pnas.org/content/113/41/11441.full

(It has pictures and everything... ...)...


<O>

[ikalugin]

a RAM chip of 1MB was a really big deal

From what I remember there are physical limits for microelectronics and current technology sort of reached them, and so you shouldnt project the old trends into the future.

[vienna]

...and the state of current technologies shouldn't be a projection they, in turn, cannot be exceeded; history is full of "unexceedable" limits...




<O>

[Skybird]

AI + swarm intelligence = my biggest worry.

[Mr Quatro]

IBM sold it's PC business a long time ago and what are you doing up so early in the morning?

http://www.pcmag.com/article2/0,2817,2483557,00.asp

For most of the 1990s and early 2000s, IBM had a strong PC business, and the ThinkPad was the anchor of its portable personal computer line. However, by 2004 IBM's business had changed, and it was interested in getting out of the PC hardware business. So on May 1, 2005, IBM sold this business to Lenovo and over the last 10 years Lenovo has become the No. 1 PC player in the world.

Not to mention that it is a Chinese company

There was an initial hiccup when the U.S. government challenged the idea that a Chinese company would have access to U.S. technology, particularly via IBM's U.S. government contracts. It took awhile but it got various governemnt approvals, and within a year the integration of IBM's PC business into Lenovo was in full swing.

[Skybird]

Originally Posted by ikalugin

From what I remember there are physical limits for microelectronics and current technology sort of reached them, and so you shouldnt project the old trends into the future.

http://www.subsim.com/radioroom/showthread.php?t=231593

[ikalugin]

Originally Posted by Skybird

http://www.subsim.com/radioroom/showthread.php?t=231593

Thanks.

But my point stands - even if you shift from a two dimentional structure to a three dimentional one (and that is what they are effectively doing now) you still do not get over the actual physical quantum limits.