REGISTER | LOGIN
Breaking News
News & Analysis

Startup Schools Machine Learning

Software explores faster, cheaper route
2/14/2017 08:00 AM EST
2 comments
NO RATINGS
1 saves
Page 1 / 3 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Max The Magnificent
User Rank
Author
Re: This development likely strongly shortens time to disruptive AI
Max The Magnificent   2/15/2017 9:39:52 AM
NO RATINGS
@DeeJee0: ...the benefits of advandced AI will be so huge and some are/will be specifically desinged to be let lose on the world, so counting on it that it will be sandboxed by any and all researchers, companies, governments or rogue organisations is a failing strategy. It's time to prepare!

I have my aluminum beanie at the ready!

DeeJee0
User Rank
Author
This development likely strongly shortens time to disruptive AI
DeeJee0   2/15/2017 9:07:02 AM
NO RATINGS
In Max blog about "FPGA-Based AI System Recognizes Faces at 1,000 Images per Second" http://www.eetimes.com/messages.asp?piddl_msgthreadid=50159 I argued that this was an example of an AI system needing far less computation power then humans to perform the same task.

I see a dozen of reasons why an advanced AI, once algorithms are optimised, needs far less computation power then humans do in order to reach parity with or outperform the human brain. It can be easily argued that advanced computing systems could be used more optimal then the human brain, leading to orders of magnitude less required computing power. I argued that that would bring it within the reach of current computing equipment.

Today, we see this interesting development that ports the hardest fase of machine learning, the training fase, from a massive data centre to a DIY server or garage. That does not create true AI, but it goes a very big way into lowering the hardware requirments for "true AI".

Shortly said: only the algorithmic side seems to pose any longer a true limit on the development of it. That anihilates the classic "hardware requirements for true AI are so huge that it won't happen for a long time" argument. It's hard to predict how fast algorithm development will go, but it highlightens a great and imidiate urgency to start investigating the potential threats that advanced AI's pose to our society and to see what preventive measures can be taken. As explained in my linked comment the benefits of advandced AI will be so huge and some are/will be specifically desinged to be let lose on the world, so counting on it that it will be sandboxed by any and all researchers, companies, governments or rogue organisations is a failing strategy. It's time to prepare!

Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed