REGISTER | LOGIN
Breaking News
Blog

FPGA-Based AI System Recognizes Faces at 1,000 Images per Second

NO RATINGS
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
Max The Magnificent
User Rank
Author
Re: Does "true" AI really needs as much computational power as humans to outclass us?
Max The Magnificent   2/6/2017 9:52:27 AM
NO RATINGS
@DeeJee0: ...I don't believe AI will be easy to contain. [...] So again: are we ready for it?

I'm not :-(

DeeJee0
User Rank
Author
Does "true" AI really needs as much computational power as humans to outclass us?
DeeJee0   2/4/2017 9:25:31 AM
NO RATINGS
Reading this blog I do think of a question I have lately: does "true" AI really needs as much computational power as humans to outclass us?

When phylosophers/researchers talk about "true" AI they talk about the singularity: the moment when self learning machines become as intelligent as we are and start to learn to make smarter copies of themselves at a rapid pace leading to machines that hugely outsmart us in a short time. That could lead to all sort of nasty consequences.

Luckely,  these machines need as much computational power as our human brain has. That is a huge amount and leaves us plenty of time before this "singularity" could happen.

But does it truely needs as much computational power? There are more then a dozen of arguments to be made that a machine could outsmart us in general intelligence with a few to many orders of magnitude less computing power given the wright design and algoritms.

This blog provides a nice example: face recognition. While face recognition is very far from true intelligence, vision with recognising objects and faces is one of the hardest computational tasks the human brain does perform. It happens near automatically, but it is very demanding on our brain. It requires expert skill to recognise several faces a second. Just like Max, I often fail at recognising a single faces in one minute of time. This FPGA implementation has far less computational power yet recognises them at a speed 3 to 4 orders of magnitude greater. Moreover, implementations are still young and could probably further be improved by another order of magnitude or more.

At the moment, the difficult taks of recognising objects and faces in images, understanding the meaning and context in text and speech, constructing meaningfull speech, machine translation from one language to another are either solved or on the brink of being solved.

Still, that isn't general intelligence isn't it? But understanding meaning out of data was one of the major bottlenecks that had to be solved to reach general artificial intelligence. Without it, it was plain dumb AI. Once "meaning" is generalised by linking the meaning of speech to the meaning of audio, pictures, video and other sensory data the next step of "stitching" the seperate components together to a more general intelligence that starts to learn on itself could be a lot less of a hurdle. If true, it could be done with current hardware and learn in a matter of weeks rather than years. Are we ready for it?

I don't believe AI will be easy to contain. The potential rewards are to great ranging from commercial data mining over designing revolutianairy new products or to exploit inefficiencies at the stock market to collect billions of dollars. Heck, many applications will be specifically developped to set lose on the world. Think of espionage, counter espionage to protect yourself against the former, AI electronic warfare to cripple a nations infrastructure before firing a first shot or consider the plain malware/ransomeware variant set lose by criminals. So again: are we ready for it?

 

Clive
User Rank
Author
Re: At least 2 potential buyers
Clive"Max"Maxfield   1/27/2017 2:48:23 PM
NO RATINGS
@perl_geek: ...the little voice would mutter "That's so-and-so" in my ear...

Did you see my article "Deep learning machine vision system aids blind and visually impaired" ( http://ubm.io/2fcw0kf ) -- instead of whispering in your ear, it would use bone conduction -- I honestly believe this sort of technology will be with us in a few years.

perl_geek
User Rank
Author
At least 2 potential buyers
perl_geek   1/27/2017 12:42:57 PM
NO RATINGS
If one of these was wearable and not insanely expensive, it would make social occasions much less nerve-wracking for me. Knowing that the little voice would mutter "That's so-and-so" in my ear when an acquaintance came within range would save the frantic internal database search that so often happens.

The other little voice that delights in telling me "You know that person; guess their name/context" would probably resent it, though. :-)*

Clive
User Rank
Author
Re: Low-tech conference solutions
Clive"Max"Maxfield   1/27/2017 12:29:02 PM
NO RATINGS
@Sean: ...The short story Norbert and the System by Timons Esaias (available online) explores this possibility in quite an amusing way....

I just did a search and found it in this The Best of Interzone anthology for only $0.01 (plus $3.99 shipping and handling LOL) -- I just ordered it -- thanks for the suggestion.

Clive
User Rank
Author
Re: Low-tech conference solutions
Clive"Max"Maxfield   1/27/2017 12:24:10 PM
NO RATINGS
@Sean: ...if your name badge is hanging from a lanyard, it should be printed double-sided, so your name is always visible even if it gets flipped over...

Yes, we do this at ESC now -- that's very handy -- it seems like badges naturally want to be the wrong way round.

But when someone comes up to you shouting "Max, how good to see you," it's a bit "obvious" to try to look at his badge to work out who he is.

That reminds be of the joke that a cat always lands on its feet, and a dropped piece of buttered toast always lands butter-side down, so if you duct-tape a piece of buttered toast to th eback of a cat, it will end up hovering in mid-air.

Check out this YouTube video showing how to create a perpetual motion machine using a cat smeared with grape jelly

Clive
User Rank
Author
Re: Low-tech conference solutions
Clive"Max"Maxfield   1/27/2017 12:17:42 PM
NO RATINGS
@Sean: ...First, introduce yourself with your own name...

When I first moved to the USA, it was common at the conferences I attended to just have given names on badges. When I was submitting online, I would use my given name "Clive," but thsi isn;t common in the US, so on several occasions I woudl find myself with a badge that proclaimed me as being "Olive" -- that certainly raised a few eyebrows LOL

Clive
User Rank
Author
Re: Fields subroutine
Clive"Max"Maxfield   1/27/2017 12:13:56 PM
NO RATINGS
@Marc: ...I personally need as much cheering up as possible these days...

I'm sorry to hear that

I'd tell you a joke (but you'd only laugh)

Clive
User Rank
Author
Re: Fields subroutine
Clive"Max"Maxfield   1/27/2017 12:12:58 PM
NO RATINGS
@Marc: ...In recent months, I've used Youtube clips of the candy assembly line fiasco with Lucille Ball and the first two minutes of Fellini's La Dolce Vida to illustrate concepts in economics and politics...

There are some clips that will always stand out -- in the case of I Love Lucy, out of all the (100s?) episodes, that's one that will always be with me -- I remember seeing it when I was a kid and rolling on the floor with laughter (the way most people react when they read one of my more technical pieces LOL)

seantellis
User Rank
Author
Re: Low-tech conference solutions
seantellis   1/27/2017 9:04:36 AM
NO RATINGS
Also, once mnemonic spectacles become ubiquitous, not using them will be a badge of honor. The short story Norbert and the System by Timons Esaias (available online) explores this possibility in quite an amusing way.

Page 1 / 2   >   >>
Most Recent Comments
Tim R Johnson
 
ewertz
 
antedeluvian
 
ewertz
 
perl_geek
 
R_Colin_Johnson
 
perl_geek
 
R_Colin_Johnson
 
dt_hayden
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed