h a l f b a k e r yA few slices short of a loaf.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Please log in.
Before you can vote, you need to register.
Please log in or create an account.
|
A recent DARPA robot is learning to cook from Youtube videos,
which
is possible because there are so very many youtube videos about
cooking out there. But choose a different arbitrary task, such as
car
repair, and you'll find far fewer videos about it. Specialize to
something like "how to
change the transmission on an '83 Buick"
and
you'll be lucky to find one or two. Robots need lots and lots of data
to
train from. Training a human is likely to receive as an apprentice
or
fooling around on his own car, robots can't get. Robots also can't
yet generalize across skills nearly as well as humans.
This business would pay mechanics to talk all day to a video
camera
as they work, explaining what they're doing and how and why.
Tens
of thousands of hours of video would be compiled to be fed to
robots. If successful the program could be expanded to other
topics,
such as guerrilla warfare tactics and subversive propaganda.
edited.
[link]
|
| |
This seems like a high-bandwidth way to accomplish something for which you might be able to program the robot to do, straight up, in less time. + for that! |
|
| |
Only gets a bun if you can come up with the robot
version of "Reefer Madness" (the 1936 version
thereof).. |
|
| |
//you might be able to program the robot to do, straight up,
in less time.// Surprisingly enough it takes much less time to
train a general learning algorithm than to program individual
tasks. In the past these things weren't possible at all. |
|
| |
I suppose if you're referring to a neural net type. But this seems beyond that in some fashion. |
|
| |
My view is the opposite of [Rayford]'s. This seems
like a very inefficient process. |
|
| |
Robots learning from existing videos is sort of OK,
because the videos already exist. Plus it's a clever
piece of AI to be able to learn from 'found' material.
But if you're
making the videos specially, I'm sure there are more
efficient ways - either teach the robot directly, or
create a standardised video with reference points,
barcodes or whatever the robot needs to learn from. |
|
| |
It seems to me that you are asking the mechanics to help robots take the mechanics' jobs away from them.... |
|
| |
I'm doing no such thing. I'm proposing that someone else use
someone elses money to pay mechanics to help robots to
take the mechanics' jobs away from them. |
|
| |
Max, how was that opposite of my view? |
|
| |