From: William Pearson (firstname.lastname@example.org)
Date: Sat Feb 28 2009 - 19:14:47 MST
On 01/03/2009, Vladimir Nesov <email@example.com> wrote:
> I won't look at it so optimistically. Machine learning is actually
> moving towards AGI, they just mostly don't realize it. ;-) As the
> inference algorithms and representations become more powerful, at one
> point they may produce something dangerous. And it's not 100 people,
> it's a mainstream effort.
I'm not quite sure it is moving straight towards AGI, I think it is
veering somewhat to the left ;-). Still useful movement but not on
My criticism of it is that it still deals with fixed hypothesis
spaces, AFAIK. To get AGI you need to go to systems that redefine
their own hypothesis spaces.* Or have you seen examples of ML systems
that do this or are moving towards it?
* For example when a human listens to a ticking noise, we might guess
that there is a fixed period. This shrinks the H space down to
figuring out the length of the period. If we then find the period is
not constant we expand our H space and look for different patterns.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:04 MDT