You know, if we actually do want to actually build an AGI (ie, a box
that when switched on, thinks), we really do need a solution to the
hardware problem, that is how much and what kind of computation (and IO
too!) that we need.
We also need an answer that is better than wait until 1-2-2029, then run
down to the CompuHut with $1,000.00 in your wallet...???? (I'm looking at
you, Kurzweil!)
To start with, take a look at what Jeff Hawkins had to say for himself
in that recent interview, it was really good. Listen carefully too!?? One
of the points he made is that the algorithms used by the brain to
recognize objects and such have been measured to operate on timescales
on the order of fractions of a second and can learn new concepts with
only a few examples where neural networks take thousands of examples and
end up with extremely brittle representations.
This implies that at least those aspects of intelligence are either:
very fast. (can be accomplished by very little but very clever computation)
very parellel (can be split among thousands of processors with little
overhead)
Evidence from neural science is somewhat indeterminant on this point. It
can be argued that the algorithm is fast because it doesn't take much
time and there wasn't that much in the way of mesaurable signaling, or
it can be argued that it is paralell because even though only a little
activity was detected, large brain regions received parts of the
stimulus and acted in concert to solve the problem quickly.
We are looking for a new algorithm, or an algorithm that is unrecognized
for being relevant to AGI.
Actually, a useful AGI agent that is complex enough to learn like a
child must have all the same learning capabilities and a complex enough
environment.
So a great deal of work (eek!) needs to be done before the theoretical
stuff can even be brought in, stuff like robotics and/or simulation and
stuff. much of that has been done already but a hell of a lot remains.
It needs to be developed to the point where I can obtain it for a sane
amount of money and actually run it on computers I actually own... --
all the software I've tried recently errors out in one way or another
while I'm simply trying to set up the development environment, this is
before actually getting a development environment running and to the
point of testing example modules...
example FPGA cards, both are probably far too expensive for mere
mortals. =\
https://www.xilinx.com/products/boards-and-kits/alveo/u250.html#overview
https://www.intel.com/content/www/us/en/programmable/products/boards_and_kits/dev-kits/altera/intel-fpga-pac-d5005/overview.html
--
Clowns feed off of funny money;
Funny money comes from the FED
so NO FED -> NO CLOWNS!!!
Powers are not rights.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T1846529a07d4cff4-M89b58b7b0cbdc4e68385442a
Delivery options: https://agi.topicbox.com/groups/agi/subscription
