[FoRK] Numenta research release

Jeff Bone <jbone at place.org> on Fri Mar 9 12:00:58 PST 2007

On Mar 8, 2007, at 3:16 AM, Eugen Leitl wrote:

> On Wed, Mar 07, 2007 at 08:23:30PM -0500, Jeff Bone wrote:
>>
>> So...
>>
>> I'm a big fan of Hawkin's theory, I think he's got great intuitions
>
> I'm not.

The argument against Hawkins is an argument about method, tradition,  
and culture, not about theory --- and those are really weak and whiny  
arguments to have.  We've had it before.  I won't indulge again  
except to say that I think he gets it more right than not on the  
essential role of prediction in cognition, and on the role of  
architecture in enabling that sort of thing.  Beyond that, the  
specifics are less interesting;  nobody's got a really good, solid,  
working, implementable idea yet.  At least he's approaching it from a  
unique perspective.  I'm a bit alarmed at the distance between the  
"theory" and the actual implementation of the theory as evidenced in  
the toolkit.

>> I'm also a big fan of Python.
>
> You can't state connectionist architectures in a conventional  
> framework.
> At best you can define a billion of asynchronous message passing  
> objects,
> but that's not particularly supported in Python. It's probably better
> supported in Fortran/MPI.

Didn't say you could.  I'm not sure Fortran / MPI makes a good  
substrate either, though.  Something like an evolved Erlang coupled  
with massive array operators (with arrays of processes, etc.) might  
be interesting. (Think X is to Erlang as F-script is to Smalltalk and  
APL.)

> Did you look at the license? No sane person will touch the package
> with a ten feet pole.

Nope.  I don't read licenses as a matter of principle. ;-)  At least  
not until I decide to maybe actually use something.

jb


More information about the FoRK mailing list