Hi,

On Sat, Nov 16, 2013 at 12:40 PM, Stewart Mackenzie <[email protected]>wrote:

> ...
> When things get rough am I correct in assuming the lower levels in the
> hierarchy become saturated with signals to the point nothing can be made
> out. At this point the patterns instilled at higher levels of the hierarchy
> take over.
>
I don't think this is correct. Do you have any support?

I believe we have "hardwired" centers for critical operation - think of a
special CLA region whose output has priority (=higher permanences) before
others, we've developed these as "animals". These centers recieve low-level
input (bcs it's fast), that is directly from sensors, or from low level
CLAs.

The examples are, when you put your hand in fire, you move your hand, no
thinking. When we try to drown you :), you shut the mouth under water and
stop breathing.

My support for why not higher level concepts - empirically, the idea of
Al-Queida bombing your house does not trigger such panic, while the
examples above do. And practical too, if you need an autopilot, you dont
build it on Windows 8 which asks FB API, but in an embedded device.

Back to your training example, I think hard training (Pavlov's dogs, 10yrs
of kung-fu) makes the trained pathways in brain so strong to act
sub-counsiously like the survival centers. So if you were to scare such a
trained soldier, he can react automatically and kill you, even if you're
his friend. Because he's been trained to kill.

You mention lower levels become (over-)saturated with signals to conclude
we must use higher levels for quick reactions. I think this is false,
because low levels operate always at the same speed - your eyes do percieve
at a given "frame-rate", no matter what happens, they don't become
oversaturated. When we act real quick (==automatic) it's because you don't
wait for high level concepts to decide "Hmm, this man with a gun pointing
at me. He wears enemy uniform. But he's smiling, maybe we should chat?"



-- 
Marek Otahal :o)
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to