Re: hello?

curt@xxxxxxxx (Curt Welch) wrote:
casey <jgkjcasey@xxxxxxxxxxxx> wrote:
On Mar 10, 5:22=A0pm, c...@xxxxxxxx (Curt Welch) wrote:
casey <jgkjca...@xxxxxxxxxxxx> wrote:
On Mar 10, 1:55=3DA0pm, c...@xxxxxxxx (Curt Welch) wrote:
casey <jgkjca...@xxxxxxxxxxxx> wrote:

Consider the simple visual target recognizer. It exists in the
same "complex" environment we do. However it only deals with,
reacts to, some small part of that environment. The complexity
is filtered out. It can only deal with the end product, the
target, in terms of say its color, size and relative location.
That is all its environment consists of.

As a follow up to the long post I just wrote, let me talk about what you
write above and how it irritates me when people write in this style.

You wrote some visual target recognizer software which runs on a real
computer and I assume, takes a visual image and analizes it to try and find
a specific type of target in the image.

You ask me to "consider the target recognizer". Are you talking about the
software you wrote, or the physical machine you built which has the power
to see a target located in the room? I don't really care which you are
talking about, but the software doesn't exist as something separate, so no
mater how you are thinking about it, I know you are in fact talking about
the physical machine you built.

Then you write:

"It exists in the same "complex" environment we do.

Yes, very true.

"However it only deals with, reacts to, some small part of that
environment."

Here's where you start to talk nonsense. The physical machine "deals
with", and "reacts to" the entire complex environment. The light from the
TV is falling on the video camera, and the hardware is responding to that
light just as much as it is responding to the light reflected off the
target. How can you say the hardware is not "dealing with" the light from
the TV next to target? It makes no sense to say that. Your computer is
dealing with it, and it is reacting to it.

Lets say your machine has two major reactions. Lets pretend it has a
light, and will turn the light on when it sees the target, and keep it off,
when it doesn't see the target. Your machine is reacting to the TV by
clasifying it as "not a target" just as much as your machine is reacting to
the target, and classifying it as "a target". It's doing just as much work
to turn the light off, as it is to turn the light on. How can you really
justify your position that the "target recognizer" is only "dealing with",
a "small part of the environment"?

"The complexity
is filtered out. It can only deal with the end product, the
target, in terms of say its color, size and relative location.
That is all its environment consists of."

The environment of the machine is the room it's located in. How it reacts
to that environment can accurately be described as a "filtering" effect.
The light sensor in the camera filters out most the information in the
light when it transforms the light into an electrical signal. The
electrical signals are just a gross average of the complex changing light
patterns. The rest of the computer hardware will then filter out most of
that information as it transforms those signals (which represent the TV and
the shelf of books as well as the target), into the signal sent to the
light to make it turn on and off when the computer sees the target.

There's just no way this device (as a whole) can accurately be described as
"dealing with only some small part of the environment". The device, and
the software running on the device deals with the full complexity of the
environment.

If by "target recognizer", you are limiting your idea of "it" to the
software you wrote, then it's valid to say that part of the hardware is at
least dealing with the simplified representation of the environment found
in the digital output signals of the camera. But even that is still a
fairly complex environment. And in doing this, you are simply limiting
your view of the "agent" to the part of the computer hardware which is
receiving the signal from the camera (aka you have moved the camera into
the environment and only consider the computer which is processing the
signal the agent).

All my comments here are just me trying to get you to look at what you are
saying, and thinking, from a different perspective - one which I believe in
very important to anyone trying to build hardware which will duplicate what
a human can do (aka AI projects).

It's far to easy, and far too common, for people to get lost in these
abstractions we all use (like "software") and believe they are something
other than a fancy way of talking about the behavior of computer hardware.
I think everyone needs to stop an look at their ideas by translating them
back to hardware terms as a way to double check the validity of anything
they come up with for AI.

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Relevant Pages

  • Re: What is Forth?
    ... the target, and then tell me why those aren't advantages for seaForth. ... But now there is Forth hardware, ... support both a Forth development environment and the target application. ... My only point was that he said he wanted to put Forth compiler into ...
    (comp.lang.forth)
  • Re: [RFC] Kernel version numbering scheme change
    ... setup a dedicated machine for a build environment (and all these hyped ... customers' various distros). ... the guy building the full distro in a chroot on an embedded ARM or MIPS ... just because it's the target. ...
    (Linux-Kernel)
  • Re: [RFC] Kernel version numbering scheme change
    ... cross-compilation environment. ... build completely OS-agnostic packages, ... The build environment in the chroot is the correct release. ... if your build environment mimmics so well the target except ...
    (Linux-Kernel)
  • Re: ER: weird behaviour when using s=6 option in sqlhosts
    ... The environment consists of three instances, ... The cdr tool is an ESQL/c program and would need to use an SQL connection to connect to the server. ... Since part of template realization is to verify that the data is correct, it must connect to those servers. ... After that I defined and realized a template on ER primary and tried to realize it for target instance also. ...
    (comp.databases.informix)
  • Re: [RFC] Kernel version numbering scheme change
    ... cross-compile> 12.000 source packages. ... It's easier as long as you have one build environment very close to the ... target, which tends not to be possible anymore as time passes or for too ... Problems range from auto-detection of kernel version to ...
    (Linux-Kernel)