Re: Google AI: well at least it's not not evil



casey <jgkjca...@xxxxxxxxxxxx> wrote>
...

What are you imagining is going on for a machine to
be aware of time?

It is a machine!! It changes states at each tick
of the clock.


Yes, but we go to great efforts to make our programs
time insensitive.

I don't remember going to great effort to make my
programs time insensitive. If it was an issue I
made them time sensitive (like games to run on pc's
with different clock rates) but never thought about
_making_ them time insensitive.


If you take a program like all the millions and
millions of lines of code in Microsoft Windows,
and put it on a machine that runs twice as fast,
the program works exactly the same way as it did
on the slower machine. The great bulk of the
code is time-insensitive - it's function and behavior
doesn't change when you make it run twice as fast.
(parts of it are very time sensitive - but the bulk
is not).


The bulk of it is not time sensitive because there
isn't requirement for it to be time sensitive just
as there are many tasks we do that are not time
sensitive.

However providing the computer ticks over fast enough
for the particular task at hand it can always adapt
to anything slower or faster just as we do. If I play
a tune fast or slow you will still recognize the tune
within the limits set by the rate at which my brain
can input and process the data.

If it was possible to slow the brain down the tune
would seem to speed up and if you could speed up the
brain the tune would seem to slow down. Either way it
is not a major issue for the brain and neither should
it be for a program designed to recognize and respond
to temporal patterns like tunes (or Morse code).

Of course for a particular task there is a minimum
rate at which the data must be processed. That is
also a limit for us. You cannot process a flying
bullet but you can process a thrown ball.

As for Windows it has to run at sufficient speed to
match the human interface. Like any other process
there is a minimum practical speed. If Windows
flashes up a message it must time the duration to
be independent of its clock speed to give the user
time to read it but not leave it there too long.

We and computer programs are time sensitive _when_
it is required.


There is no magic soul-like need for a "sense of time".


:)

If you look at any typical computer programming language,
you will see that we have designed the language to remove
all concepts of time from the code.


The code is a list of instructions that essentially wire
up a particular machine. That machine will tick over at
some rate. Providing that rate is fast enough for the
task it will be able to adapt to achieving some time
critical goal.


The primitive functions of the language, produce the
exact same behavior no matter how fast you run the code.


Which is true for us also. The faster we process the data
the faster the input can become. When you first learn Morse
code you can only process a slow input, with practice you
can process a faster input. You adapt to the speed at which
the input is coming in _regardless_ or your own clock speed.


Only when you extend the language by adding system calls
that return the value of a real time clock do the programs
become time sensitive.


But often that is not an issue. The only issue is that
the fastest the machine can run at is fast enough. It
adapts to the speed of the input. And it is no different
for us. Our experience of the rate of time is determined
by the rate at which our neurons respond to an input.
Our intuition of the duration of time is determined by
the rate neurons can process data. We may compare that
with a real clock but it is not determined by a real
clock.


But in all the code, it's that one call to the clock
function, and nothing else, which makes the program time
sensitive.


Concentrate on what the machine does, not the code that
wires up the machine. The machine ticks over the way we do.
When a set of input pulses to a neuron it triggers an output
pulse and the speed at which that happens if fairly constant.
The only issue is, is it fast enough for the task at hand.


Our abstract programming languages are sequence aware,


I don't see programming languages as being "aware". But
again I say concentrate on the machine not the code that
determines how it is wired up. The instructions for
building a clock are not time sensitive but the clock is.


and sequence awareness is a weak stand-in for real time
awareness. I have to wonder if maybe a sequence aware
AI program might be able to use it's understanding of
sequence (the order in which things change over time)
might be good enough to allow it to understand enough
about time to not look too stupid when trying to
communicate with humans.


You use terms like "time awareness" and "sequence awareness"
and "understand" you really have to tease that out into the
physical you are talking about. I think your problem is
imaginary.


A simple example of a time aware program is one that
allows you enter a number on a keyboard. It prints out
"slow" if you take more than 5 seconds to enter the next
number, and it prints out "fast" if you take less than 4
seconds.


That is a simple example of a "real time aware" computer
program.


If your programming language doesn't include a real time
clock function, you can't write that program.


Sure you can. What is meant by fast or slow? If your brain
ticked over twice as fast as mine the relative times it took
something to happen would be the same, only the absolute
times would be different. If a turtle was "slow" and a
rabbit "fast" we would attach the word to whatever speed
the turtle or rabbit had relative to our own brain's
measure of time determine by the reactive times of neurons.


It's easy to add _some_ awareness of time to a program.
A simple call to a clock function or input data which
comes with time stamps allows you to do it.
But to duplicate what the brain does,


I wouldn't assume that the brain doesn't use time stamps.
Same problem similar solution.


we don't just need to call a clock function now and
again. The processing must be time aware at the core
of the algorithm.


What makes you assume everything the brain does requires
it to be "time aware"?


However, many AI projects have a history of not being
time aware. The behavior they produce is not sensitive
to the timing of events in their environment, and the
behavior it produces tends not to be temporal in nature.


Our behavior is also not sensitive to the timing of
events either, unless it is relevant to the problem.

I don't know how long it has taken me to type this post.

Most of the things I do are not time critical in the
sense of catching a ball or balancing while standing,
walking or jumping.


As we interact with our environment, we have an innate
sense of how much time is passing. If we see something
moving, and then close our eyes, we have an innate idea
of where the thing is even after we have stopped
watching it. We can make a good guess where to reach
with our hand to grab it.


I don't think this is an innate sense which you seem to
be implying is a result of information as pulses. This
innate sense is a high level computation that allows
us to predict the path of a moving object. An ordinary
gofai program can compute the future positions of a moving
object, as they do with the ball catching robot. Without
this computation no amount of "temporal" gates will enable
the machine to catch the ball.



We can listen to a repeating rhythm, and notice when
the tempo changes. We can predict when the next beat
will happen. When we talk to someone, we pick up on
their speed of talk, on how long they pause between
ideas, or how fast they respond after we say something.
All these subtle temporal clues carry a lot of
information for us.


There is no question that AI must be able to represent
the temporal patterns in its world model. I just don't
see that as anything special or requiring the dynamic
movement of pulses.


From> c...@xxxxxxxx (Curt Welch)
Date> 01 Apr 2008 17>34>08 GMT
Local> Wed, Apr 2 2008 4>34 am
Subject> Re> Google AI> well at least it's not not evil


A bit, is a spatial concept of information.


A bit is a binary digit. It is not a concept of
information, spatial or temporal.


It is static, it has no location in time, only a
location in space.


Everything has a location in time even if it is
the same location time after time.


It's our spatial concept of information which
is the result of years of having to record
all our information in a media which doesn't
support temporal encoding.


No, it is a binary digit. It is not a spatial
concept of information.


But a pulse, is a temporal event marker. It has
a location in space, and a location in time.


Just like everything else has a location in space
and in time. Everything exists in space-time.


It encodes most of it's information in the time
domain. It's value is not its static state (on
or off), all pulses are the same in that regard.

The value of the pulse is either there or not at
some particular place at some particular time like
the value of a bistable device. Just because the
pulse is on the move doesn't make it temporal.
It exsit fully in space at some point in time
even if some device can only read it in a serial
manner.


It's value is a function of where it exists in
space, and when it exists in time - with a huge
amount of information encoded in that time domain.


Physically it exists at some location at some given
time like everything else. If you write a 0 or 1 on
a piece of paper you can carry it somewhere else.

To say that a pulse's value is a function of where
it exists in space and when it exists in time isn't
the case. As a wave of pulses is carried along by
whatever medium (even moving strips of paper) the
information contained in the pattern of the pulses
remains the same regardless of it current position
in space and/or time.

I suspect you are thinking of a pulse being processed
in your "temporal" gates? A pulse at the output of
the gate at time_n-1 has the information (P then Q)
as opposed to a pulse at the entry of the gate
at time_n meaning just P?

Well as the values of a static, non moving block
of registers changes, the information in that
block also only exists at some particular time.

A moving wave may carry a "temporal" pattern for
millions of years.


The fact that the brain encodes much of the
information it processes in the temporal domain,

No sorry it exists physically somewhere in the
spatial domain at some point in time. It may
be being moved by the nature of its carrier
such as a wave but it still exists spatially.

Just because a device is limited to reading the
data in a serial fashion doesn't mean the data
itself doesn't exist in the spatial domain.

and the fact that all the nodes which process this
data (the neurons) perform time sensitive processing
functions, creates an understanding of time as a
basic primitive to the brain.

It's why understanding movement, and speed, and
change per time, is so easy and natural for us.

"Understanding" occurs at a higher level like the L
made up of D's.

Did you know that damage to a certain part of your
brain will result in you not experiencing motion?
All the rest of the brain will be happily doing all
the "temporal" stuff you talk about but the brain
will not see objects as "moving". A moving object
will be seen as a succession of static , strobe like
snapshots. So what does that say about your notion
of time sensitive processing neurons making it
easy and natural for us to understand change per
time?

JC
.

Relevant Pages

  • Re: Einsteins Most Illustrious Red Herring.
    ... the very definition of a 'frame or reference' on logical grounds. ... L/during the transit of a LH moving light pulse across L and by ... Einstein) according to his own clock readings, (since the pulses both move ...
    (sci.physics.relativity)
  • Re: Einsteins Most Illustrious Red Herring.
    ... the very definition of a 'frame or reference' on logical grounds. ... L/during the transit of a LH moving light pulse across L and by ... Einstein) according to his own clock readings, (since the pulses both move ...
    (sci.physics.relativity)
  • Re: language, perception and time
    ... I have a clock which consists of a mass of bismuth 212. ... It doesn't have any moving parts -- when the bismuth turns into something else, the fact that the something else immediately leaves the clock at high speed is not indicative of any motion in or of the clock itself. ... I can't deny that there are moving things in my brain, but I don't know how, or if, they are part of the clockwork. ...
    (alt.usage.english)
  • Re: Consciousness: whats the problem?
    ... this raw sensory data. ... The clock doesn't tell you the time. ... type of decision our hardware makes by channeling energy flows? ... information flow rates in the brain is not very impressive ...
    (comp.ai.philosophy)
  • Re: More On The Flaws in Einsteins RoS.
    ... moving one's reading, in the stationary frame. ... Yes, in the coordinate system of the stationary frame, this is true ... and that the moving clock is at the rear of a moving rod of length L ... or clock synchronization. ...
    (sci.physics.relativity)