On Thu, May 24, 2018 at 4:04 AM, Alexey Potapov <[email protected]> wrote:

> Linas,
> OK, let's continue the discussion about probabilities.
>
> 2018-05-22 21:18 GMT+03:00 Linas Vepstas <[email protected]>:
>
>> Do not assume that a probability is what you actually want.  Let me give
>> three examples.
>>
>> In real life, when you see a crow, and it is dark, and you want to talk
>> about it, you just say "black crow" as an identifier of the object in the
>> scene.  You don't pull out your photometer and measure it's darkness at
>> 87.68% and a blueish hue of 77%. Why? Because you don't need to do that to
>> have a conversation about it's presence, location, movement, etc. You only
>> need to evaluate crow-ness and blackness sufficiently to distinguish it
>> from all other elements of the scene, and then you can assign P==100% for
>> most practical purposes.
>>
>
> I cannot agree. This is controversial at least.
>

I was referring specifically to the interaction between language and
perception and conscious-attention-awareness. Most of your rebuttal is
about perception, only. I'll make some random unstructured commentary...



> My vision system does perform photometry.
>

Well, yes, of course, at both low and high levels.  Another interesting
thing that it does is to solve differential equations for motion
prediction. Apparently, this is done in the cortical columns. One of the
most interesting modern applications is the teaching of the "still eye"
technique to athletes. The goal of still-eye is to minimize head-movement
and eye movement as much as possible, to obtain the longest possible
duration of accurate motion capture data. A tenth of a second additional
data can make a huge difference in motion prediction. The most dramatic
example of this are the photos of football receivers catching the ball with
their eyes closed: this is to halt visual input, once the visual data is no
longer accurate/usable (because they are being hit.)

http://www.espn.com/espn/e60/news/story?id=4407415
http://www.science20.com/mark_changizi/wide_receivers_who_catch_their_eyes_closed_explained
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0172061

Baseball players don't close their eyes, but are taught still-eye
techniques; the difference between college players and league players has
been measured: the league players watch the ball for an additional tenth of
a second, or so.  You can try this at home: shake your head while someone
throws something at you; its very noticeable.



>  If somebody tells me: "Hey, there is a white crow", then my vision system
> will try to detect
>

No, almost certainly not. You would probably spend a few moments thinking
to yourself, "what the heck is a white crow?" and then decide to look for
something that looks like a dove.  Or a bird, but not a small bird, that is
somehow (?) light-colored. Or something. Kind of confusing. Only after
spotting something that might be appropriate, you would do the visual
pat-down - is it really a crow, or just a white dove, or something else? If
its not a crow, keep looking some more.

"Looking", here, is in the sense of performing (semi-)conscious decisions
about the perceived object; your visual cortex never stopped working, but
the judgment system has to review each nominated object for "whiteness" and
"crowness".


>  Your "crow-ness" is a rough estimation of P(crow|...).
>

Well, yes, in this kind of "what the heck are you talking about" situation,
the optical subsystem has to nominate one or more candidates, while the
judgmental subsystem then accepts or rejects each. This goes back to my
Medieval Scholastic court-room drama: several objects are accused of being
white crows, but does the various evidence actually support that
conclusion?  There's shape, there's size, there is posture.  If there is
insufficient evidence, you discard the claim.

By the time the perceived object is in conscious awareness, ready to
interact with the language subsystem, it is effectively assigned a P=100%,
probability with confidence=100%. (there are corner cases where this would
not be the case) But for the large majority of every-day situations, this
is the case - like when someone says "please pass the salt".  You don't
stop to think: "geee, I think he said salt with 78% probability", and "I
think I see a salt shaker with 92% probability."  And "therefore I will
move my arm with 0.78*0.92 probability".   You just reflexively grab the
object, and perform the action.  Some 5% of the time, you will incorrectly
pass the pepper shaker -- which provides indirect evidence that the
judgment system made the wrong judgment 5% of the time --due to bad
lighting, poor language comprehension, being tired, being drunk, etc.


>
> When I'm saying 'probability',
>

And probabilities are great, at certain subsystem levels.  But, by the time
you start having to perform conscious decision making, you are no longer
computing probabilities, per se: you are evaluating a large network of
inter-related evidence, to see if there is sufficient support to arrive at
a particular conclusion.

You may entertain multiple competing hypothesis: maybe "white crow" is this
object, or maybe its that object. Maybe the speaker is making a funny joke,
and I should be looking not for birds, but old ladies dressed in white.
These are possibilities, and its not so much that the possibility is
assigned a numerical value, per se, but rather, the entire network of
evidence of support, pro or con, is evaluated.  The network with the most
comprehensive evidence is selected, and then the verbal conversation
proceeds with the assumption of 100% correctness.  The verbal conversation
proceeds with assumptions of 100% correctness, until either the
conversation halts or moves on or points of disagreement are found.

BTW, I am not claiming that this is how the human brain actually works (or
am I? I want to have it both ways) - for that we need neuroscience. But I
am claiming that this is a practical way to engineer a system that can have
conversations about the world.  I am not claiming that this is AGI. I am
claiming that, as a device that could actually be built within the next few
years, it would be a better approximation to AGI than what we currently
have.

Linas


-- 
cassette tapes - analog TV - film cameras - you

-- 
You received this message because you are subscribed to the Google Groups 
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/opencog/CAHrUA36-f%2Bn2bnEGq6L7N6Wav3mKZkNxA6d5A%2BT66zPoDEVYSw%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to