Philosophy Forums


Just what is a cat?

PrintPrint


Page: 1 2

Just what is a cat?
123savethewhales
Assistant Professor
Avatar

Usergroup: Members
Joined: Mar 25, 2009

Total Topics: 7
Total Posts: 346
Posted 04/05/09 - 02:57 AM:
Subject: Just what is a cat?
quote post
#1
If any of you follow the development of AI, you will know how difficult it is to program an AI to recognize a cat. It takes huge amount of logical commands and massive computing power just to do this with some accuracy. Yet human seems to be able to do so effortlessly.

By definition (according to dictionary.com), philosophy is "the rational investigation of the truths and principles of being, knowledge, or conduct". If we consider a computer as a tool of pure logic, this will mean that it shouldn't be too hard for philosophers to teach a computer what a cat is.

So, what is a cat? How do we rationally tell the difference between a cat and a dog? Is rationality alone enough to distinguish the difference between the two?

Keep it simple.
Death Monkey
Tenured Poster
Avatar

Usergroup: Members
Joined: Sep 18, 2003
Location: Eindhoven, The Netherlands

Total Topics: 7
Total Posts: 2598
Posted 04/05/09 - 03:15 AM:
quote post
#2
A major factor in the easyness or difficulty of teach somebody something, is the learning ability of the person or thing you are trying to teach. Even our most sophisticated AI's have learning capabilities far below those of a human being. In fact, I doubt that we can currently even match the learning ability of a mouse.

DM

Pseudoscience makes Baby Jesus cry.
swstephe
Tenured Poster
Avatar

Usergroup: Moderators
Joined: Apr 20, 2006
Location: borneo island

Total Topics: 28
Total Posts: 3304
Posted 04/05/09 - 05:09 AM:
quote post
#3
123savethewhales wrote:
If any of you follow the development of AI, you will know how difficult it is to program an AI to recognize a cat. It takes huge amount of logical commands and massive computing power just to do this with some accuracy. Yet human seems to be able to do so effortlessly.


I guess you are reading National Geographic from the 1950's? Who says it is difficult? You can't define how complex the solution is unless you know the problem set. It is actually pretty simple if the computer is only ever going to be looking into a room full of cats. If it moves, and it is a different color than the background, it is a cat. Visual recognition has progressed far beyond that of human beings. You can have a computer look for faces of suspects in a football crowd, and identify them instantaneously. Something a human expert would be very hard pressed to do unless they knew the person quite well.

Humans don't use reason to recognize things. That process is at a much lower level. We associate images and superficial attributes with concepts, either words or symbols. A computer can do the same thing with a properly trained or self-trained neural net. You recognize something and automatically have a strength of association before it even enters your conscious mind. Reason has to come into play to explain away recognition later. You instantly recognize, "there is a cat, I'm sure of it" -- then you have to reason why the cat seems to be brushing its teeth with your toothbrush, and reason out an alternate explanation. Obviously, it is a witch!

Anyway, how do you explain to a computer what a cat is? Just say it is a dog-computer that has a bug which makes it ignore commands sent to it and suffers long periods of downtime or self-cleaning process. Sometimes it eats mice.

Ethics is the measuring of morality. Morality is the measuring of good. Good is the measuring of benefit. Benefit is the measure of values.
123savethewhales
Assistant Professor
Avatar

Usergroup: Members
Joined: Mar 25, 2009

Total Topics: 7
Total Posts: 346
Posted 04/05/09 - 10:16 AM:
quote post
#4
swstephe wrote:


I guess you are reading National Geographic from the 1950's? Who says it is difficult? You can't define how complex the solution is unless you know the problem set. It is actually pretty simple if the computer is only ever going to be looking into a room full of cats. If it moves, and it is a different color than the background, it is a cat. Visual recognition has progressed far beyond that of human beings. You can have a computer look for faces of suspects in a football crowd, and identify them instantaneously. Something a human expert would be very hard pressed to do unless they knew the person quite well.

Humans don't use reason to recognize things. That process is at a much lower level. We associate images and superficial attributes with concepts, either words or symbols. A computer can do the same thing with a properly trained or self-trained neural net. You recognize something and automatically have a strength of association before it even enters your conscious mind. Reason has to come into play to explain away recognition later. You instantly recognize, "there is a cat, I'm sure of it" -- then you have to reason why the cat seems to be brushing its teeth with your toothbrush, and reason out an alternate explanation. Obviously, it is a witch!

Anyway, how do you explain to a computer what a cat is? Just say it is a dog-computer that has a bug which makes it ignore commands sent to it and suffers long periods of downtime or self-cleaning process. Sometimes it eats mice.


Yes I know about the trained neural net. It is possible to write a program that somewhat mimic learning, which you can then train it through trial and error. While neural net is able to solve the cat or dog problem, it was a disaster when we try to hard code an AI to do the same.

I guess the real thing I want to bring up is our rationality and how often we rely on it. Now that we established that we recognize a cat without the use of it. Are we really logical when we judge other events? Or do we do the same as we recognize a cat, only later to rationalize it when being asked for a reason?

I really didn't know about computer being able to recognize a specific face in a football crowd, do you have any particular source where I can read up on that?

Keep it simple.
swstephe
Tenured Poster
Avatar

Usergroup: Moderators
Joined: Apr 20, 2006
Location: borneo island

Total Topics: 28
Total Posts: 3304
Posted 04/05/09 - 05:28 PM:
quote post
#5
123savethewhales wrote:
Yes I know about the trained neural net. It is possible to write a program that somewhat mimic learning, which you can then train it through trial and error. While neural net is able to solve the cat or dog problem, it was a disaster when we try to hard code an AI to do the same.


So now AI is only valid if hard-wired to recognize cats? Do you think humans are born with the ability to recognize a cat without any prior experience at all? Actually, hard-wiring a neural net is no problem. I've seen some new models of scanners coming out which have a hard-coded neural net for OCR, (reading text from scanned-in image), since the problem set is relatively small -- and remember, there are many humans which don't have the ability to read, (let alone the number of languages supported).

123savethewhales wrote:
I guess the real thing I want to bring up is our rationality and how often we rely on it. Now that we established that we recognize a cat without the use of it. Are we really logical when we judge other events? Or do we do the same as we recognize a cat, only later to rationalize it when being asked for a reason?


Our neurons activate in an area associated with cats, so we associate it with words or symbols we use to represent cats. A person with a severed connection between the two spheres of the brain can recognize the word "cat" with one side, and the image "cat" with the other, but has difficulty making the connection between the two concepts. This is apparently a physical ability that we are incorporating into reason.

123savethewhales wrote:
I really didn't know about computer being able to recognize a specific face in a football crowd, do you have any particular source where I can read up on that?


Sure ... http://www.wired.com/politics/law/news/2001/02/41571 ... it is already decade-old technology. I was playing around with the technology myself. We had a project for a police profile database. There are open-source libraries for facial recognition, (google for "opencv"). I had this system where they could take your picture with their mobile phone camera and the phone would tell you if the person was in the profile database. The recognition library is supposed to match around 85% of the time, even with attempts to disguise your appearance. People don't realize how close they really are to "Big Brother is Watching You" ... see my avatar?

Ethics is the measuring of morality. Morality is the measuring of good. Good is the measuring of benefit. Benefit is the measure of values.
J. Random Hacker
banned
Avatar

Usergroup: Sponsors
Joined: Apr 11, 2009

Total Topics: 12
Total Posts: 479
Posted 04/11/09 - 09:03 PM:
quote post
#6
I think it's a two part process. First we have a heuristic pattern matching of the shape, size, texture and color of cats and dogs. Then based on this analysis, we tentatively assign cat or dog status until disconfirmed by some other cue such as that "cat" suddenly barking or that "dog" hissing. I don't think any of this is difficult for a computer to do once it knows how. Making it plastic enough to learn how to do it on its own is the hard part.
123savethewhales
Assistant Professor
Avatar

Usergroup: Members
Joined: Mar 25, 2009

Total Topics: 7
Total Posts: 346
Posted 04/12/09 - 12:15 AM:
quote post
#7
swstephe wrote:


So now AI is only valid if hard-wired to recognize cats? Do you think humans are born with the ability to recognize a cat without any prior experience at all? Actually, hard-wiring a neural net is no problem. I've seen some new models of scanners coming out which have a hard-coded neural net for OCR, (reading text from scanned-in image), since the problem set is relatively small -- and remember, there are many humans which don't have the ability to read, (let alone the number of languages supported).

Well, no it's not so much the validity of AI. Clearly bottom up approaches still does everything it needs to do, so as a product it's a success. It is more that, if we can hard-code the AI for a cat, then 'we' have all the logic to what a cat is. Otherwise, we don't.

Sure ... http://www.wired.com/politics/law/news/2001/02/41571 ... it is already decade-old technology. I was playing around with the technology myself. We had a project for a police profile database. There are open-source libraries for facial recognition, (google for "opencv"). I had this system where they could take your picture with their mobile phone camera and the phone would tell you if the person was in the profile database. The recognition library is supposed to match around 85% of the time, even with attempts to disguise your appearance. People don't realize how close they really are to "Big Brother is Watching You" ... see my avatar?

Well I was reading this book "On Intelligence" a while back, and even downloaded the program numenta.com to play around with (at the time only the old one was released). Now they already have their full black and white image recognition demo up for download, which is suppose to be harder for the computer than with color. I didn't know similar technology is used on low resolution surveillance camera for a while already. Big brother is always watching now I guess.

It will only be a matter of time before large corporation have access to this technology (or maybe they already did). AI in automatic surveillance for market research? Hey it can happen.

Edit: With regard to 85%, it's error margin is actually quite large. Take a room with 100 people, and one of them is on the database. It will catch the one guy 85% of the time, but will also spit out an average of 14.85 false positives. So it's accuracy rate is only about 1 to 17.5 in this given situation. Now put that in the scale of the a stadium, and you can guess how many false positives are being watched just to pull out maybe one guy. To the government this fact doesn't matter beyond the cost of extra surveillance, and the need to hire real people to sort out the true hits from the false positives. But for us the false positive means a 15% chance that we will be closely monitored in public areas for the entirely duration.

Edited by 123savethewhales on 04/12/09 - 12:45 AM

Keep it simple.
123savethewhales
Assistant Professor
Avatar

Usergroup: Members
Joined: Mar 25, 2009

Total Topics: 7
Total Posts: 346
Posted 04/12/09 - 02:12 AM:
quote post
#8
J. Random Hacker wrote:
I think it's a two part process. First we have a heuristic pattern matching of the shape, size, texture and color of cats and dogs. Then based on this analysis, we tentatively assign cat or dog status until disconfirmed by some other cue such as that "cat" suddenly barking or that "dog" hissing. I don't think any of this is difficult for a computer to do once it knows how. Making it plastic enough to learn how to do it on its own is the hard part.

The first type of cat vs dog recognition is actually harder to do than a learning program, since it got to pick out the cat from rest the background, the computer got to know where the cat ends and where background begins. Plus there are really a lot of overlapping when consider what shape, size, texture, and color, when it comes to a cat or dog, making it logically difficult to program in any line that is either cat or dog specific.

As far as learning program, check out the site below. It has descriptions and a free demo to play around with.
http://www.numenta.com/

Keep it simple.
J. Random Hacker
banned
Avatar

Usergroup: Sponsors
Joined: Apr 11, 2009

Total Topics: 12
Total Posts: 479
Posted 04/12/09 - 09:11 AM:
quote post
#9
123savethewhales wrote:
The first type of cat vs dog recognition is actually harder to do than a learning program, since it got to pick out the cat from rest the background, the computer got to know where the cat ends and where background begins.


Recognizing an object can be done by training a neural net with examples. A similar looking animal with different backgrounds is easy to pick out because there is a constant shape that is common between the pictures. A general learning program is a lot harder. That's why there are plenty of image recognition programs. Many are available commercially.
ying
I ♥ Scepticism
Avatar

Usergroup: Moderators
Joined: Feb 20, 2004
Location: Lichtenvoorde, Netherlands

Total Topics: 18
Total Posts: 1485
Posted 04/12/09 - 05:52 PM:
quote post
#10
123savethewhales wrote:

So, what is a cat? How do we rationally tell the difference between a cat and a dog? Is rationality alone enough to distinguish the difference between the two?

These days, it's down to genetics. Used to be phenotype, back in the day (of Aristotle). From a phenomenological perspective though, a cat might just be a dog with a different set of characteristics. Hard to tell, without checking genotype. How do we distinguish between the two though, without invoking (modern) science? Well, you get one thingy which looks and does things differently from another thingy. Then we paste different words onto them, conceptualising the phenomenological objects. Easy, really, and not much rationality going on, there. grin

"I determined nothing."
-Sceptical expression
Download thread as

Page: 1 2



Sorry, you don't have permission to post. Log in, or register if you haven't yet.