You are not logged in.
Pages: 1
I was thinking about adjectives like big. When we say something is big is it inherently big or big with respect to some set. Like if I say a big cat, do I mean a lion or a cat that is bigger then the average house cat? Is this ambiguity a strength or a weakness in languages? What challenge do such ambiguities present for computers dealing with languages. What kind of solutions do people foresee?
Dig into the [url=http://child-civilization.blogspot.com/2006/12/political-grab-bag.html]political grab bag[/url] at [url=http://child-civilization.blogspot.com/]Child Civilization[/url]
Offline
Like button can go here
I was thinking about adjectives like big. When we say something is big is it inherently big or big with respect to some set. Like if I say a big cat, do I mean a lion or a cat that is bigger then the average house cat? Is this ambiguity a strength or a weakness in languages? What challenge do such ambiguities present for computers dealing with languages. What kind of solutions do people foresee?
Minsky talks a bit about this in _Society of Mind_
http://www.amazon.com/Society-Mind-Marv … 0671657135
He blackboxes the decision with a _More_ agent (i.e., decides "is greater than") that is located in a network of agents (the whole network of agents is the society of mind, agents can be complex or single-neuron simple). There might be just one _More_ agent but there are probably many of them - maybe no single agent can compare apples and oranges, maybe it can. The comparison agent _More_ may be different from, but linked to, the language agent _More_ (observe each of listening, speaking, reading and writing seem to eventually connect to the same language agent). Each individual mind may structure the network differently.
The actual decision more/less, big/normal, etc can, by all accounts, be well modeled with Bayesian networks with such agents as nodes. Bayesian -> probabilistic decision making ~ ambiguity, "fuzzy logic."
Some people say that this is why a computer will never truly understand human language - because it is not embodied in a human body. What is "big" on a human scale maybe insignificantly small on some scales and inconceivably large on others. And it gets even weirder when you move on to concepts more abstract than big and those intimately tied to human emotion (insert _Blade Runner_ quote). To buy this you probably need to privilege human consciousness as Searle does in his famous Chinese Room thought experiment ...
http://en.wikipedia.org/wiki/Chinese_room
Automatic language translation is where the state of the art is for this stuff. So you can see we've still got a ways to go. To get useful machine translation, it looks like you need to model the human experience pretty deeply. It may be that there will emerge an intermediate language between computer programming languages and natural languages that will be less ambiguous/human-centric. Lots of people are working in that direction.
Fan of [url=http://www.red-oasis.com/]Red Oasis[/url]
Offline
Like button can go here
It has always intrigued me that human beings are midway in size, magnitude-wise, between the Universe and the so-called particles that compose it. Could it be that conscience, self-aware intelligence needs to fall within this size region in order to function?
Offline
Like button can go here
I’m curious. How interested are people here in artificial intelligence and databases? I wanted to learn some stuff about Microsoft Access and artificial intelligence. I started creating a relational database:
http://www.geocities.com/s243a/iq/iq1.mdb
http://www.geocities.com/s243a/iq/iq1.zip
There is not much in the database right now and I am working on some forms to make data entry easier. The idea of the database is to only have one peace of information per table. Thus one table could be a table of types, another table of words and another table a table of links. Each row of each table has a unique index so you can refer to the table and row with only a single number. This makes it easier to express relationships and take advantage of the relationship features of Microsoft Access.
First I want to express relationships between data. Like for instance a cat is an animal or a cat is a noun. Once a bunch of relationships are built into the database I want to see if I can ask it questions and get back intelligent answers. Anyway, I can explain my design choices or take design suggestions if anyone is interested. That is I don’t mind being the teacher student or both.
Dig into the [url=http://child-civilization.blogspot.com/2006/12/political-grab-bag.html]political grab bag[/url] at [url=http://child-civilization.blogspot.com/]Child Civilization[/url]
Offline
Like button can go here
It has always intrigued me that human beings are midway in size, magnitude-wise, between the Universe and the so-called particles that compose it. Could it be that conscience, self-aware intelligence needs to fall within this size region in order to function?
I am not sure how the scale of size of a life form affects conciseness. I could see there being a minimum size but as for maximum size I see it not as a limitation but as a reason for greater division between thought processes. Perhaps as the size of a mind increases it becomes more of a collective then a mind for reasons of efficiency.
Dig into the [url=http://child-civilization.blogspot.com/2006/12/political-grab-bag.html]political grab bag[/url] at [url=http://child-civilization.blogspot.com/]Child Civilization[/url]
Offline
Like button can go here
First I want to express relationships between data. Like for instance a cat is an animal or a cat is a noun. Once a bunch of relationships are built into the database I want to see if I can ask it questions and get back intelligent answers.
You might be interested in Douglas "Intelligence is 10 million rules" Lenat's Cyc project.
Fan of [url=http://www.red-oasis.com/]Red Oasis[/url]
Offline
Like button can go here
It has always intrigued me that human beings are midway in size, magnitude-wise, between the Universe and the so-called particles that compose it. Could it be that conscience, self-aware intelligence needs to fall within this size region in order to function?
There's a theory that says intelligence is inevitable once you have sufficient complexity to allow evolution. You haven't really left much room outside of the region you defined (whole universe to constituent particles), but I imagine that the vibrating multidimensional membranes of M-theory are complex enough to support evolution.
Fan of [url=http://www.red-oasis.com/]Red Oasis[/url]
Offline
Like button can go here
Minsky talks a bit about this in _Society of Mind_
http://www.amazon.com/Society-Mind-Marv … 0671657135
He blackboxes the decision with a _More_ agent (i.e., decides "is greater than") that is located in a network of agents (the whole network of agents is the society of mind, agents can be complex or single-neuron simple). There might be just one _More_ agent but there are probably many of them - maybe no single agent can compare apples and oranges, maybe it can. The comparison agent _More_ may be different from, but linked to, the language agent _More_ (observe each of listening, speaking, reading and writing seem to eventually connect to the same language agent). Each individual mind may structure the network differently.
I’ve heard about multi agents before. A student that had the same supervisor as me was doing his thesis on multi agent systems. He was building a system for an oil company to combine multiple fault detection techniques. I don’t think human’s work in this fashion because personally I can’t concentrate on very many things at once. Although the human mind has a highly parallel architecture.
Today I was thinking about how AI could be applied to a diet program. I was thinking how on packaging they list the ingredients in food but they don’t tell you how much of each ingredient. However they give the composition of the food in terms of vitamins and if you assume the composition of the food in terms of vitamins is made up of a linear combination of the ingredients you should be able to estimate how much of each ingredient there is.
One might try to solve this problem with matrix inversion but some ingredients with virtually no nutritional value will lead to an ill conditioned matrix inversion. Say you have access to visual basic, MATLAB and not a lot of money to by extra packages for visual basic. You might write a short script in visual basic to solve the problem based on ridge regression because it keeps you having to worry about numerical issues like pivoting.
You might write one or more complex techniques in MATLAB to deal with the problem because it has built in algorithms created by experts in their field. Now the visual basic algorithm you can probably run as a thread in the background or maybe it will be fast enough that you won’t even to bother to run it as a thread. It will give you a crude quick answer and let you proceed.
MATLAB is a large process and you might want to access it via a server on another computer. It can only handle one process at a time and two instances of MATLAB running on one computer don’t share resources well. IF the algorithm in MATLAB is sufficiently more complex it might take a while and even if it runs quick, the MATLAB resource could be tied up solving a more difficult problem that you previously requested it solve.
As a consequence you might have to wait for the answer. The main program can do two things. It can create an ID representing the problem and each answer can be uniquely stored and linked to the question so it can be used when needed. With this kind of architecture you can use the crude answers to provide early solutions to problems and will have access to better answers latter to improve the solution. Thus in my diet example as you enter in new food information, you can get a quick answer of what you ingested and later in the day if you are curious you can check to see if more sophisticated techniques changed that answer much.
Another point is have the program set up to use several techniques will allow people to utilized what packages they have available. Not everyone may have access to complex software packages running on other computers so if these auxiliary systems aren’t configured the software will still be able to function and return an answer.
Dig into the [url=http://child-civilization.blogspot.com/2006/12/political-grab-bag.html]political grab bag[/url] at [url=http://child-civilization.blogspot.com/]Child Civilization[/url]
Offline
Like button can go here
First I want to express relationships between data. Like for instance a cat is an animal or a cat is a noun. Once a bunch of relationships are built into the database I want to see if I can ask it questions and get back intelligent answers.
You might be interested in Douglas "Intelligence is 10 million rules" Lenat's Cyc project.
Here is an odd piece of Tria:
"Big Cyc ("Cyc" is Polish for "Tit") is a Polish rock band formed in 1988.
The band is well-known in Poland for their controversial behaviour. The cover of their first album, Z partyjnym pozdrowieniem (Polish for With a Party greeting), was an image of Vladimir Lenin with a Mohawk hairstyle. The title of their second album, Nie wierzcie elektrykom (Don't believe electricians) refers to the Polish president Lech Wałęsa, an electrician by education. Their fourth album cover, Wojna plemników (War of spermatozoons) featured a nun drying condoms on a clothes line. In May 1999, the band leader Krzysztof Skiba was charged with indecent exposure and fined the equivalent of $308 for mooning the Polish prime minister Jerzy Buzek during a festival in February 1999."
Perhaps Cyc is the mother of AI.
http://en.wikipedia.org/wiki/Big_Cyc
Dig into the [url=http://child-civilization.blogspot.com/2006/12/political-grab-bag.html]political grab bag[/url] at [url=http://child-civilization.blogspot.com/]Child Civilization[/url]
Offline
Like button can go here
Researchers Use Wikipedia To Make Computers Smarter
http://www.physorg.com/news87276588.html
Fan of [url=http://www.red-oasis.com/]Red Oasis[/url]
Offline
Like button can go here
Pages: 1