Wednesday, January 02, 2008
The Need for a Client View of Context-Independent Symbols
In the paper Interactively converging on context-sensitive representations: A solution to the frame problem by Robert M. French and Patrick Anselme, the authors describe a robotic simulation of context-sensitive representations:
"Agre and Chapman (1987) developed a simulation, Pengi, in which an agent — a penguin — makes use of context-dependent representations in order to avoid being attacked, in this case, by a bee. Piles of ice cubes are used to allow Pengi to protect himself from being stung by the bee. In addition, Pengi can fight back against the bee by means of a well-placed kick to any ice cube that the bee happens to be directly behind. As for the bee, it has two means to kill the penguin: either by striking an ice cube that Pengi is hiding behind or by stinging the penguin. Space does not allow a full development here of the strategies that the penguin can adopt, nor of his actual sensori-motor abilities (for a detailed description, see Agre, 1997). It is, however, important to note that the penguin has well-developed attentional capabilities that allow him to focus on the salient factors of the situation in which he finds himself.
In the Pengi simulation, all events are contextually determined. There are not context-independent representations, like “ice-cube wall” or even “bee.” The penguin’s representations are, to use Agre’s term, deictic, meaning that they depend on the circumstances in which they are used. Pengi’s representations take the form of the-wall-behind-which-I-will-hide or the-bee-I-want-to-kill (Agre, 1997, p. 267). These representations do not describe context-independent entities; rather they describe some aspect of the environment that is in a particular relation to the agent at a particular moment in time. In other words, rather than a context-independent “bee” representation, the system produces a representation for a particular bee in a particular place at a particular time. The Pengi model avoids the frame problem because the agent is manipulating context-sensitive representations for “bee” that include various salient aspects of situation at hand."
I suspect that biological agents show this kind of continuous context-sensitivity in their sense of the referents of symbols. As the "Great Global Graph" of linked data on the web grows, our automated agents will need to show this kind of selectivity as well. But rather than over-use the word "context", perhaps a better term here would be "personalization". So each clients view of the GGG, the world wide graph of data, will be different, skewed towards the needs of that client. This will probably be done by a collaboration between new kinds of search agents and each client themselves.
"Agre and Chapman (1987) developed a simulation, Pengi, in which an agent — a penguin — makes use of context-dependent representations in order to avoid being attacked, in this case, by a bee. Piles of ice cubes are used to allow Pengi to protect himself from being stung by the bee. In addition, Pengi can fight back against the bee by means of a well-placed kick to any ice cube that the bee happens to be directly behind. As for the bee, it has two means to kill the penguin: either by striking an ice cube that Pengi is hiding behind or by stinging the penguin. Space does not allow a full development here of the strategies that the penguin can adopt, nor of his actual sensori-motor abilities (for a detailed description, see Agre, 1997). It is, however, important to note that the penguin has well-developed attentional capabilities that allow him to focus on the salient factors of the situation in which he finds himself.
In the Pengi simulation, all events are contextually determined. There are not context-independent representations, like “ice-cube wall” or even “bee.” The penguin’s representations are, to use Agre’s term, deictic, meaning that they depend on the circumstances in which they are used. Pengi’s representations take the form of the-wall-behind-which-I-will-hide or the-bee-I-want-to-kill (Agre, 1997, p. 267). These representations do not describe context-independent entities; rather they describe some aspect of the environment that is in a particular relation to the agent at a particular moment in time. In other words, rather than a context-independent “bee” representation, the system produces a representation for a particular bee in a particular place at a particular time. The Pengi model avoids the frame problem because the agent is manipulating context-sensitive representations for “bee” that include various salient aspects of situation at hand."
I suspect that biological agents show this kind of continuous context-sensitivity in their sense of the referents of symbols. As the "Great Global Graph" of linked data on the web grows, our automated agents will need to show this kind of selectivity as well. But rather than over-use the word "context", perhaps a better term here would be "personalization". So each clients view of the GGG, the world wide graph of data, will be different, skewed towards the needs of that client. This will probably be done by a collaboration between new kinds of search agents and each client themselves.
Sunday, December 30, 2007
The Semantic Web Meets the Abundance Problem
The paper Authoritative Sources in a Hyperlinked Environment (2002 PDF) by Jon Kleinberg proposes the following definition of the problem of efficiently and reliably finding documents on the web out of the billions available, a problem he names the Abundance Problem: "The number of pages that could reasonably be returned as relevant is far too large for a human user to digest. To provide effective search methods under these conditions, one needs a way to filter, from among a huge collection of relevant pages, a small set of the most "authoritative" or "definitive" ones."
As the semantic web will build a much more granular set of linked data than the current set of linked documents, this problem will only become worse. So even as search is improved by having more detailed and accurate semantic annotations added, the problem of finding the most authoritative and definitive linked-data will increase. Look for the semantic web version of page-rank algorithms. We will be just as much in need of datum-rank algorithms perhaps?
Swoogle may already be doing this.
As the semantic web will build a much more granular set of linked data than the current set of linked documents, this problem will only become worse. So even as search is improved by having more detailed and accurate semantic annotations added, the problem of finding the most authoritative and definitive linked-data will increase. Look for the semantic web version of page-rank algorithms. We will be just as much in need of datum-rank algorithms perhaps?
Swoogle may already be doing this.
Saturday, December 29, 2007
Solutions to the Frame Problem
In the paper Global Abductive Inference and Authoritative Sources, or, How Search Engines Can Save Cognitive Science* by Andy Clark 2002 (PDF), Clark presents a problem that looks very much like the one I mention in Words (or URI) as Locations in the Fabric of Context, but he refers to it has the "Frame Problem", and attributes its origin with John McCarthy and and Pat Hayes. Clark also presents a potential, partial solution related to searching on the web.
Here is the Abstract:
"Kleinberg (1999) describes a novel procedure for efficient search in a dense hyper-linked environment, such as the world wide web. The procedure exploits information implicit in the links between pages so as to identify patterns of connectivity indicative of “authorative sources”. At a more general level, the trick is to use this second-order link-structure information to rapidly and cheaply identify the knowledge-structures most likely to be relevant given a specific input. I shall argue that Kleinberg’s procedure is suggestive of a new, viable, and neuroscientifically plausible solution to at least (one incarnation of) the so-called “Frame Problem” in cognitive science viz the problem of explaining global abductive inference. More accurately, I shall argue that Kleinberg’s procedure suggests a new variety of “fast and frugal heuristic” (Gigerenzer and Todd (1999)) capable of pressing maximum utility from the vast bodies of information and associations commanded by the biological brain. The paper thus takes up the challenge laid down by Fodor ((1983)(Ms)). Fodor depicts the problem of global knowledge-based reason as the point source of many paradigmatic failings of contemporary computational theories of mind. These failings, Fodor goes on to argue, cannot be remedied by any simple appeal to alternative (e.g. connectionist) modes of encoding and processing. I shall show, however, that connectionist models can provide for one neurologically plausible incarnation of Kleinberg’s procedure. The paper ends by noting that current commercial applications increasingly confront the kinds of challenge (such as managing complexity and making efficient use of vast data-bases) initially posed to biological thought and reason."
My fear was that "the meaning of each term is contained in other documents, and the meaning of each of those terms is again contained in yet more documents and so on until, in the end, in order to understand any URI, you must import some extremely large subset of the entire semantic web."
Here is Clark on the the problem: "The problem of global abductive inference is best appreciated against this backdrop. It is the problem, in a super-compressed nutshell, of how to do inference to the best explanation in a way that is sensitive to whatever is most relevant in the massive body of belief and knowledge that underlies commonsense thinking and reasoning."
Here is the Abstract:
"Kleinberg (1999) describes a novel procedure for efficient search in a dense hyper-linked environment, such as the world wide web. The procedure exploits information implicit in the links between pages so as to identify patterns of connectivity indicative of “authorative sources”. At a more general level, the trick is to use this second-order link-structure information to rapidly and cheaply identify the knowledge-structures most likely to be relevant given a specific input. I shall argue that Kleinberg’s procedure is suggestive of a new, viable, and neuroscientifically plausible solution to at least (one incarnation of) the so-called “Frame Problem” in cognitive science viz the problem of explaining global abductive inference. More accurately, I shall argue that Kleinberg’s procedure suggests a new variety of “fast and frugal heuristic” (Gigerenzer and Todd (1999)) capable of pressing maximum utility from the vast bodies of information and associations commanded by the biological brain. The paper thus takes up the challenge laid down by Fodor ((1983)(Ms)). Fodor depicts the problem of global knowledge-based reason as the point source of many paradigmatic failings of contemporary computational theories of mind. These failings, Fodor goes on to argue, cannot be remedied by any simple appeal to alternative (e.g. connectionist) modes of encoding and processing. I shall show, however, that connectionist models can provide for one neurologically plausible incarnation of Kleinberg’s procedure. The paper ends by noting that current commercial applications increasingly confront the kinds of challenge (such as managing complexity and making efficient use of vast data-bases) initially posed to biological thought and reason."
My fear was that "the meaning of each term is contained in other documents, and the meaning of each of those terms is again contained in yet more documents and so on until, in the end, in order to understand any URI, you must import some extremely large subset of the entire semantic web."
Here is Clark on the the problem: "The problem of global abductive inference is best appreciated against this backdrop. It is the problem, in a super-compressed nutshell, of how to do inference to the best explanation in a way that is sensitive to whatever is most relevant in the massive body of belief and knowledge that underlies commonsense thinking and reasoning."
Friday, December 28, 2007
The Semantic Conception of an Information Resource
From Section 4 of Tarski's 1944 paper, hosted on John Sowa's website, titled The Semantic Conception of Truth: "the fundamental conventions regarding the use of any language require that in any utterance we make about an object it is the name of the object which must be employed, and not the object itself. In consequence, if we wish to say something about a sentence, for example, that it is true, we must use the name of this sentence, and not the sentence itself."
The same can be said about the web: to say of a web document that it is an information resource, for example, it is the name (the URI) of the object which must be employed, and not the object itself. What the current version of the web architecture seems to be saying is that web objects, as opposed to everything else, say what they are themselves. In other words, to make utterances about a web object, you use the web object itself (including the 200 OK), and not its name. This would be like insisting that, when talking about a car, you had to insert the car itself in your speech somehow, as the subject of all propositions about it. This, of course, would seem to have the great advantage that everyone would always be referring to the same car - unambiguously - regardless of any local context. But actually, I think this is an illusion. Objects can't insert themselves into ontological categories. Agents use names to put categories around objects like cars or web documents for particular purposes. The universe just exists, agents say what objects are by using language. So I don't agree it would break the web for each utterance of the name (URI occurrence) of some web object to be used for any useful purpose, even if different from other uses, all while HTTP returns the same object for the same URI.
But even as I try to wrap this up and hit the post button, I see problems with this. By analogy, for example, the fly (object) buzzing around in front of the frog reflects light that physically changes the frogs vision transducers in a distinct, fly-like way. When the frog processes this as a fly, that is, when the frog's brain puts it in the 'dinner' category, it appears very much like it is the very essence of the fly, the fly's fly-ness that does so. Any frog that arbitrarily put it in the 'mate' category, for example, would lead a very short life.
The same can be said about the web: to say of a web document that it is an information resource, for example, it is the name (the URI) of the object which must be employed, and not the object itself. What the current version of the web architecture seems to be saying is that web objects, as opposed to everything else, say what they are themselves. In other words, to make utterances about a web object, you use the web object itself (including the 200 OK), and not its name. This would be like insisting that, when talking about a car, you had to insert the car itself in your speech somehow, as the subject of all propositions about it. This, of course, would seem to have the great advantage that everyone would always be referring to the same car - unambiguously - regardless of any local context. But actually, I think this is an illusion. Objects can't insert themselves into ontological categories. Agents use names to put categories around objects like cars or web documents for particular purposes. The universe just exists, agents say what objects are by using language. So I don't agree it would break the web for each utterance of the name (URI occurrence) of some web object to be used for any useful purpose, even if different from other uses, all while HTTP returns the same object for the same URI.
But even as I try to wrap this up and hit the post button, I see problems with this. By analogy, for example, the fly (object) buzzing around in front of the frog reflects light that physically changes the frogs vision transducers in a distinct, fly-like way. When the frog processes this as a fly, that is, when the frog's brain puts it in the 'dinner' category, it appears very much like it is the very essence of the fly, the fly's fly-ness that does so. Any frog that arbitrarily put it in the 'mate' category, for example, would lead a very short life.
Wednesday, December 12, 2007
Neither Brain nor Ghost - Book Chapter Summaries
Neither Brain nor Ghost - Book Chapter Summaries by Teed Rockwell.
In the post, Words (or URI) as Locations in the Fabric of Context, I speculated that a full interpretation of a name (or word, or URI in the semantic web) would require access to a web of associations extending out very far in computational space. But Teed Rockwell, in his book, Neither Brain nor Ghost, points to a solution.
"However, if the self is embodied by the brain/body/world nexus, rather than by the brain alone, there is no need for the world to get inside the head in order for the self to be aware of it. However, in order to avoid the vacuously mystical claim that “we are one with everything” , we need a very specific and technical definition of “world”. It would be trivial to claim that the entire causal nexus responsible for a mental state embodies a mental state. But once we see “world” as a biological concept defined by function, it can have borders that go beyond the brain without encompassing so much as to be meaningless or trivial. Every organism has a symbiotic relationship with specific aspects of reality in its immediate spatial vicinity. It is these elements which the science of ecology calls an organism’s environment , and which Heidegger and others have called a creature’s “umwelt”."
Then with respect to the problem of context, the extent to which attention to the web of associations needs to extend is limited by the purpose of the agent. Just as he does with "world", once we see "context" as "a biological concept defined by function", then its limits are potentially computable.
In the post, Words (or URI) as Locations in the Fabric of Context, I speculated that a full interpretation of a name (or word, or URI in the semantic web) would require access to a web of associations extending out very far in computational space. But Teed Rockwell, in his book, Neither Brain nor Ghost, points to a solution.
"However, if the self is embodied by the brain/body/world nexus, rather than by the brain alone, there is no need for the world to get inside the head in order for the self to be aware of it. However, in order to avoid the vacuously mystical claim that “we are one with everything” , we need a very specific and technical definition of “world”. It would be trivial to claim that the entire causal nexus responsible for a mental state embodies a mental state. But once we see “world” as a biological concept defined by function, it can have borders that go beyond the brain without encompassing so much as to be meaningless or trivial. Every organism has a symbiotic relationship with specific aspects of reality in its immediate spatial vicinity. It is these elements which the science of ecology calls an organism’s environment , and which Heidegger and others have called a creature’s “umwelt”."
Then with respect to the problem of context, the extent to which attention to the web of associations needs to extend is limited by the purpose of the agent. Just as he does with "world", once we see "context" as "a biological concept defined by function", then its limits are potentially computable.
Monday, December 10, 2007
Name, Sense, Reference, Attention, and the World
A name does not move from place to place. There is no 'Transfer of information'. However, signals are transfered, by transducers. There is no information before the sense of it occurs. A sign becomes a name with a reference when it coincides with a shift of attention towards an object in the world. The name, the object, and the shift of attention to the object - these three all happen at once, it is only a convenience for us to separate them. They are really a unit. Another way to think about this would be to say that the shift of attention to the object is the reference of the name.
Attention to an object in the world is unique to that object. Attention to a red apple is different from attention to a blue drinking glass. So each name is formed by the qualities of the world.
Attention to an object in the world is unique to that object. Attention to a red apple is different from attention to a blue drinking glass. So each name is formed by the qualities of the world.
Tuesday, November 27, 2007
Nomen est numen - To Name is to Know
To name is to know. Paglia extends this with, " to know is to control."
From Wikipedia's Name article, we hear from Shakespere on names,
Tis but thy name that is my enemy;
Thou art thyself, though not a Montague.
What's Montague? it is nor hand, nor foot,
Nor arm, nor face, nor any other part
Belonging to a man. O, be some other name!
What's in a name? that which we call a rose
By any other name would smell as sweet;
Naming Conventions (Programming)
From Pat Hayes on IKL: " if names be not correct, language is not in accordance with the truth of things. If language be not in accordance with the truth of things, affairs cannot be carried on to success, ... What the superior man requires is just that in his words there may be nothing incorrect." Confucious, Analects XIII, 3
From Wikipedia's Name article, we hear from Shakespere on names,
Tis but thy name that is my enemy;
Thou art thyself, though not a Montague.
What's Montague? it is nor hand, nor foot,
Nor arm, nor face, nor any other part
Belonging to a man. O, be some other name!
What's in a name? that which we call a rose
By any other name would smell as sweet;
Naming Conventions (Programming)
From Pat Hayes on IKL: " if names be not correct, language is not in accordance with the truth of things. If language be not in accordance with the truth of things, affairs cannot be carried on to success, ... What the superior man requires is just that in his words there may be nothing incorrect." Confucious, Analects XIII, 3
