Know Ledge
Know LedgeKnowledge is power. -- Locke
Data that can be interpreted is called information. When that information can be placed in relevant context it is called knowledge. A message (finite signal) can contain data, information or knowledge depending largely on how it is processed. This rank ordering of terms can be extended on both ends to form the following hierarchy ...
A (new?) idea about it. Information is disposable (in all meanings of the term), whereas knowledge should be sustainable. This could be a guideline to build tools that make distinct DisposableInformation versus SustainableKnowledge. See some developments of that idea at: http://www.universimmedia.com/sk.htm
Traffic analysis is the process of studying patterns in messages that cannot otherwise be understood. For example, in warfare, if enemy radio traffic increases sharply after bombing a particular location then it is probably worth bombing some more. Similarly if radio traffic is silenced then one has probably bombed enough. Silence can be considered a message in this context (as pointed out below with an appeal to information theory) but a better name for the type of message would be pattern, as in recognizable sequence or correlation.
(Note: on this site the word pattern often refers to a possible element of a PatternLanguage which is a message located in the above hierarchy somewhere around knowledge or wisdom.)
We should revisit this spectrum. Silence between noise and data does not make much sense. If noise is one end of the spectrum, hence silence is the other one, therefore beyond wisdom. Wisdom is the path from knowledge to silence.
Go ask Mozart, J.S. Bach, John Cage, and many more.
Noise > Data > Information > Knowledge > Wisdom > Silence ... a progression towards getting rid of what is useless ... that's what interpretation is all about.
-- CrisBer - December 3, 2001
Are we wasting our time with SETI? Or is the sequence Silence > Noise > Data > Information > Knowlegde > Wisdom
Or perhaps Silence > Noise > Silence > Information > Silence > Data > Silence > Knowledge > Silence > Wisdom > Silence?
IMHO, Noise -> Silence -> Data... works. Silence is some information, because you know that nothing is there, while noise hides even that fact.....
deliberate silence is aka "pause" deliberate noise is aka "signal"
combination of pause and signal is "information"
Perhaps this applies:
People can only hear (a) if they are willing, (b) if they have the preconditions for understanding.
I saw a FarSide cartoon today with scientists studying the vocalizations of dolphins, recording each with a tally mark. Scientist says to other scientist, "I got another 'ah blah es pan yol' sound here." Other sounds on the board are 'booen os dee os', etc. The scientists have (a) but not (b). You are trying to keep (a) alive while building toward (b). -- AlistairCockburn
It would be good if people could add links here to writers that have embraced this notion of a spectrum of knowledge.
It is interesting to consider the sorts of activities that "bridge" the terms in the spectrum. For example, some time ago some person explained to me that it takes interpretation for "data" to become "information", and that It takes experience for information to become knowledge, and for knowledge to become wisdom, it was something like being able to stand the test of time.
Any good epistemologists among us??
I don't find "wisdom" to be a very useful concept. There are a fair number of possible reasonable-sounding definitions, but I've never seen a reason to prefer one over another, or to even use the word except when suggesting that a source should be trusted. Wow, metawisdom. -- DanielKnapp
I agree, from the list above, the distinctions between data, information and knowledge are the most important. It irks me when they get treated as synonyms.
How about this one: There is no true noise or silence on the path to wisdom:
Data extends Object Noise extends Data Silence extends Data implements Information Knowledge extends Data implements Information Wisdom extends Knowledge
'Wisdom' is not well defined; the closest one can come to it is a "generally correct, assumptive prediction". Understanding answers a 'Why?' question. The closest I can tell, Wisdom is simply knowing how to answer a whole lotta 'Whys'.
I've been studying knowledge for the last two months as preliminary to designing a (truly) generalized, distributed learning system (for everything from Starcraft to Storytelling) and I've come to the conclusion that knowledge, itself, is very multi-dimensional; one cannot quantify it on the sort of one-dimensional ladder presented above. Or, more accurately, one may quantify it as such, but the consequences to both efficiency and efficacy will be both profound and negative.
I made the Solipsist viewpoint fundamental to my own study of knowledge and the learning process. This is a very sound philosophical and mathematical foundation, solipsism being tautological. Solipsism is not nihilism.... It is merely complete acceptance of the fact that it is impossible to deductively prove the existence of an external world or anything about it. The best one can do is make inductive proofs on the signals one receives and, with a little faith, assume that they are coming from an 'external' environment.
With a Solipsist foundation, there are fundamentally only two actions: send signal and receive signal. Any 'boxed' process or agent may fundamentally only process incoming signals and attempt to deliver signals to (a not-provably existent) external environment. The only reason to do the latter is in an attempt to influence future incoming signals, ideally in a predictable manner. This is as true for you and me as it is for any process in a computer. By 'boxed' process, I refer to conceptually drawing a nice, big box around any set of processes. The result is a sort of black box that can accept incoming messages on some vectors and sends messages outwards (into an environment) on other vectors. Any process without any incoming or outgoing signals may be described as a value as it is timeless and functional, having neither dependency on environment nor environmental side-effects... by definition. The empty set of processes is also a value by definition, and is the void value in particular.
Messages traveling to and from these boxed processes are always well-typed values with well-known representations. However, there is a certain arbitrariness to the assumptive 'level' of this type... i.e. whether one knows it as goban representations vs. integers vs. mere bit-strings vs. analog noise. 'Continuous' messages (such as analog signals) must be somehow stored if one is to process them... i.e. as a discretized or piecewise-defined stream of (value,time) pairs.
If knowledge is to exist meaningfully, it must exist within a boxed process. This might not be precisely what you or I think of as knowledge (e.g. one could box two people and refer to all the knowledge in that system, even though it isn't well shared between them). However, this does place a limit on the bounds of knowledge... and it is just as relevant to speaking of cultural knowledge and such as it is to knowledge of an individual brain or even just part of that brain.
Anyhow, there are four fundamental components of knowledge:
(1) Data -- typed values represented internally with intrinsic or contextual meaning (as propositions)
and held with some level of confidence. In a skeptical learning system, data should be
sourced... because some sources aren't reliable (i.e. they lie, they might be compromised,
etc.)
(2) Pattern -- A function f from A->Bool (or A->Confidence, or A->Fuzzy), where A is a collection of
knowledge. Patterns, themselves, aren't useful... and learning new, useful patterns
necessarily requires a large data set. This function must come from somewhere, too...
and is one of the more interesting problems to solve. Neural networks, genetic programming,
and human support may create them. A fitness function naturally exists in a learning
environment -- the function's value in producing useful abstractions.
(3) Information -- A piece of knowledge that says a particular pattern or abstraction holds over some other
particular collection of knowledge. To gain information naturally requires processing of
data, including other information (thus information output may always be data input).
Information is generally held with non-unitary confidence, unless it is tautological in
nature. Efficient capture of information requires being able to rapidly determine which
patterns and abstractions to test within a particular collection of data, which implies
that this must be learned, too, and that a truly outstanding indexing system must exist.
(4) Abstraction -- an inductive (and therefore fallible) observation that some instantiated patterns accurately
(with some degree of confidence) predict the existence of some other instantiated
patterns, possibly with contingencies or probabilities (i.e. "f(A) and g(B) predicts h(C)
or j(D)" for variable knowledge-sets A,B,C, and D). This is fundamental to the inference
system. Also, while it isn't fundamental, Abstraction itself may be of a sort to
describe exo-patterns "f(A) predicts g(B)" with some set of constraints between A
and B or endo-patterns "f(A) predicts g(A)". Both sorts of abstractions are quite valuable.
Abstractions (not information) are the true heart of knowledge. They provide a sort of semantic compression that is used for analytical, constructive, and communicative purposes. All words in the English language constitute abstractions (generally of endo-patterns). If one discusses "apple", one is discussing the patterns predicted by the hypothetical existence of an apple -- taste, shape, color, structure, origin, and destination. To observe the supposed existence of an "apple", meanwhile, requires identifying certain patterns -- which requires processing data to create information -- about a particular instance. If one knows it is tart and came from a tree, one might suspect it is fruit... perhaps a not-quite-ripe plum or apple. If one adds that it has a shape that is peculiarly apple, then one may believe, with pretty good confidence, that a particular instance is an apple. That is information. Used constructively, ask a machine (or human) to draw an image of an apple, and that machine (or human) will probably draw one of many possible facsimiles based on what the abstraction "apple" implies wrgt patterns regarding its image.
To communicate efficiently, two distributed knowledge systems that have made contact with one another must identify commonly known, provably useful abstractions and create common names for them. (This can be done with low-level data analysis and such... but isn't trivial.) Afterwards, any 'discussion' between two systems is simply building concepts in one another via the use of known abstractions. This entire discussion may later be named and treated as data... creating books and such... and whole books may later be abstracted, if many people have discussions about the same sort of thing. This 'distributed' part of the generalized learning system has been causing me headaches in my last two weeks of study, but I think that developing something like a Robopedia or Robowiki (where the learning systems can cooperate on identifying commonly known, useful abstractions then share this information with others and offer criticism and more faithful abstractions) shall be a good start. Essentially, they need to develop their own language on the fly.
One should note that Abstractions, Patterns, and Information, too, must be typed values that are somehow represented internally. They are not -necessarily- part of the knowledge base and learning system, but they -should- be.
Making Abstractions, Patterns, and Information into 'Data' has many advantages... such as allowing the learning system to experiment with new patterns, new abstractions, and new types of information, in order to find find patterns of patterns, find new patterns that are easier to calculate and nearly as faithful, etc. All these things are critical to developing various sorts of 'understanding' in a learning system... the ability to learn to make accurate, predictive assumptions whereby a few small observances allows one to accurately predict other patterns one should expect to find. Similarly, whenever an inference is performed, that entire inference tree should also be simply turned into data and dropped into the knowledge-base in order to allow for processing, abstraction, and optimization of the regular inference paths. This allows the learning system to determine where it can make logical leaps with reasonable confidence, allowing it to gain massive speed in the thinking process. The fitness functions for these things may be quantified for both speed and fidelity... and these same functions apply to assumptive leaps in both inference and in pattern identification.
'Fidelity' is all about avoiding 'surprise' and may always be checked by auditing the forward leaps via: backwards chaining to unused evidence in the forward leap, plus via accurate prediction of likely future signals, possibly in response to signals of one's own. (An example of the latter might be asking: "Am I correct to assume ____", and correctly predicting the answer to be "yes"). The fitness of speed is more obvious.
As a note, any successful answer to 'Why?' implies understanding, and answering 'why' requires that assumptive leap-forward to an abstraction that implies constraints on a system. "Why do the two people sit at the goban?" A generalized learning system should come up with a whole set of mostly-correct answers, based on its experience: "To test each other at the game of (Go | Connect4 | etc.)" or "To experience the human concept of 'fun'". A generalized learning system might not be able to word these answers in English (since it might not have learned to communicate with English), but the answers it arrives at should be useful in predicting future signals, making sense of present signals, and likely predicting any past signals (e.g. if the AI starts observing halfway through the game, the constraint of 'why' is necessary for constructive backwards chaining of likely previous moves... but to have a 'why', any intelligence (be it artificial or not) must make assumptions that the stones are placed the way they are for a reason, not randomly.)
Thus knowledge plus assumptive inference leads to wisdom.
So what was it I speak of regarding 'more than one dimension'? Well, primarily I reference the different sorts of data that one should use in creating a knowledge-base will grow upon each-other in a non-linear and somewhat orthogonal fashion, with utility generally rising as abstractions and information reach higher levels (e.g. "collection of pixels" -> "video-image of goban and players" -> "conceptualized board with predicted future moves, possibly with other data on who is playing whom, who is winning, and by how much"). At the lowest level, the signals themselves constitute data. Add to that patterns, information, abstractions, and prior inferences (along with their success ratings and speeds), and one has some more data at a conceptually 'higher level', but that level on something of four or five different dimensions... with 'information on signal data' not following the same branch as 'information on pattern data'.
Oh, and the creation of all these patterns, and abstractions including testing their utility via creation of abstractions, plus ridding the knowledge base of unnecessary data that may now rapidly be predicted via fast, faithful abstraction... should be a separate process from the expert systems that require the inferences. Just have it run at some priority, continuously, possibly at idle. That is, ultimately, the 'learning process'. The 'expert system' that uses the system, though, might be responsible for adding data, information, and new 'inference tree dumps'... possibly with varying priority and auditing to speed things up (e.g. a learning-robot-eyes-camera-expert-system might prioritize and send the highest level information, such as recognized faces, as opposed to lower level information... as a means of semantic compression of the video-stream to reduce both processing-costs and bandwidth costs upstream. Anyone who can understand this semantic stream can reproduce the image with a /known/ fidelity via constructive use of abstractions, much like a person can reproduce with some predictable fidelity the basics of a situation. And such a camera audit itself by occasionally sending full frames to a system, allowing the 'learning process' to audit it when time permits.)
Real knowledge is semantic compression. That's all.
... Of course, I'm no epistemologist. I've studied this for almost exactly two months now. However, this looks good mathematically and philosophically, and I plan to implement it. I'm of the opinion that nobody really understands something until they prove it through good, ol'fashioned utilization, and that 'nobody' especially includes me.
Oh, and wrgt the idea that 'silence' is above wisdom... I suppose it is, if you entertain the notion that useful information (as measured in bits) is fundamentally related to surprise. I.e. roll 1d6, and the specific value rolled is good for lg(6) = 2.585 bits. Confucius would simply note that the value will be between 1 and 6, and the world will move on... and thus has only lg(1) = 0 bits. True silence, eh?
knowledge, itself, is very multi-dimensional;
Hmmm, where are the UgLyPeople who should delete this page and change it to something like KnowledgeIsSomethingBecauseWikiDoesntLetUsUseSingleWords?.
IdeaSpaceAsAnEvolutionarySystem? KnowledgeProliferation
EditText of this page
(last edited July 9, 2010)
or FindPage with title or text search