2014.02.15
Knowledge. What is this? Knowledge has two components. First, it is a system structure, obtained through experience. The second is the ability to "manipulate" the knowledge in a most general way, to posses algorithms to work with knowledge.
Here we come to the notion of "Understanding". What is it? For humans it is intuitive ability to work with knowledge. Until knowledge (structure and algorithms) becomes available on intuitive level, there is no "understanding" yet. All sorts of "working with knowledge" are helpful in understanding, like looking from different point of view, discussing, just sorting. But by far the best way to obtain understanding is the "problem solving". The more difficult, unusual, inventive, creative problems are solved, the better. I conclude from there that a "problem solving" module is essential to any artificial system claiming to be an "AI with knowledge processing". Let me call it "An AI of a Second Level" in contrast to definition of a simplest AI system I have given before which defines what AI is in general.
But still, problem solving itself does not provide the ultimate "Understanding". The Ultimate Understanding is obtained, when knowledge is used for a creative process on that knowledge. When rules and methods are found of creation of certain systems within the domain of available knowledge. Through this mechanism of "Creation" as well as through process of problem solving, contradictions between knowledge and reality and different models within knowledge are found. Which is the source of further improvement.
But Problem Solving has one advantage over the Creation Process. It is "outside" nature of problems. They come from outside of the knowledge domain, therefore testing it for completeness. While creation process is possible on available knowledge, problem solving quickly reveals that usually "knowledge" obtained is not enough. Not enough depth. How to obtain "Depth" of knowledge? It is consideration of knowledge from different points of view. I can say that it may be "restructurizing it with different structures".
The difference between problem solving and creation process is this. Creation process operates from within. Problems are usually set up from outside.
Another way to obtain knowledge depth is what in TRIZ is called "multiscreen model". It means, shortly, considering knowledge of a system with its relations to oversystems, subsystems, past, and future, knowing its evolution, knowing its contradictions, applying operator STV ( Size-Time-Value ) to reveal system parameters range and constrains. The next, it comes an interaction of the system with other systems, immediate or possible.
I have found that the history of the knowledge about a system is also very important. It helps to view the knowledge beyond the current "background" general level of knowledge, which we usually do not do. This "background" knowledge puts a lot of psychological inertia. This also includes the history of terms, language used to describe the knowledge.
2014.02.14
Analogy... So many definitoins.. It is thought of from ancient times. It is considered the "the core of cognition". That is how people think. They do not think logically. It is an achievement of scence. They think analogically.
To me that is one more cofirmation that the system science is the root of all. Because, analogy, from the systems view is nothing but finding the same system structure in two areas.That is how our brain works - it works to comprehend systems and their structure. And that what supposed to be a target of an AI design.
PS
"Similar" means "the same" in one thing and "not the same" in something else.
2014.02.03
I was asked a question "what is an appliction of AI? Does AI suppose to replace humans?" I think the most important application of AI is a control of systems which humans are not able to control efficiently. We already have overwhelming number of such systems, from technical to social. We are on the edge. A little further, and we will not be able to control them because of the speed of processes there, the complexity, and, last but not least, human factor. Paradoxically, but to provide efficient control and monitoring from society over AI control system is significanlty easier and transparent then to provide the same thing over human control organizations.
Another application of AI is dessimination of unique experience and methods. For example, in pedagogy, personality of a genious teacher is paramount. Because he is able to grow up his students by constantly monitoring their sucess, set backs, tiredness, mood, and through that develop an optimal learning curve for a particular pupil. But such teacher is only one. He can handle only so many students and no more. If it would be possible to create such AI system which may function efficiently as a replacement for such teacher, it would be a greatest achievement. Since, it means, the teachers personality is preserved in a live, acting system forever.
The third application of AI is knowledge management. Everyone knows, that the amount of knowledge accumulated, yet used to solve problems grows exponentially and long ago overpassed abilities of any human to use it extensively. Narrowness, specialization is the way to deal with it so far. While the solution of all problems with the level above 1 requires broader view. Renaissance mind is getting more and more rare.
Surprisingly, in this area AI achieved the most significant results. It is internet search engines. They do possess all the component of AI. They accumulate data actively going through websites in forms of text files, they do produce the output in relevant terms for a user, and they do actively measure the success of the search results. For this purpose they do have a model of internet data inside. Other knowledge AI systems, like a private company systems may be more targeted, since they can know about structure of the company, its problems, equipment, personnel, customers.
Again, I'd like specifically note that knowledge management is not the same as data management. It is a very different thing. In data management system, a human makes a decision, he conducts the search, structurizes information ( that is knowledge), and solves problems. In an
AI system, AI does it.
Another application of knowldge management AI is a "personal knowledge manager". It is reads all you read, it records it all, it provides you everything your forgot, it suggests everything you like, it structurizes information the way you like, it searches for everything you need, it is another you, because it has your model of knowledge inside. You can discuss the things together, and come to better understanding.
And it is not connected to internet.
For a scientist, the PKM can "translate" from different fields, provide some bridges to other unfamiliar areas of terms, find analogies, find applications, problems, examples, contacts.
PKM can contact with an organization-wide knowledge manager.
2014.02.01
Artificial Intelligence is a an ability of a computer program to design a model of reality given with sensor(visual, tactile, aural, etc) and/or logical input (commands, texts. etc) and act upon it successfully. The key points here are:
- Designing a Model. This is the most important thing distinguishing AI from regular programs. AI program goes have a model of reality inside. This model is built by the program itself and is adjusting, changing, being redesigned continuously;
- Sensors (availability of input information);
- Acting (providing output actions or information);
- "Successfully". An ability to estimate output back via input, existence of a criteria that the actions were "successful" in some sense;
An AI system "lives" by continuously transforming information into actions.
As you can see AI is related to system science: model, feedback, sensors, actions, development.
Role of TRIZ is to help building the Model
This is my working definition of AI.
2014.02.01
IA and AI. I like it.
2014.01.13
What is pattern? Mathematical definition. What can be used to generate patterns. Is there relation between patterns and groups? How to generate pattern of patterns? Are there aperiodic patterns? What is what could be called "break in a pattern". There is no tag "pattern" on this site. Is there any synonim of "pattern" in mathematics? What studies patterns?
I have difficulty to find answers to these questions. At least something! Please. e-mail me. Leave your opinion in the Patterns community on Google+.
I do not still know a defintion of what is pattern. I do not think patterns are nesessarily symmetircal or repeatable. Patterns are elements interconnected with some governing regularity, law, rule, formula. A pattern as a collection and its rule form a system. AxxAxxAxx is a pattern, periodial pattern, but AxAxxAxxxAxxxx is a pattern too.
A pattern can manifest, show, represent a rule: when there is a pattern it should be a rule. On another hand, a pattern can set up a rule, a rule defined by a pattern itself.
Intelligent things seach for patterns in sourroundings to discover governing rules of it. Patterns happen where there is repeatability. A pattern is a footprint of a system acting around.
[ Wow, isn't it an example of thought process: It left its record, pattern in the previous pragraph: from "I do not know what is pattern" to "A patter is.."]
From "A pattern is a footprint of an acting system" follows the answer to several questions. As you can see the definition of pattern is very simple in system science. May be mathematicians should look more closely what comes first: math or theory of systems.. Another remark, If there is an "acting system", it should be a "media" to record the pattern. Two elements. The media may have its own structure and "laws of operation" which might affect the pattern. Ok, answers.. How to generate pattern of patterns.. Assume that an acting system leaves a footprint which is its pattern. If this system is a subsystem of another system, The pattern will be a combined pattern of actions of these two systems. It is a "pattern of patterns". "Break in the pattern" happens when the operation of its system changes.
2014.01.10
An approaches to AI based on systems science and philosophical works of Altshuller, Bogdanov.. Our goal is to learn the body on knowledge from these areas, find mathematical formalization, and use it for break-through advances in AI. I am inviting TRIZ specialists as well as software ones and mathematicians to take part. I expect this community be mainly about AI.
First posts:
For those who do not know what is TRIZ. TRIZ is an independent part of system science dealing with qualitative development of systems, their design, transformation, utilization. It has huge heuristic values and numerous practical methods to solve complex problems in all areas. These methods were thoroughly tested on thousands of cases. Also TRIZ pays a lot of attention to psychological conditioning of a problem solver. Please, contact me to find out more about what is TRIZ.
* * *
Let TUProgram be a program to undertand text. What is "Text Understanding"? It is matching of a hypothetical model presented by a text writer to a model in the TUProgram. Text is a content of an information channel, it is not the model itself. It can be misleading, contradictory. TRIZ provides methods to deal with contradiction. This is where conventional mathematics fails... Of course, TUProgram model is dynamic, but it operates on defined rules. Expansion of these rules is "Machine Learning of the TUProgram". But it still has to have very basic concepts. These concepts have to be of system science, mathematics, basics of other sciences and arts. TRIZ is a part of system science. Therefore it must be a part of the basic concepts for Text Understanding. Text Understanding will obtain its power from the most powerful and correct basics. If the basic are choosen incorrectly, Text Understanding system will not be successful.
* * *
Importance of Text Understanding for TRIZ. Text Understanding is the mean to verify TRIZ concepts and laws, add new knowledge, and the most importantly, continuously monitor the all available knowledge for new things. Naked human effort will fail in the avalanche of information, it is subjective, prone to errors. The process may go iterativelly. With solid basic concepts available now which still should be formalized, a simple TUProgram can be made. With this, more information can be obtained, it can learn. Thus, the basic concepts can be verified and expanded making possible the next version of TUProgram, and so on. The successful TUProgam can build itsself. How can we assure that this program is correct? One way, I see, to make it work with Progam Solving module. TUProgram will extract problems and pass it to Program Solving module which will solve it. Then the result will be compared, automatically, of course, with available answer. They should match, and certain criterial should be used for ranking the answer, since there should be many of answers, and known aswers are not nesessarily the best. TRIZ can give such criteria.
2014.01.08
"Calling a three-dimensional universe “flat” means its shape is well described by the Euclidean geometry familiar from high school: straight lines are parallel and triangles add up to 180 degrees. Extraordinary flatness means .." -- Link
2013.03.20
There is Space.
It has 4 types of transformations:
1. Mirror. A part of space which stays the same is called a surface.
2. Rotation. A part of space which stays the same is called a line.
3. "Turning inside out". A part of space which stays the same is called a point.
4. Shift. No part of space stays the same.
Postulated that if two transformations map the space in the same way - the surface, the line and the point is the same.
A surface divides space into minimum 2 parts excluding the surface itself.
A plane is such sufrace, which no matter how it is transrormed, divides the space like this: 2 planes may divide space into maximum 4 parts. 3 planes may divide space into maximum 8 parts. 4 planes divide the space into maximum 16 parts. In this case, one of these parts is completely enclosed by the planes. Minimum 4 planes is required to create such figure.
There is also Squeeze - Expansion - Twist kind of transformations ( the one which changes the metric ) called Metric transoformations. ( Wha t is Metric? I am looking for a definition )
1. Mirror tears the space
2. Rotation may twist the space
3. "Turning inside out" does what? It tears the space too.
4. Shift may squeeze-expand the space.
A surface divides the space into 2 parts. A line and a point do not divide the space except vyrozdeno. A line has 2 ends( means directions) !!! [is it an axiom or a Conjecture?]
What is a plane and a straight line?
There are such Rotations and Shifts which transorm the space gradually. But Mirror and Turning-inside-out transorm abruptly, not gradually.
Double mirror over the same plane P1 maps the space into itself:
Sp -> Mirror over P1 --> SP`-->Mirror over P1 = Sp or mir( P1, mir(P1,S) ) = S
A plane P2 is different then P1 if the result is not the mapping of space onto itself:
If Sp ->Mirror over P1 -- SP` --> Mirror over P2 != Sp => P1!=P2
A non-zero metric transformation inserted into transformation chain which maps Sp onto itself breakes the equivalence:
Sp ->Mirror ober P1 --> SP1 -->any Metric transformation --> There is not Mirror transofrmation which maps -->Sp.
Opposite metric transformations are those which compensate each other.
Mirror of Mirror result can be obtained by Rotation of S around a line L, which is L = P1 intersection with P2 ( prove it ! )
mir ( P1, mir(P2, S ) = rot(L, S) , where L=P1 and P2
mir(mir(mir))) - is it the same like "Turning insdie out"?
mir(mir(mir(mir))) - can it become a Shift?
Is it true that there are such not equal 4 planes such that mirs over them will map S into iself?
Parallel planes and lines are such planes and lines that there is .....
It two mirrors produce the result which can be obtained by a shift ( not rotation ), these surfaces are called parallel. Does it mean that all shifted surfaces are parallel? Yes. Parallel surfaces and lines do not intersect. This is a result of the definition.