 |
|
|
 |
|
|
|
|
Dariusz Daćko
[Guest] |
| | | | | | posted 1/28/2008 15:35 |   | | | | | | | | | Hi,
I want to describe my idea of building a large ontology and thus the Semantic Web. Because it can be a very large database of structuralized and reliable knowledge it can be also used as knowledge for artificial intelligence development. I think that lack of knowledge understandable for computers can be major obstacle to build AI. It can be also a new kind of information exchange medium for people. I introduced the idea on following web page: http://tarest.republika.pl/semweb.html and described it in this document: http://tarest.republika.pl/files/centralized_ontology.pdf.
Dariusz Daćko
http://tarest.republika.pl/
| | |
| | |
heraclitus
[Guest] |
| | | | | | posted 1/29/2008 12:46 | | | | | | | | | | Ideas can be voted in the Democratic Science Wiki, everyone has a possibility now, the road is open.
But, if you want that people would vote for your idea, you must thoroughly and correctly describe it somewhere, and you must formulate your idea so that many people would understand. You also must compare it to other similar ideas, provide foundations for your idea, and provide references to these, either to peer reviewed papers, or to other works approved by voting. You must explain why your approach is better than any other existing one, why your solution is unique, and why do you expect that it would be successful. My question would be, how is your idea better than efforts like cyc, which did not have any remarkable success. If you can do all that, answer to all questions, and you can show that you have an approach which is truly better than any existing one, in general or in some special niche, and can also explain it clearly to people, then you would have a reason to expect that people would support you.
| | |  Democratic Science Wiki | | |
| | |
GrandBotMaster |
| | | | | | posted 2/2/2008 15:16 | | | | | | | | | | Save yourself the time spent in re-inventing the wheel and download the OpenCYC database and engine.
http://www.OpenCYC.org
YOURS -- Christopher Doyon
---------------------
| | | MLAI Foundation | | |
| | |
tarest
[Guest] |
| | | | | | posted 2/3/2008 17:50 | | | | | | | | | |
No team, regardless of its size, is able to build very large knowledge database that could be used to build the Semantic Web or other truly intelligent system. The only way is to build it by whole community. It is not only my opinion, but also opinion of team implementing myOntology, McKinstry (Mindpixel project) and I am sure that also many other scientists. I have started reading about similar approaches recently so I don't have vast knowledge about them yet.
I am familiarized with Cyc ontology since 2002. Its main limitation is inability to build really large (and complete) database because of reason mentioned above (Wikipedia [1]: “The current incompleteness of the system in both breadth and depth and the related difficulty in measuring its completeness”, “A large number of gaps in not only the ontology of ordinary objects but an almost complete lack of relevant assertions describing such objects”). Apart of this, using CycL makes impossible to insert some information that can not be expressed in it.
Currently I know three projects in which knowledge is gathered by community. These projects are: Mindpixel, Open Mind Common Sense and myOntology. Differences with myOntology were described in my document. OpenMind Commons Sense project has following disadvantages:
- it is susceptible to vandalism (you can see it easy if you look at statements with largest number of votes). I suggest that users should be verified as real physical persons. Thanks to it vandals could be permanently deleted and there will be no vandalism. It will make the process of database creation much faster and much more pleasant. This solution is also proposed by Larry Sanger for Citizendium project (aiming at creation of more reliable Wikipedia).
- all statements are inserted in natural language, next natural language processing is used to build very imperfect ontology from them. So it is the opposite of Cyc. I suggest building system in which information will be stored not so formally as in Cyc, but also not in natural language.
- In OpenMind, information should be confirmed by large number of users to be reliable. In system proposed by me users will be reliable, so they will try to insert true information (it seems that in Mindpixel similar approach is used). So information also will be reliable. Information reliability can be calculated on the basis of reliability of users supporting it. It only one person supports information, but this person is a professor, than it is sufficient for me to believe in its correctness. No other comments about information correctness are required!
I am currently reading about Mindpixel and Open Mind Common Sense projects. When I read about Mindpixel I will post comparison with this project.
And finally I want to ask you: did you read pdf written by me? I wrote there about main principle i.e. about community-driven building and about consensus. It is obvious difference from Cyc. I admit that I should wrote in this pdf also about Cyc writing that “building a big centralized ontology isn’t currently planned”. But I am thinking about ontologies much larger than Cyc. It is true that Cyc is big but from other point of view it is still small.
Dariusz Daćko
[1] http://en.wikipedia.org/wiki/Cyc#Criticisms_of_the_Cyc_Project
PS. Why registering to this forum doesn't work? No email is send to me with password.
| | |
| | |
tarest
[Guest] |
| | | | | | posted 2/3/2008 18:31 | | | | | | | | | |
>But, if you want that people would vote for your idea, you must thoroughly and
>correctly describe it somewhere, and you must formulate your idea so that many
>people would understand.
I think that I did this in pdf, is something unclear?
>You also must compare it to other similar ideas, provide foundations for your
>idea, and provide references to these, either to peer reviewed papers, or to
>other works approved by voting. You must explain why your approach is better
>than any other existing one, why your solution is unique, and why do you expect
>that it would be successful.
Comparison to myOntology is in pdf. I posted comparison to Cyc and Open Mind Common Sense some minutes ago.
Maybe I will also describe differences with Wikipedia:
-----
I think that storing information as long text in web pages, articles, books and Internet encyclopedias (as in Wikipedia) has one fundamental disadvantage. It is extremely
hard (or rather it is impossible) to write the text in such a way that: 1) it is in 100% true and it covers all essential information 2) nobody wants to add something to it or change something. Article will be in permanent edition state (as it can be observed in Wikipedia). I think that if
supervision by experts will be added (as proposed in Citizendium) than articles will be reliable but permanent editing will still have place (it will only be slower). So I was wondering what can be done in order to enable adding information in this way that it wouldn't have to be changed in the future. And I have realized that if information will be divided into very small, indivisible parts (for example simple sentences such as "Cat has four legs") it will be so small that if it turns out to be false than we can simply delete it and replace it with new correct information. What is more, true information may exist in the system without any changes - no edition is required. Such information pieces can be organized hierarchically into ontology to enable semantic search of them. I was also wondering what to do to discover the truth i.e. that only true information will be added to the ontology. I think that requirement of long enough discussion about individual pieces of information will result in reaching consensus which will be truth. So proposed information source will be reliable. Proposed ontology can be viewed as centralized version of the Semantic Web as well as it can be used to create independent Semantic Web. It can be viewed also as computer implementation
of so called collective intelligence.
-----
I think that I don't know other ideas. Please let me know if there are some other ideas. I am not a scientist and I didn't read many peer reviewed papers, thus I cannot provide references to them. But did idea needs them? I wanted only to ask you if it is good? :)
Dariusz Daćko
| | |
| | |
purplefrap |
| | | | | | posted 6/4/2008 20:19 | | | | | | | | | | | |
Dariusz Daćko wrote @ 1/28/2008 3:35:00 PM:
I think that lack of knowledge understandable for computers can be major obstacle to build AI.
| |
What about another major obstacle for AI - the lack of understanding what "understanding" actually is.
The Semantic web will be a great resource to tap and will be proven to be very useful but semantics will only get you so far. Even if you build a central ontology every entity within it will be making an assumption. Example: "Cat has 4 legs". I've seen cats with 3 legs due to injury and I'm sure there are 5+ legged cats due to mutation.
Top-down approaches to AI will always be flawed as assumptions are carried with all elements of knowlege. How can something undertand and have an open mind when everything it receives is tainted from the start?
p-frap
| | | Defining Understanding for AI | | |
|
|
|
|
|
