One of the basic ways that people make sense of the world is by breaking up the world into objects and grouping those objects into 'classes'.
Experience and common sense about the properties of one class are used everyday to make predictions and determine behaviour when interacting with other members of that class.
When given an apple, you know from experience of many other apples that it can be eaten and that it is good to eat. The way we interact with people is fundamentally different to the way we interact when we come across a building or a motorcar. This common sense is fundamental to our interaction with the world and is the basis of all intelligent acts.
Amazingly, the world’s computer systems have no ability to do this. For them, “apple”, “empire state building”, “barrack obama” and “photo corners” are just sequences of letters, patterns of 1’s and 0’s. They can be processed according to a few rules, hard-coded into the application by the software engineers that wrote it but they have no perception of the basic properties of these entities or how they fit into the real world. This is why computers systems are generally dumb and why internet search is still largely about searching for keywords in a sea of documents.
However, the True Knowledge platform now has a comprehensive world ontology - which is continuing to grow. Excluding species of plant and animal life (about 100k classes) and products (another half million classes), the system now knows about more than 20,000 real-world classes.
These 20,000 classes cover everything in human experience: places, people, planets, substances, concepts: all carefully arranged to show what is a kind of what (all doctors are people,all villages are places), which classes never share members (integers are never substances; indian restaurants are never hairdressers) and which sometimes do (some women are lawyers, some lawyers are women).
For every entity it knows about (some 6 million things) it also knows which of those classes they are members of and which they are not.
Importantly for interacting with humans, the platform also contains comprehensive lexical knowledge: what words are used to denote which classes and objects.
This knowledge underpins the platform’s ability to make sense of human beings' questions and answer sensibly. It knows when asked Who built the titanic? that the user must be asking about the famous ship but when asked Who directed titanic? that the question is about the movie.
A small section of the True Knowledge ontology is shown below:
It is a tiny section of the class tree beginning with [track]: long, thin geographical features such as roads, canals and railway lines. For clarity some classes aren’t shown.
This section is a tiny corner of the entire ontology that has as its root [object] and with subclasses [physical object], [conceptual object]….
With the opening up of the True Knowledge api recently, the good news it that this common sense knowledge is now available to every automated system on the internet.
Bravo, TrueKnowledge!
Posted by: Jack D. Logan | 11 May 2009 at 09:32 PM