Knowledge of Domain | Artificial Intelligence

In this article we will discuss about the need for knowledge of domain.

Efficient AI systems require knowledge of the problem domain. Knowledge is more than simply data or information. Data are raw facts or elementary description of things events, activities, transactions etc. which are captured, recorded, stored and— classified but not recognized to convey any specific meaning. Example of data include grade point average, bank balance or number of hours an employee worked in his company.

Information is a collection of facts (data) organized in some manner so that they are meaningful to a recipient. For example students name, with GPA, customers names with bank balances is the information and comes from data when processed.

Upon further refinement, analysis and the addition of heuristics information may be converted into knowledge which is useful in problem solving and from which additional knowledge (wisdom) may be inferred.

Knowledge is central to intelligence. We need it to:

1. Use or understand a language

2. Make decisions

3. Recognise objects

4. Interpret situations

5. Plan strategies

Knowledge can be of declarative or procedural type. Declarative knowledge means representation of facts or assertions. This tells ‘what’ about a situation. Procedural know ledge means representation of actions or consequences and tells tell how of a situation.

The difference between declarative and procedural schemes may be illustrated by a simple example. A table of logarithms is an explicit enumeration of this (numerical) domain knowledge and would be considered a declarative representation. On the other hand, a stored sequence of actions indicating how to compute the logarithm of a number must be considered a procedural representation.

Consider another example from real time experience:

Suppose a NRI friend wants to visit you. Giving him the directions that I live in Kurukshetra sector 5. HUDA colony and my house number is 607, is not sufficient. In addition to giving this factual knowledge (declarative knowledge) we must pass on to him the mode and method of reaching me (procedural knowledge).

We would have to tell him that while coming from Delhi by a bus, or a taxi he must drop at the bus station, Pipli. From there he must be told that he should follow Pehowa/Kaithal (the two cities in Haryana) road OR he should call me and I shall pick him or escort him in my car.

The above two examples illustrate the trade-off in using declarative or procedural representation. Declarative representation are usually more expansive (and expensive), in the sense that enumeration may be redundant and inefficient. However, modification of declaration representation is usually quite easy: one may add or delete the knowledge easily. Procedural representation on the other hand may be compact, at the expense of flexibility.

The division is not that rigid and practical representation may include both declaration and procedural components. Modern approach is that versatile that the intelligent machines may need to use both procedural and declaration representation.

Another method of categorization of knowledge can be:

The knowledge we have on a particular subject (domain specific knowledge) and the general or common sense knowledge which applies through our experience (domain independent knowledge).

The fact that the number 301 bus goes to University is an example of the former:

It is knowledge which is relevant only in a restricted domain in this case, Haryana roadways- going to Kaithal or Pehowa (two cities near Kurukshetra).

New knowledge would be required to deal with transport in other cities since different bus numbers go on different routes. Knowledge that a bus is motorised means of transport is a piece of general knowledge which is applicable to buses through our experience.

General or common sense knowledge like domain knowledge also enables us to interpret situations accurately. Imagine that someone asks us “can you tell me the way to station?” Only a deliberately obtuse person would answer literally ‘yes’. But common sense knowledge expects from us that the person wants a set of directions for reaching the station.

Commonsense Knowledge:

This is the kind of knowledge which we humans routinely deploy in our day-do-day existence when we are not engaged in tasks which require technical knowledge and skills which have been acquired through specialist training.

Thus, for example, the knowledge of the physical world which is required is not what is presented within the academic discipline of physics, but rather the ‘naive’ physics which is shared by all humans by virtue of their intuitive understanding of how physical objects behave under the ordinary circumstances encountered in everyday life -for example, the knowledge that unsupported objects fall towards the ground, that unpowered moving objects normally come to rest eventually, that solid objects cannot occupy overlapping locations simultaneously, that liquids can be poured, and so on.

In keeping with this emphasis on common sense rather than specialist or technical understanding, it has been the view of many practitioners of knowledge representation that this commonsense understanding is largely qualitative in nature.

With reference to time and space, this has the implications that the standard mathematical models which have been of such great service in the natural sciences may be less appropriate for the purposes of emulating the spatial and temporal aspects of common sense knowledge, and this in turn has sometimes led to conflict or misunderstanding.

Some aspects of everyday knowledge which have been found to be particularly problematic – and which have therefore engaged what might be thought to be disproportionate amount of the attention of KR researches- include vagueness, uncertainty and granularity.

The problem of vagueness (or indeterminacy) is that whereas much of ordinary language is imprecise in various ways, the symbolic systems devised for representing in a computer the knowledge encoded in such language are by nature precise. An example of this which has been much studied in the context of GI science (Geographical Information) concerns the application of place names to geographical regions.

Many place-names seem to be somewhat indeter­minate in their spatial reference. For example, terms such as ‘Central Mumbai or the west of India which are often used in everyday life, do not correspond to any precisely delineated geographical regions, since it is in some measure indeterminate just which locations they cover, yet most geographical information systems and spatial cover; yet database are only able to assign a precisely determined spatial region as the referent of such a name.

This problem has been studied in relation to such examples as forests, mountains, and town centres, and various technical solutions have been proposed for how to represent vagueness formally, e.g., using fuzzy set theory. ‘egg-yolk’ theory rough sets, super valuation semantics and anchoring.  

Uncertainly is different from vagueness in that whereas the latter involves the intrinsic indeterminacy of certain terms, the former is concerned with the limitations in our knowledge. The locations of Central Mumbai is vague rather than uncertain, because there simply is no fact of the matter as to where the boundaries of Central Mumbai are — it is not as if we could discover these by unearthing new facts.

On the other hand, the location of Archimedes’ tomb is uncertain, not vague; it must have had a precisely definable location, but we do not know for sure exactly where it was. Approaches to uncertainty include probabilistic methods and various forms of non-monotonic reasoning.

There are thousands of ‘facts’ which are obvious to us from the experience of the world and clarify the difference between domain-specific knowledge and common sense knowledge.

For example:

1. A Prof, is senior to associate Prof.

2. A person’s age increments by one each year.

3. Children are always younger than their parents.

4. People don’t live much beyond 150 years.

5. Pure air is useful for health.

These are the examples of the common sense knowledge or general knowledge which humans experience through shared experience.

The specific knowledge will depend upon the application. For language understanding, we need to provide the knowledge of syntax rules, words and their meanings and the context- for expert decision making, we need knowledge of domain of strategies: for visual recognition knowledge of possible objects and how they occur in world is needed. Even simple game playing requires knowledge of possible moves and winning strategies.