Knowledge Acquisition is a crucial phase in building KBS but it is usually not a simple or straightforward activity. The required knowledge can exist in many different forms and be spread over number of different sources. The most usual source of knowledge is a domain expert and the very first problem is that the domain expert needs to be both willing and able to express their knowledge.

For knowledge to be useful it must available in a clear, explicit and detailed form. It must also be capable of representation within the chosen knowledge representation system. Typical useful forms of knowledge include facts, theories, procedural and declarative knowledge. Unsuitable knowledge is usually based on either motor skills or is 'compiled' into a compressed representation.

Knowledge itself is available from many sources, the most typical being humans, existing information, books and journals, training materials and through simulation or practice. Do remember that knowledge may itself be contradictory and incomplete. Where human experts are involved it is vital to ensure that they are both willing, able and available to be a knowledge source.

Given that the expert wants to be involved in building the KBS and can articulate their knowledge then there is need to adopt a suitable approach to collecting their knowledge. To some extent this will be dependent on the domain, the types of problem to be solved and the expert's success in expressing himself. The most usual way of KA is through interview though other approaches such as video recording can be used in conjunction with interviews.

There are many ways to collect knowledge ranging from approaches used in traditional systems analysis (interviews, questionnaires, observation etc), through to the use of videos, simulation and apprenticeship. Each has benefits and drawbacks and is suitable for collecting different forms of knowledge. There are cognitive approaches such as repertory grids that can be used to aid experts by improving their recall of associative knowledge. They can also help to clarify relationships between various aspects of knowledge.

Where the expert has problems expressing their knowledge or where they are unwilling then other techniques such as machine learning can be used to build KBS. Other problems such as the limited time an expert has can also be addressed through using computer based tools. It is generally claimed that collecting knowledge is major bottleneck in building KBS. The major evidences for this claim are the lengthy process involved in collecting knowledge from humans, the inability of experts to describe their expertise, misunderstanding and misinterpretation between the KE and the expert, the inability of the expert to relate to the resulting system etc.

Automated Knowledge Acquisition approaches have been developed to counter these problems. Some examples of these include the use of machine learning, computer based knowledge acquisition tools and data mining. One effect of using computer based tools is to change the traditional roles of the expert and KE, generally the KE reverts to the role of computing expert and is involved in building and maintaining the computer based tools. The expert becomes more directly involved in building the KBS and assumes a much higher level of ownership. Their role becomes that of supplying examples to the computer tools and validating the resultant KBS.

Machine learning typically requires the presentation of sets of examples with known outcomes so that knowledge can be 'learned' through example and correction. Inductive learning algorithms e.g. ID3, and neural nets are examples of this approach. In this approach the expert interacts directly with the learning tool and hence has a strong sense of involvement in the project and its outcomes. Here the KE acts as a computer expert supporting the learning tools and the expert.

Tools such as OPAL have been developed to support the generation of KBS. Here the tool has knowledge of the problem domain and offers a form based interface to the expert as an aid to collecting suitable knowledge. This is then used to build the KBS. Here the KE functions more as a computer expert supporting the tool and the expert.

Data mining is an approach that takes existing data and searches for previously unnoticed patterns that occur within that data. Once again the role of the KE shifts from being the collector of knowledge to being the administrator of the tool set.

The result of Knowledge Acquisition should a body of knowledge that will typically be unstructured, and contain a variety of knowledge forms. It will have been gathered over a period of time. During this period the KE appreciation of the problem domain should have increased quite markedly.

The KE needs to be able to extract the knowledge from the various sources and transform it into something suitable for use within the chosen representation system. This is a process fraught with errors in transforming information from one representation to another.