Misinforming Knowledge through Ontology

Article excerpt

Introduction

The World Wide Web Consortium (W3C) has now a recognised ontology modelling language (W3C 2004). Tools exist such as Chimera from Stanford University (Stanford, 2004) that allow us to [perform analysis on ontologies embedded in web sites. These developments allow us to use software agents to deliver meaning rather than flat text. This meaning can come from the web, or from documents within an organisation, analysed through the organization onotology. When we rely on computer systems to support decision making with derived knowledge, the underlying model used by the computer system must be trusted. Systems based on ontology have a serious flaw in that the ontology grows from a seed that defines the breadth of knowledge that can be organised. If this seed (called the bootstrap) includes domain knowledge, but no knowledge of the people working in the domain, then we cannot trust results.

Webster's Dictionary defines ontology as "the branch of metaphysics dealing with the nature of being, reality, or ultimate substance." However, Abou-Zeid (2003) recognises that, for the AI (artificial intelligence) community what 'exists' is that which can be represented, and ontology from this perspective is "an explicit specification of conceptualization." Knowledge management systems can 'create' knowledge by semantic analysis of a database (often called an 'organisational memory') of data gathered from interactions between members of an organisation and the knowledge management system. To perform such a semantic analysis the knowledge management system must have created a formal language specification of all the concepts used in the organisation. This formal specification of 'what is' for an organisation is the ontology. An illustration taken from software of Sheth, Thacker and Patel (2003) shows this:

Earthquake (latitude, longitude, region,
event Date, description,
damagePhoto,
numberOfDeaths, magnitude);

This definition is of a class that will define the ontological entity earthquake. In this problem domain, whenever the term 'earthquake' is used it must have the parameters: latitude, longitude, region etc and nothing outside this list of parameters has anything to do with earthquakes. By tightly specifying the language of a problem domain we allow the knowledge management system to perform analysis amongst the data captured.

Software to date has been capable of gathering fairly mundane engineering aspects of the processes and operations of organizations. Current research into knowledge acquisition focuses on enriching knowledge capture and creation, and this enrichment requires more sophisticated ontology to underlie the process. What can happen if the ontology is flawed is that analysis is based on oversimplified concepts leading to conclusions that either have little relevance or are flawed as they leave out important factors. This is particularly important when relationships are omitted from an ontology. Imagine a virtual organisation containing two companies; a wine maker in the Barossa Valley and a wine distributor with warehousing in Melbourne and Brisbane. Analysis of captured data from a time and motion study is in the form "shipping to Melbourne will take 24 hours, shipping to Brisbane will take 36 hours" and this is used to make scheduling decisions. In operation the system is found to deliver to Melbourne in 24 hours and to Brisbane in 52 hours. Detailed study of the delivery shows that the trucks going to Brisbane are always loaded last at the winery because a competing company shipping to Brisbane pays more for the wine and is given preference. This class of problem with ontology is due to the difficulty of starting an ontology. Once seeded with useful data an ontology will grow to include all, important factors. The problem is that seeding or 'bootstrapping' of ontologies often ignores whole areas of data, particularly that data to do with relationships between actors. …