ODONNELL@arcc.afrc.ac.uk (05/31/89)
* Andrew Wallace suggests a knowledge domain limited to the * sequence-function relationship of genes. From another hereto passive reader, who has been enjoying the debate, I contribute my own opinions, whatever the value: ******************************************************************************** The human way of acquiring knowledge is to learn a large number of specialised topics (domains) in great detail, and obtain less detailed information about the links between them. From the whole we derive inferences and a greater understanding of the interrelationship between two different domains. (eg: protein structure affects enzyme kinetics, but the detail is lacking as a generalised system) The success of the bio matrix project depends upon its ability to make use of the less detailed **linking** information available for biological domains. Indeed it will be expected to draw inferences similar to that of human reasoning. It is the information in the links, and how we represent it in relation to the defined domains, that will be important. The link information is a subset of the whole matrix, so maybe the concern about representing all knowledge in one way is misplaced. In the light of the present discussion this concern seems to be preventing acceptance of the idea. We may only need to standardise the links, by a sort of "input/output" standard. We could start by creating specialised domains, with their own internal structures and rules, as in a 'real' biological system. This is also a practical way of prototyping a larger system. As we improve our means of representing well defined domains we can tackle more difficult areas. Development of the links can proceed independently of the knowledge in each domain. Each domain will remain self consistent and will be of practical use before the rest of the matrix is in place. In programming terms this is "block structuring", and maybe what the instigators had in mind all along. This approach eliminates the need for one type of structure on all types of knowledge. I presume it could lose the strength of analysis that existing AI methodologies provide. But a methodology that looks for 'logic planes' in biology seems quite absurd. It is predictive power and interrelationships that biologists require. ************************************************************************* Cary O'Donnell Tel: (in UK) 05827 62271 ext 226 AFRC Computing Centre (+44 for UK) 5827 62271 ext 226 West Common Harpenden Herts AL5 2JE email: ODONNELL@UK.AC.AFRC.ARCC U.K. (Molecular biology software support at AFRCCC)