On the members area of this blog, a discussion has begun. About ontologies and their role.
The problem with ontologies is that they have to be formal and correct. As one of Gödel’s theorems says, a logical system cannot be complete and, at the same time, correct. There is a similarity with ontologies, here.
One cannot use them to establish ‘observational’ relationships – such as for instance the relationships between genes that always appear to be up-regulated together in high-throughput experiments, or between concepts co-occurring in the same sentence in text. Let alone hypothetical, inferred, relationships, based on overlapping ‘concept-clouds’.
What the Alliance is about is to try and achieve a uniform data structure that makes it possible to establish, with different levels of certainty, relationships between concepts in a given domain (we start with biomedical). The idea of a single, overarching, monolithic ontology doesn’t fit this model. Instead, several ontologies must co-exist, and yes, if combined, they can yield contradictions.
The triple store that at least some Alliance members have in mind allows several ontologies at the same time, without them necessarily having to be consistent with one another. The resulting fuzzines does, of course, mean that there may be domains in the triple store that are less suitable for ontological inferences and even contain contradictions. But even then, they are of value in gaining insight in which elements of (biological) models are based on observations, where the theory is not yet ‘crystallized’.
The difference with Cyc is that we integrate knowledge other than formal, ontological knowledge as well: observational and inferred, hypothesized knowledge. And this knowledge can be highly significant for decisions as to whether to accept or reject hypotheses, or in the process of improving the hypotheses for new experiments. Cyc is of course a great resource to use, in addition to all the ontologies constructed by NCBO and others.
Jan Velterop and Erik van Mulligen