"Ontology" A great word, with a cryptic meaning that indemnifies the role of AI researcher.
I keep seeing this word, ontology, pop up when I'm doing research on some new artificial intelligence concepts. Most of the time it comes out of refined university types that revert to using a kind of niche lingo that allows them to talk about things that might make them famous without giving away exactly what it is they are working on. The definition I found was startlingly simple and yet perversely vague.
ontology - A specification of a conceptualization
What in the hell does that have to do with AI? Let's break it down a bit further and maybe that will help.
specification - A detailed precise presentation
conceptualization - An elaborated concept
Haha, that is speaking in circles if you ask me. A detailed precise presentation of an elaborated concept. I've done a great job here of putting together the most comical definitions of each of the words, but hopefully you see where these people are coming from. When they say "ontology" they are trying to form some agreement on how to share their data. Now that I understand. Some general ontologies are that presented by the XML based language AIML and how it can be used to define the existence of particular AI personality... This goes along with specific implementations that work on top of AIML such as A.L.I.C.E.
For me, I pretty much stop right there. I know now that the ontology is all the crap I'm going to feed to the AI that defines it's body of knowledge. I can feed that same crap to another AI and the two will theoretically be able to produce results that are in the same domain of knowledge. The theory underlying how the thing thinks, what language it was programmed in, etc... all of that isn't part of the ontology, that is something else entirely. That means this agreement on some formal structure allows these AI researchers to more easily compare results. Fine, I'll let them finish the explanation http://www-ksl.stanford.edu/kst/what-is-an-ontology.html.
Moving forward, I'll try never to use this word again. It is confusing and we've developed our own lingo in the realm of computing. We call it an interface when it is formally defined in code and a format when it is formally defined in data. With that in mind the .NET Terrarium contains an ontology. In our case it is the "OrganismBase" where all of the interfaces between the agent and the game world exist. All of the agents are different, programmed in different languages possibly, but yet they all have access to the same knowledge. I once said that you don't enable your agent by providing it access to information, instead you limit your agent by ONLY giving it access to specific information that you decide is relevant to it's task. This process of limitation is crucial to providing a consistent series of responses in nearly all applicable uses of AI (master systems, fuzzy networks, and games). After all, the first time your fuzzy smart microwave thinks of trying a new way of cooking your food and fails miserably is the last time you'll ever use it.