OWL — The Language, Not The Bird
Having just attended ISWC 2016 I can say the title sums up my general experience from the conference. Research into OWL (the Web Ontology Language) and its related technologies seems to be dying and surging forward at the same time. Most researchers are apparently focusing their efforts on OWL, RDF, and SPARQL, but whilst doing so they consistently mention the various issues with OWL.
So why not work with a completely different knowledge representation model? At Grakn, we have chosen not to adopt OWL, because it is difficult to present it as a unified stack. Engineering solutions should be simple, with a minimal learning curve. OWL is complex, even when modelling simple ontologies, since it requires understanding RDF, which represents the data, and SPARQL, to query the data. Engineers tend to favour simplicity over expressivity.
Not to say that it cannot be done with OWL. The rise of Linked Data has allowed the exploration of a more deterministic approach to AI and this provides novel approaches to many tasks that we are attempting to automate.
Deterministic vs Non-Deterministic AI
For example: at the conference, one of the posters was from a team currently focusing their efforts on home automation via semantic reasoning.
To put it simply: if your house is too warm, the system is expected to reason to switch on the air conditioning. But what would stop that system from opening my fridge instead?
With non-deterministic systems (such as Deep Learning or traditional Machine Learning approaches) the answer to that would be a complicated one, but with semantics, the answer is simple: either the ontology would prevent such behaviour or a rule would prohibit the action.
Academic Research Driving Inspiration
Despite being on “the other side”, the academic within me thoroughly enjoyed the conference. Here are some of the highlights:
The latest research in Stream Reasoning allows ontological and rule inference to be done over relational databases without needing the data to be migrated, which I can appreciate because migrating to a new database is a hard pill to swallow.
Ontology reuse and design patterns also received a lot of attention, specifically how can you reuse ontologies partially without losing semantic meaning? For this one I picture my fridge being opened again because of an incorrect ontology migration.
All fascinating subjects, many of which I look forward to following up on.
Vienna Awaits Next Year
These conferences always present a unique opportunity to exchange ideas and catch up with everyone’s fields. I was really happy to see how so many people were curious about Grakn. I was originally expecting more resistance to the idea of a new knowledge representation model. I am looking forward to the next conference in Vienna, and in the meantime, I hope to catch up on my readings about many of the great topics presented.