TRIPLE-I 2008: First Day Filled by Commonsense Knowledge
The TRIPLE-I conference in Graz today started with a keynote by Henry Lieberman, research scientist at the MIT Media Laboratory. Given that, nominally, at least a third of the conference is dedicated to knowledge managemen, Lieberman introduced an important, often overseen aspect of knowledge management right at the beginning: Managing knowledge that everybody knows already.
Knowledge management typically aims at knowledge that people do not know yet, e.g. (tacit) knowledge that people have acquired in a project and that is suppose to be made explicit and accessible to other people who don’t yet have this knowledge.
But what about the knowledge that everybody knows without them knowing they need to know it? Such as that an apple is a type of fruit, and is green and is red? Common sense knowledge?
I boldly asked Henry Lieberman for a 12 seconds definition of Common Sense Knowledge, a challenge he accomplished with perfect precision:
Henry Lieberman defines Common Sense Knowledge on 12seconds.tv
An intriguing MIT project I hadn’t yet heard about which Henry Lieberman introduced is the Common sense knowledge base Open Mind Common Sense – anyone can sign up to it and contribute. A total of 203 knowledge facts have, for instance, been accumulated about the concept “apple”, including facts such as these:
â†’ An apple is red
â†’ An apple is green
â†’ Apples grow in trees
â†’ an apple are food.
â†’ An apple has a core
â†’ An apple can fall from a tree
â†’ An apple is a type of fruit
Offered similar concepts are “egg, potato, steak, bread, spinach, frozen food, butter, appl [sic], leftover, grape”. The process of adding knowledge is guided by a list of questions that allow to conceptualize and structure the knowledge, e.g.
What is it made of?
What kind of thing is it?
What do you use it for?
What can it do?
What is it part of?
How do you define it?
But what are the roles that common sense knowledge can play in interactive applications? Henry Lieberman suggested using common sense knowledge, a system can e.g. anticipate what a user is most likely to do, or it can at least make most likely things easiest to do, e.g. by providing a map from goals to concrete actions in the interface, or by integrating appropriate applications.
Lieberman furthermore introduced a couple of tools which illustrated these benefits, e.g. the prototype for an Event Minder for improved scheduling driven by common sense knowledge. Entering a statement such as “Lunch with Charlie at Miracle next Friday” would for instance calculate the date of ‘next Friday’, call up a calendar application and also a web service to get directions for getting to Miracle.
Regarding the difference between CYC (the common sense knowledge ontology) and the MIT’s common knowledge base Open Mind Common Sense: CYC is an ontology organized by experts with a broader and deeper knowledge – the common knowledge base grants access to anyone and has, for instance, also information about kitten that might not be that relevant to experts. At this stage, there is no mapping to CYC.
Henry Lieberman’s keynote tied in nicely with a presentation by Andrew S. Gordon about “Envisioning with Weblogs”. According to Andrew Gordon, there have been three waves in the 50 year history of common sense knowledge in artifical intelligence:
First wave: Logical formalizations of commons sense knowledge (e.g. CYC)
Second wave: volunteer contributions from web communities (e.g. Open Mind Common Sense)
3rd wave: Knowledge acquisition from the social web (e.g. Envisioning with Weblogs)
First off, what is envisioning? Andrew Gordon described it as a form of reasoning about states and events in time and space, generating answers to questions such as “What’s happening in the world right now?”, or “What is going on in the audience’s mind right now?”, or “How did this person get into the room?”, or “What am I going to have for dinner tonight?”
At the Institute for Creative Technologies (University of Southern California), Andrew is involved in a project called Story Representation and Management, which among other things, is doing research on story interpretation, i.e. “techniques for integrating automated commonsense inference into the processing of narrative text documents, and methodologies for creating very large scale commonsense knowledge bases.”
One of the paths towards the creation of this knowledge base is gathering up stories on weblogs. But can we really gather up all stories ever written in a weblog? In the research conducted and cited by Andrew (Gordon 2007), 4,5 million stories, made up of 66,6 million sentences and 1,06 billion words were extracted from weblogs.
In Gordon’s recipe for envisioning with weblogs, the retrieval of the closest situation provides the best results. Take for instance the quest of formalizing this particular problem in common sense physical reasoning: cracking an egg into a bowl (as described by Morgenstern 1998, Lifschitz 1998, Shanahan 1998).
There are so many things to be considered: Is the bowl big enough? What if the bowl is made of cardboard? What of the egg is hardboiled? Common sense knowledge in stories on weblogs does offer many answers, for instance this story from Amit Asaravala – which also generates further knowledge as to what would happen to a person who does this:
Seeing the little weirdo reminded me of one Saturday morning, a year or so ago, when I cracked an egg into a bowl and found three yolks inside. After tossing the triplets, I cracked another egg from the batch and found yet another three yolks jiggling up at me. Another egg, another trio of blondes.
This continued through all twelve eggs â€” I kid you not.
Though the episode had me thoroughly creeped out, I must say that I am somewhat intrigued by the thought that, on some farm somewhere, there is a crotchety old hen that consistently lays triple-yolkers.
In the following discussion, some people wondered if weblogs aren’t an unreliable source for a common sense knowledge database. Andrew however doubted that the difference true/false or the difference true/fictitious did really matter. Instead he suggested that in 99% of the cases the same physical reasoning applies in, say, the Star Wars Universe as does apply in the real world.
Common sense knowledge is not about the velocity of spacecrafts crossing the milky way, it’s about what happens if Leia punches Han.
Which is yet another point sustaining that common sense knowledge is so obvious that most of the time we don’t even know we know it. And that’s a challenge to knowledge management.
Oh, and something very nice happened to me today: While I sat in our booth preparing this blog post, someone approached me very politely saying that he had read my name somewhere before, on some blog. Turns out this person – Stefano Bertolo, Project Officer at the Information Society Directorate of the European Commission – has in the past also left a comment on the Flickr page of our “Escape from the Data Silo” logo (which can be used freely by anyone on a CC license). It’s a small world, thanks to Social Media:-) We had a nice conversation at our booth, during which he also recommended the NeON project: Lifecyle Support for Networked Ontologies: a recommendation which I herewith pass on to you, reader of this blog:-)
P.S. There were many more interesting talks and sessions, but the scope of this blogpost is, sadly, limited by the rules of physics: I could only attend one talk at a time.