Jo Walsh is creator and curator of the spacenamespace project, an adventure in collaborative mapping. As an ambitious first step, she is overseeing the building of an interactive map of London, accessible via an instant messaging bot. WriteTheWeb asked her to explain the roots of the idea, and how it is progressing. Her answer follows.
I started working on spacenamespace as a spin-off from logicmoo, a project by Doug Miles, an IRC friend of mine. Logicmoo is a MOO environment, harking back to the old days of online text games and chat environments; what makes it special is its use of the ontology in OpenCYC, the partially open version of the CYC 'commonsense' artificial intelligence project, as its substructure.
So when you walk into a Room, you can find out that it's a RoomInAConstruction, a SpaceInAHOC (Human Occupation Construct), which is a ConstructionArtifact, which is an Artifact, accordng to openCYC's ontology (a kind of concept map)... this may seem like a geek version of 'common sense', but could provide a great learning environment for young children, or for online 'bots' looking to learn more about how the world fits together. Rather than the virtual world supplanting, distracting the user from the physical world, in logicmoo the virtual retains an isomorphic mapping to things in the physical world.
Logicmoo is built from logic languages which I don't know well enough to program in, but wanting to contribute, I started writing a model in my favourite language, perl, based on RDF - the 'Resource Description Framework' model that is the basis of the semantic web, a gleam in Tim Berners-Lee's eye as the WWW was over ten years ago. RDF could easily be translated into the CYC-MOO model, and vice versa.
I was also motivated by the lack of freely available location data - it costs a couple of thousand pounds annually to license the postcode database, for example, and online services like streetmap strictly forbid the re-use of their information. With mobile services taking off, with open wireless networks like the (slightly moribund) consume becoming more popular, GPRS making tracking and targeting viable, it seems important to me to keep as much data as possible in the peer-owned, public domain.
If location-based, mobile services are going to become essential to my lifestyle, like my mobile phone has, I'd like to have an alternative to near-monopolistic commercial offerings. When I'm walking down a London backstreet and my mobile-of-the-future points out that I might enjoy that beer in that pub over there, I'd rather that was a friend of a friend, and not News International telling me so.
So I decided to build an RDF model of London, with as much location data as I could find on the web, or persuade people to contribute. I found a site which listed the postcode of every Tube station with the lines it's connected to; this became a starting grid, which I and others could use as a starting point to map out the local area. The model is available on the web, with a REST-style interface which you can GET places and connections from, and POST new places to. As it provides a semantic web resource to identify locations, it can be plugged in to other RDF- or RSS- enabled sites, and into collaborative recommendation websites like knowhere and grubstreet.
I've been obsessed with bots - small, dumb agents - for a couple of years, so building a bot interface to mudlondon was an inevitable next step. The bot talks to Instant Message services - jabber and AIM at the moment - and uses the web interface as its 'brain'. A nice side-effect of this is whenever I want to add a new function to the bot, I have to provide it on the web, so it becomes publically accessible.
I've used the web service side of it when i'm going out, and want to know "what's really the nearest tube station to this postcode...?" and managed to get random people building their own personal mental models into my map - it "went viral" in my partner's office for a few days. Having the interface available over IM, as well as the web, provides an inadvertent new work-avoidance strategy.
The London model uses an ontology similar, but different, to the one in logicmoo and CYC (though the ontologies connect together at a few points). I'm persuading a small coterie of my friends to vote on additions to the ontology, which are submitted automatically by the mudlondon bot whenever a new kind of thing is added to the London map. The idea of a constantly evolving, consensus-driven ontology which is available for other bots to use and adapt appeals to me greatly.
In my more florid moments, I like to think I'm acting as a bootstrap layer for my own artificial intelligence; the intelligence that I construct around me from things that I find; an emulation of the process my bots will go through in their own world, where the means of making reconnections is much more obvious.
PostcodePosted by: gordo at 2004-05-30
Postcodes, along with Streetmap and Multimap are the new navigation system. Printouts are often seen on the Tube. As the pubs disappear (in the East End), navigation has to evolve.
But there are pitfalls. Postcodes sometime lead us astray. Some are purely pointers to P O Boxen.
I was very interested to see one to one mapping between OpenCYC anything!