conversational interfaces

Trapped in Venice

Earlier this year I managed to get stuck in Venice, Italy at 12 pm midnight, literally babe-in-arms without a hotel or any place to stay.

We'd taken a ferry from Fusina, rather than driving all the way to the kind of parking garage thing that they have on the mainland.  After doing the requisite tourist milling around Venice for a while we then went back to get on our ferry and I realized that the small '*' beside the schedule meant 'summer only'.

My son eventually conked out and we had to take turns carrying him; making a baby holder out of a jacket.  All the hotels were booked.  I called family in Canada, and had them search on the net for me as I wandered around, and they were unsuccessful as well.  Flipping on my ESR911 emergency beacon, or breaking out the backpack full of survival gear didn't seem fruitful.  Ultimately we took a bus back to the mainland and managed to catch a ride with a police-officer back to our car (which was 30 miles away by land).  I then quaffed a couple of redbull and drove all night back home to Pinerolo; not finding any hotels en-route either since it was too late.

Maps Lie

It really felt like this was yet another case where 'maps lie'...  where cold rational line drawings removed of all human doubt, removed of all scrawling in the margins, are in fact worse than nothing.

And the big annoyance was knowing that there must have been other options.  I know I must have carried my sleeping son past a hundred empty rooms.  And past a hundred people; whom; if had heard the slightest inkling of this predicament would have had some clever solution.  People who had guest houses or were friends of friends, or who had rooms to let somewhere near or even in Venice.

It sometimes feels like maps work very well but have these edge catastrophic failure positions; like being blind and trusting a friend who is really great most of the time but occasionally sticks his leg out and trips you.

Bed Finder

It seemed like, for this particular problem, there needed to be some kind of bed-finder service.

Simply some kind of brokerage that aggregated available hotel rooms as rss-feeds and that matched buyers with sellers automatically.

Clearly such a service could be applied to other domains, but even just beds by themselves; being such a human necessity, could probably make a service worthwhile.

The current hotel search services put the burden on the buyer to do searches and to contact hotels by themselves, often hard to do when trying to do things in an emergency situation.  Their own referral databases are privy to themselves, rather than being open to any person who has available rooms, and they don't have any kind of automated bidding capability or match-making capability...

There are aggregators around, and such a service could be built on them (say placedb (which friends and I wrote) and or mapfacture (which other friends led by Mikel wrote)).

And such an aggregator could become a fabric underneath all of the place sharing services, and could encourage them to compete on services rather than also on content.  In fact Platial and Plazes both endorse the idea of basic aggregation (Platial provides support for placedb).  Anybody could contribute or consume; it would be like a smart craigslist or a location-saavy ebay.

A thought for extracting location from unwashed rss

One of the issues in aggregation is that sometimes raw content you aggregate does not have appropriate metadata - like say 'location'.  In the 'bed finder' application this isn't a problem, but if you're aggregating content in general then it makes sense to try to have some strategy for finding the location(s) of a document.  The dumbest way to find where a document is, is to look for place-names or addresses in the document.  A slightly smarter way is to do document comparison.  If you have a corpus of documents that have well pinned locations { say for example the corpus of documents  that users have contributed to platial, or the comments associated with geo-coded flickr pictures } then you can geolocate unpinned documents by simply finding the pinned documents that the unpinned documents are most similar to.  One way to do this is LSI (latent semantic indexing) - which is hugely computationally intensive.  Another way is CNG (contextual network graphs) which is pretty fast and also is 'reversible' - you can see why it has a certain outcome.  There might be a much faster way however, which is to use bloom filters to generate a fingerprint of each document, and then to simply compare document bitmasks...  this could at least reduce the filter set.  I'd be curious to see if that idea works.

Lighter weight interfaces

A bed-finder service would typically run via cell-phone or other device with very limited displays.  It might even be voice.  It would be quite different from most social digital maps - which are heavyweight - and somewhat at odds with the domain they speak to.  Personally an interface that was actually more like a chat room, with human participants, and also instrumented by bots, seems more appealing.  This would be not entirely unlike the #geo irc.oftc.net channel that many of the geo types hang out in... bots could remember, mediate and match...  and the chat based ui is effectively "no interface" rather than form fill-out-boxes interface...

Force Majeure

In general, that night in Venice, it was surprisingly obvious how poor our social place sharing tools were, that they couldn't help and that I couldn't signal for help effectively. Certainly if all bed seekers had had this magical tool then there would still have been similar problems of hotel saturation, but I'm also sure that other choices would have become available opportunistically.

I don't really know if this is the answer...  applying yet more technology to fix a problem... it's just one take...  perhaps just planning better would be good enough...  but I think there are other similar kinds of problems, and these all seem to lead in a direction of some kind of non-siloed distributed signalling network.  As zool said: what is the 'simplest least useless thing' that would work?

Praxis is expensive

Of course it is easy to talk about ideas.  To go a bit further I put up a test of some of these ideas on my hook.org domain; testing the idea of a map centered conversational interface.  I grabbed a whole bunch of pieces from different places and integrated them as a single service.  A test room is at:

  http://venice.hook.org

Here you can stake out an area and converse within it; if you enter a line with "/map hello" it will map that item.  It does not do a lot else yet - please be patient.

I hope to roll in some aggregation related features so that rooms you create can also sum up external content; and then I'd like to try turn on some digital bots to help support matchmaking.

Here, rather than trying to do something that might not reflect what people would really use, I'm hoping to get a bit of feedback, continue improving this, and see if it can't actually be something useful for a broader community.

Comments

[Josh Cooper]

I like the idea of something like this from a perparedness standpoint. I'd like to see something like this as an iphone app.