Big Brother — Learning and Inferring Transportation Routines

Last week I went to AAAI-04 in San Jose. On Wednesday, the “outstanding paper” award winner was presented. It should also be the “scariest future technology paper”. Here’s abstract (emphasis mine):

L. Liao, D. Fox, and H. Kautz. Learning and Inferring Transportation Routines. Proc. of the National Conference on Artificial Intelligence (AAAI-04).
Abstract
This paper introduces a hierarchical Markov model that can learn and infer a user’s daily movements through the community. The model uses multiple levels of abstraction in order to bridge the gap between raw GPS sensor measurements and high level information such as a user’s mode of transportation or her goal. We apply Rao-Blackwellised particle filters for efficient inference both at the low level and at the higher levels of the hierarchy. Significant locations such as goals or locations where the user frequently changes mode of transportation are learned from GPS data logs without requiring any manual labeling. We show how to detect abnormal behaviors (e.g. taking a wrong bus) by concurrently tracking his activities with a trained and a prior model. Experiments show that our model is able to accurately predict the goals of a person and to recognize situations in which the user performs unknown activities.[The full text can be downloaded from here]

I saw the presentation and the method is pretty cool, but also kinda scary. From 30 days of continuous GPS traces they were able to build a model of a user’s movements around Seattle, including the locations of his grocery store and several of his friends’ houses, and predict, in real-time, the user’s most-likely destination when travelling around the city.

The authors present the method as part of a handheld wireless assistant for the elderly that could, for example, tell them when they got on the wrong bus. To be fair, I’ve met Dieter Fox, and I don’t think the he or his students have any nefarious ulterior motives. That said, once something like this is published there’s nothing to stop, say, your cell-phone provider from implementing it and integrating it with their location services. Many of us, including me, are carrying around little network-connected GPS units in our pockets at all times. I have Sprint’s location services turned “off”, but my location is still available to 911, so I wonder if it’s really hidden from Sprint’s network.

Maybe David Brin is right, and technology will soon make privacy a thing of the past. He has an article in Salon this month on that topic, suggesting that the way for us to protect our liberty is not to jealously guard our privacy (a losing battle), but to make sure that we the people are empowered just as much to watch as to be watched.

Advertisements
Posted in AI. 3 Comments »

3 Responses to “Big Brother — Learning and Inferring Transportation Routines”

  1. daniel luke Says:

    Not to brag, but I’ve had this same idea for over a year. Unfortunately, it is something that has remained just a theory for me becasue my knowledge of CS is non-existent, and I wouldn’t have the first clue about how to conduct such an experiment. Having said that, there is a lot of stuff you can do with GPS data that people are only beginning to realize. I personally think it will usher in a whole new era of public transportation (where the automobile is an agent of public transportation), but everyone thinks I’m nuts. Below are my thoughts on the paper talked about in the post:

    I came to thinking about such matters while thinking about how to improve public transportation. In Portland, Oregon, where I live, there is a service called Flexcar which allows a person to rent an automobile for very short durations (as little as ½ hour). The problem with this service is that one-way trips are not permitted because cars must be returned to a specific, fixed location. One way trips are desirable because they would serve to reduce the overall expense of using the service, and they would make the service much more flexible. So I came up with a theoretical solution that relies on GPS so that cars are freed from fixed locations are thus able to make one-way trips.

    As it is, the cars in the Flexcar fleet are equipped with GPS, but this information derived from it is not supplied to the customer, but instead used for fleet tracking purposes. But I thought the information might be put to better use. I thought, for example, that the people who use this service might be interested in seeing, over time, where they have gone using Flexcars, how long they’ve stayed at a particular location, that sort of thing. Then I thought further that even more information could be obtained from this kind of location data if the locations were tagged according to activity. If GPS data shows that I’m parked in the parking lot of Safeway, there is a high probability that I’m there shopping. Over time, I reasoned, this information could be collated, quantified, and put into some kind of context. Believe it or not, I’ve even thought about how this information could form some kind of compatibility criteria for internet dating concept! Using such information, it could become very easy to determine who you live in proximity to who enjoys doing the same activities you do as measured by how much time they actually spend doing that activity.

  2. Jefferson Says:

    Interesting. I definitely agree with the idea that users should have access to the tracking data. Though it seems like you’d have to use Flexcars for pretty much all your travel in order for the data to build a decent model of your activities. Seems like cell-phone tracking would be more complete.

    Another idea we were talking about after seeing the paper presented at the conference is to track many people en masse and, say, automatically detect traffic jams and direct people to alternate routes.

    [btw, the authors have probably been thinking about this stuff for at least a year as well. The paper was submitted for review back in January, and I’m sure it took them a while before that to collect the data develop the model, debug it, analyse the results and write up the paper.]

  3. daniel luke Says:

    If everyone has GPS cellphones, that’s clearly the way to go for the reasons you mentioned. I got in to thinking about GPS and is possibilities while thinking about car sharing, and during a time when no phones had GPS. Now, of course you can get them with GPS or some kind of locative technology, but how long it will be before your whereabouts is acurately tracked via cellphones, and this data put into any kind of meaningful context such as I described is anyones guess.

    When there are a multiplicity of such devices, undoubtedly they will of great service for giving a very accurate picture of traffic. But I think that GPS and cellphones can be combined, actually, to change the whole paradigm of transporation. Wouldn’t it be great, for example, for all transportation to placed under one digital umbrella? It is perhaps easier to imagine the usefulness of this if you live in a city with a relatively dense urban core, and where public transportation is in abundance. I mentioned car sharing, and I certainly think a modified version of this could be quite useful. The concept basically boils down to driveless taxis. So ask yourself, with GPS, bluetooth, wi-fi, (or some other kind of portable broadband), and cellphones, why will taxis need drivers (assuming there are in a given population a sufficient pool of skilled drivers)? Imagine in the future transportation being a collection of services. Anyway, I ramble. For more information, check out my link: http://wirelessfuturenow.blogspot.com


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: