mining reality

[Eugen Leitl]

http://www.thefeature.com/article?articleid=101579

MIT researcher Nathan Eagle is helping mobile phones get to know you better
than you know yourself.

How much does your phone know about your life? Perhaps enough to predict your
behavior and even match you up with a mate, says Nathan Eagle, a graduate
student at the MIT Media Lab. As part of the Reality Mining project, Eagle
has collected approximately 40 years of continuous data on human behavior by
capturing communication, proximity, location and activity information from
100 cell phone-wielding subjects at his school. Hidden in that data are
insights about complex social systems that could affect our relationships
with our mobile phones and each other.

TheFeature: What is Reality Mining?

Eagle: We're at a point where we have the potential to capture an enormous
amount of data on everyday human life, on the reality of our behaviors.
Traditionally, that kind of data is collected through self-reported surveys
or sociologists' observations. But now we can gather much more objective data
on human behavior.

TheFeature: How do you gather the data?

Eagle: I primarily look at mobile phone data that can be broken down into
three types: location, communication and proximity patterns. We use cell
tower IDs to get approximate locations within a few blocks. Communication
logs reveal who is calling and texting whom and how often. And Bluetooth
scans every five minutes show who is proximate to you.

TheFeature: What new functionalities are enabled by this data?

Eagle: We can do behavior prediction. Depending on the life you lead, I can
predict what you're going to do next based on very limited information.
Whether it's your morning Starbucks fix or your Saturday afternoon softball
game, everyone lives life in routines. One of our algorithms extracts these
routine patterns from everyone's daily lives.

TheFeature: What makes someone's daily routines particularly interesting
though?

Eagle: There has been a lot of work on building more user-centric interfaces.
So the kind of data we gather could automatically change the phone
functionality according to a certain demographic. For example, Nokia, one of
the sponsors of this research, is selling the same phone to soccer moms,
power executives and texting teenagers. With just a few days worth of data,
we can characterize the user and their usage. Once we do that, we can
customize how the phone looks and operates for specific groups of people.

TheFeature: In what way?

Eagle: Here's a basic example: Out of 100 MIT students in my study, about 30
of them use the phone's clock as an alarm every day. When you first get the
phone, the clock functionality is 10 keystrokes down into the user interface.
Several users have figured out how to bring this functionality to the front,
but the rest of these MIT students have not. So every day, they spend the ten
keystrokes to set the alarm. By looking at the usage pattern, the phone could
assign a single button to the alarm clock.

TheFeature: How does your application Serendipity leverage Reality Mining
data?

Eagle: We're uncovering affiliations between people. I have a similarity
metric based on distance in behavior states. The end idea is that the
software would notice, say, that you typically hang out at the B-Side Lounge
on Friday nights. So do I and perhaps you also do other behaviors similar to
me. Those things in common may mean that we would want to be introduced.
That's one method of matchmaking. Another is based on proximity. The
Bluetooth addresses of those people running our client get pushed to our
server. Then we do a comparison based on their profiles.

TheFeature: It sounds like Friendster for the physical world?

Eagle: That's the general idea. But Serendipity is based not just on explicit
user profiles (that you enter) but also implicit behavioral information.

TheFeature: So even if I don't explicitly state in my profile that I go to a
particular club whenever a certain DJ is performing, the system would know
that anyway?

Eagle: Yes. And that would be weighted when matching you with other people.
Of course, it might not tell you all of the reasons why you're being
introduced to someone due to the privacy implications.

TheFeature: Serendipity seems like it could be mired in privacy issues.

Eagle: There's certainly a lot to talk about. When we were designing the
system, we talked with Match.com about the right way to pull something like
this off in a way that would keep people safe and not freak them out. To
avoid profile spoofing, we felt that there's a need for a trusted third party
to mediate the introductions. That way, I wouldn't be able to see anything
about who a nearby girl is or what she's looking for. There would just be
anonymous MAC addresses that get sent to our server. Then you might get
offered an introduction or not depending on the settings. But I do like the
idea of potentially broadcasting some of this information. I think of it as a
proximity Web page. Men might have the pages -- I have no problem revealing
my name or interests -- and women could go into a bar and do a social scan
and potentially meet someone.

TheFeature: What's your next step?

Eagle: As a researcher, I have hundreds of thousands of hours of continuous
human behavior and I've just scratched the surface analyzing it. This data
set was never available before, so I'm going to collaborate with a variety of
different disciplines -- computer scientists, physicists, social scientists,
people who are interested in organizational behavior and chaos theory and
others. I've also launched a startup, SenseSix, to commercialize Serendipty.
We're developing a client that will work on many more mobile devices.
Hopefully, we'll have our first product out by the end of summer.

--
Eugen* Leitl <a href="http://leitl.org">leitl</a>
______________________________________________________________
ICBM: 48.07078, 11.61144            http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
http://moleculardevices.org         http://nanomachines.net

[demime 1.01d removed an attachment of type application/pgp-signature which had a name of signature.asc]