[00:01.540 --> 00:02.120]  Hi, everyone.
[00:02.220 --> 00:02.660]  Thanks for coming.
[00:03.040 --> 00:04.040]  Again, my name is Paul.
[00:04.120 --> 00:04.640]  It's Paul Vett.
[00:04.780 --> 00:05.780]  I do have a last name.
[00:06.100 --> 00:07.000]  I'm a grad student.
[00:07.160 --> 00:08.840]  I just didn't give it to anybody at first.
[00:09.000 --> 00:13.560]  I'm a grad student in Toronto, and I'm looking at geotagging and social media.
[00:14.100 --> 00:18.320]  And in particular, I've been looking at Twitter, but I've been chatting with people about other services.
[00:18.700 --> 00:20.340]  And so I've been doing a project.
[00:20.340 --> 00:20.860]  It's a tool.
[00:21.020 --> 00:23.380]  It's a website that aggregates tweets.
[00:23.640 --> 00:26.900]  And I'm going to show you some examples of what I've been doing with it.
[00:27.040 --> 00:31.360]  I'm not going to demo it live because I don't trust the network with my laptop.
[00:31.700 --> 00:37.740]  But if you want to see it, feel free to come chat to me after, and we'll go somewhere, grab a beer, and play around with it.
[00:39.220 --> 00:41.680]  So the quick overview of what I'm going to talk about.
[00:42.580 --> 00:44.220]  I'm just going to give you a quick intro to social media.
[00:44.320 --> 00:48.380]  I'm sure everyone here knows what Twitter is, but I just want to make sure everyone's on the same page.
[00:48.880 --> 00:53.940]  Then I'm going to scare you with what I've been doing and make you rethink using geotagging if you've been doing it.
[00:54.000 --> 00:58.440]  And I know some people here at the conference have been, so maybe you'll rethink it after you see this.
[00:59.400 --> 01:03.800]  And then once I've done that and scared you, I'm going to try to cheer you up a bit with some good aspects of geotag.
[01:03.880 --> 01:04.820]  So it's not all bad news.
[01:04.880 --> 01:05.640]  And there's a balance.
[01:06.100 --> 01:08.280]  And I think it's important that we take a bit of a look at that.
[01:08.520 --> 01:10.720]  And then after that, again, I'll be around.
[01:10.880 --> 01:11.260]  We can chat.
[01:11.260 --> 01:11.980]  We can go for beers.
[01:14.260 --> 01:16.580]  So here are some typical social media sites.
[01:16.780 --> 01:19.900]  Like I mentioned, I'm focusing primarily on Twitter for my project.
[01:20.980 --> 01:28.160]  But I've chatted with people about Foursquare and Yelp, which are sites that let you check into locations or give reviews of restaurants.
[01:28.440 --> 01:30.520]  Things like that, which are basically social media.
[01:30.720 --> 01:33.220]  And again, you're tagging your location if you check in that you're at HOPE.
[01:33.420 --> 01:35.420]  And I think I saw that somebody is the mayor of HOPE.
[01:37.780 --> 01:38.900]  So that's one example.
[01:39.040 --> 01:42.680]  And then Flickr is another one that you might not think of as a geotagging social media site.
[01:42.800 --> 01:49.900]  But if you're taking lots of photos, you can embed your GPS coordinates in them if you're using an iPhone, for example.
[01:50.080 --> 01:52.580]  And then when you upload that to Flickr, it can expose that for you.
[01:52.880 --> 01:54.720]  And that's a good and a bad thing.
[01:54.720 --> 01:58.340]  And so there's one big social media site that's missing from this list.
[01:58.600 --> 02:00.060]  Kind of conspicuous in its absence.
[02:00.900 --> 02:02.680]  And I'll talk about it a bit later.
[02:02.960 --> 02:05.100]  It doesn't really have geotagging right now.
[02:05.540 --> 02:07.560]  But it's going to be a problem in the future, I think.
[02:09.040 --> 02:12.120]  So for somebody that just got their first smartphone, right?
[02:12.300 --> 02:13.600]  And they got an iPhone.
[02:13.840 --> 02:14.840]  It comes with all these apps on it.
[02:14.900 --> 02:16.760]  There's now an official Twitter for iPhone app.
[02:17.160 --> 02:20.160]  And they fire up their first tweet just because they're playing around with their new phone.
[02:20.260 --> 02:21.160]  They're like, I wonder what's going to happen.
[02:21.160 --> 02:22.640]  And it pops up this prompt here.
[02:22.940 --> 02:23.980]  It says, geotagging disabled.
[02:24.240 --> 02:24.840]  What do you want to do?
[02:25.220 --> 02:27.180]  And so configure is the default option.
[02:27.600 --> 02:28.960]  And the other one is cancel.
[02:29.120 --> 02:35.020]  So clearly the typical person who just wants to get rid of dialog boxes as fast as possible is just going to click configure.
[02:35.200 --> 02:36.560]  And then it's going to say, do you want to do this?
[02:36.620 --> 02:37.720]  And they're going to be like, yes, go away.
[02:37.840 --> 02:38.440]  Stop asking me.
[02:38.480 --> 02:39.500]  I want to tweet about my meal.
[02:40.200 --> 02:42.300]  And so that's going to be the typical case, right?
[02:43.400 --> 02:45.200]  But that's not the only case, right?
[02:45.200 --> 02:51.760]  I mean, other people, like, I'm sure there are a lot of gadget-aholics in this room right now, right?
[02:51.940 --> 02:57.100]  I mean, I think somebody asked for a show of smartphone hands earlier, and there were a lot of them, right?
[02:57.280 --> 03:05.660]  So even if you're aware that there are implications to these decisions you're making, or if you're consciously making them, it might not be for the right reasons.
[03:05.720 --> 03:09.240]  And I did a bunch of interviews as I was talking with people about geotagging.
[03:09.240 --> 03:12.380]  And this is kind of a typical example of why people start.
[03:13.640 --> 03:16.720]  I started to get into it when I got the iPhone 3G.
[03:17.040 --> 03:17.900]  I heard about Twitter.
[03:18.020 --> 03:21.600]  I signed up when I got the phone itself, and I figured might as well try it out.
[03:22.520 --> 03:23.260]  So there you go.
[03:23.360 --> 03:24.300]  Might as well try it out.
[03:24.440 --> 03:29.680]  That's the typical train of thought that that's as hard as people think about geotagging before they start using it, typically.
[03:30.220 --> 03:31.900]  And so does he understand the implications?
[03:32.140 --> 03:36.000]  He's a tech podcaster, so I'm sure he has some sense of what's going on.
[03:36.200 --> 03:38.380]  But does everybody who makes this ooh, shiny decision?
[03:38.380 --> 03:38.720]  No.
[03:38.900 --> 03:39.500]  No, they don't.
[03:40.560 --> 03:43.440]  So a brief timeline of geotags on Twitter.
[03:44.000 --> 03:47.100]  Originally, Twitter just let you put in a location field, and you could set it to whatever you want.
[03:47.220 --> 03:47.800]  It was just text.
[03:47.960 --> 03:49.340]  So my location was Toronto.
[03:49.680 --> 03:52.200]  Other people had put in a city or a country or a state.
[03:52.620 --> 03:56.060]  And eventually, clients started updating this with GPS coordinates.
[03:56.380 --> 03:59.020]  And so that's when I started getting the idea that this might be worth looking at.
[03:59.100 --> 04:05.240]  And I thought I was going to have to scrape Twitter and extract this field and figure out who was actually updating it and when it was getting updated.
[04:05.800 --> 04:13.060]  But luckily for me, later on in 2009, they added metadata to each tweet that you could embed a GPS coordinate in it.
[04:13.180 --> 04:19.480]  And so that made it much, much easier for me for this project because I could just get the tweets and then pull the metadata right out of them.
[04:21.100 --> 04:23.920]  And originally, there was no way to ask Twitter just for geotagged tweets.
[04:23.980 --> 04:25.360]  You got all the tweets, no matter what.
[04:25.460 --> 04:27.300]  And you had to discard them and filter them on your own.
[04:27.440 --> 04:29.100]  And eventually, the APIs got richer.
[04:29.120 --> 04:32.880]  And I stopped having to get gigs of data for the occasional tweet, which was super.
[04:33.920 --> 04:35.420]  So this is on the BlackBerry.
[04:35.540 --> 04:40.920]  This is what happens when you enable or when you try to set up Uber Twitter, which is the client I use for geotagging.
[04:41.100 --> 04:43.080]  And you'll see that it has a bunch of options.
[04:43.220 --> 04:46.280]  It gives you some choices as to how much you want to share and how accurate you want it to be.
[04:46.720 --> 04:50.980]  So for instance, I've picked the second one, not the cell tower, but the assisted GPS setting.
[04:51.380 --> 04:53.940]  And here's a geotag tweet I did at a concert.
[04:54.520 --> 04:58.800]  And so the push pin is where the microphone reported me as being.
[04:58.840 --> 05:01.320]  And that red dot in the corner is where I actually was.
[05:01.320 --> 05:03.580]  So you can see that I'm about three streets off.
[05:03.800 --> 05:05.520]  So that's pretty inaccurate, right?
[05:05.700 --> 05:07.480]  And that may be reassuring to somebody.
[05:07.660 --> 05:10.840]  But as we're going to find out, it's not actually that reassuring.
[05:11.540 --> 05:13.440]  Because one tweet sure is inaccurate.
[05:13.640 --> 05:16.360]  But many tweets start to add up to a little more accuracy.
[05:16.980 --> 05:19.380]  And so this is what the tweet looks like from the API.
[05:19.600 --> 05:22.500]  You can see it's just got the GPS data right in there.
[05:22.640 --> 05:24.280]  And it's super easy to use.
[05:24.440 --> 05:25.620]  Twitter just gives you everything you need.
[05:27.040 --> 05:28.580]  So why is this a big deal, right?
[05:28.680 --> 05:29.440]  Like, who cares?
[05:30.040 --> 05:33.180]  I took a photo of myself eating ice cream, geotagged that, posted that.
[05:33.320 --> 05:33.960]  What does it matter?
[05:34.040 --> 05:35.660]  Who cares where I get my ice cream from, right?
[05:35.840 --> 05:36.620]  Like, nobody, right?
[05:36.820 --> 05:38.020]  But there are reasons to care.
[05:38.140 --> 05:38.400]  I don't know.
[05:38.520 --> 05:40.820]  Can anyone speculate as to why this might be a big deal?
[05:41.320 --> 05:43.040]  And just shout it out because I can't see anything.
[05:43.540 --> 05:43.860]  Stalking.
[05:44.200 --> 05:44.520]  Right?
[05:44.760 --> 05:45.660]  Stalking's a good one, yeah.
[05:45.880 --> 05:46.360]  The man.
[05:46.880 --> 05:47.940]  The man, all right.
[05:48.860 --> 05:49.420]  Robbery, yeah.
[05:49.520 --> 05:50.180]  Yeah, so exactly.
[05:50.340 --> 05:51.320]  So there are all these good reasons.
[05:51.420 --> 05:51.820]  No ice cream.
[05:52.640 --> 05:52.960]  Exactly.
[05:53.340 --> 05:54.360]  That would be brutal.
[05:54.840 --> 05:55.020]  All right.
[05:56.980 --> 05:59.340]  So, and this came up in my interviews again.
[05:59.540 --> 06:12.140]  And so the same podcaster I was talking to before has a little anecdote that he gave me that hit it right on the head and exactly what you guys have been saying and exactly what worried me when I first heard about this.
[06:12.880 --> 06:15.960]  One of the guys that I know, he's a friend of mine from Henry's.
[06:16.600 --> 06:18.660]  We do some of the work together for the video stuff.
[06:19.240 --> 06:26.780]  He tweeted me a few weeks ago saying, well, yeah, I'm looking at your tweets and by the tweets, let's see, I think you live in this area of Mississauga.
[06:26.980 --> 06:27.180]  Right.
[06:27.300 --> 06:28.760]  Which, you know, for me, it's not a problem.
[06:28.980 --> 06:30.200]  I'm pretty open with what I'm doing.
[06:30.340 --> 06:35.540]  But I can see that becoming an issue with some people that might not be aware that they're sharing that sort of data.
[06:37.620 --> 06:41.400]  So what he's saying there is basically that one geotag is just an anecdote.
[06:41.680 --> 06:43.280]  It's one little tidbit.
[06:43.440 --> 06:45.580]  But many geotags put together become data.
[06:45.760 --> 06:47.480]  You can look at them and you can make inferences.
[06:48.840 --> 06:50.840]  And so this is what...
[06:50.840 --> 06:53.580]  I basically wanted to find a way to illustrate this to people.
[06:53.720 --> 06:53.820]  Right.
[06:53.840 --> 06:55.480]  This is a kind of a nebulous concept to grasp.
[06:55.640 --> 06:56.260]  It's kind of abstract.
[06:56.260 --> 06:59.020]  You're thinking, well, maybe people can do something with my tweets.
[06:59.160 --> 07:00.940]  But you're not really sure what.
[07:01.120 --> 07:03.860]  And I'm talking about Twitter specifically because that's what I looked at.
[07:03.860 --> 07:07.000]  But this is true of other geotagging services as well.
[07:08.460 --> 07:11.780]  So I tried to figure out how I could express this to people.
[07:11.960 --> 07:13.040]  And most people are visual.
[07:13.180 --> 07:14.160]  Or a lot of people think visually.
[07:14.320 --> 07:16.280]  And it tends to make a big impact on people.
[07:16.480 --> 07:20.240]  So I decided to start collecting geotagged tweets and do some analysis on them.
[07:20.320 --> 07:21.140]  And then visualize that.
[07:21.300 --> 07:23.460]  Put it on the screen so people can say, whoa, this is scary.
[07:23.640 --> 07:27.020]  And my goal ultimately is to creep people out and to make them think of what they're doing.
[07:29.140 --> 07:29.640]  So, yeah.
[07:29.860 --> 07:33.840]  And on top of that, right, people are already doing this.
[07:33.860 --> 07:36.920]  It's the same reason we, like, for full disclosure, right?
[07:37.660 --> 07:41.260]  People are doing this and they're not telling anybody.
[07:41.260 --> 07:42.460]  But that doesn't mean it's not happening.
[07:42.640 --> 07:45.360]  So my goal is to be the person who tells people that this is going on.
[07:45.420 --> 07:46.580]  I want to shine some light on it.
[07:47.360 --> 07:50.980]  And I want people to realize that they're making that choice with long-term consequences.
[07:51.200 --> 07:52.520]  Confront it and make it consciously.
[07:52.600 --> 07:53.820]  And maybe they'll choose to share.
[07:54.000 --> 07:55.400]  And if they do, that's up to them.
[07:55.480 --> 07:56.720]  But at least they'll be fully informed.
[07:58.520 --> 08:08.580]  So an early experiment, this was an early prototype, I was just looking at grabbing tweets and plotting them just to get a sense for what was going on.
[08:08.900 --> 08:10.860]  And this person's asking if the feature is creepy.
[08:10.900 --> 08:12.140]  And I think the answer is yes.
[08:12.400 --> 08:14.160]  I absolutely think it's a creepy feature.
[08:14.920 --> 08:23.120]  So at this point, I was just grabbing from what Twitter calls the garden hose, which they have a streaming API so you can just connect to it and they'll just fire some percentage of the tweets your way.
[08:23.560 --> 08:26.940]  And I was finding about 1% of the tweets were geotagged at this point.
[08:27.020 --> 08:28.980]  This was just eight months ago or something.
[08:29.100 --> 08:30.300]  And it's gone way up since then.
[08:30.860 --> 08:36.040]  But like I said earlier, I was getting gigs of bandwidth just for megs of tweets and it was unsustainable.
[08:37.840 --> 08:41.260]  So the first step was to build on that and use filtering.
[08:41.540 --> 08:43.800]  So they added an ability to filter based on locations.
[08:44.100 --> 08:45.060]  So that's what I did.
[08:45.160 --> 08:47.160]  I just asked for geotagged tweets in certain locations.
[08:47.300 --> 08:48.580]  I asked for three cities in particular.
[08:48.960 --> 08:51.980]  Toronto, which is where I am, New York, and San Francisco.
[08:52.220 --> 08:57.420]  So I've been collecting geotagged tweets from those locations for maybe six months now.
[08:58.800 --> 09:02.840]  And except when the process crashes and I'm not around to restart it.
[09:03.200 --> 09:06.220]  I have pretty much the entire sample for that period in these cities.
[09:06.900 --> 09:09.980]  And so that's about one tweet a second, which is what I'm getting.
[09:10.200 --> 09:15.660]  Which doesn't sound like a lot, but over the course that this has been running, that's about 3.5 gigs of tweets.
[09:18.700 --> 09:24.440]  So here is an example of just one user that I was getting tweets from that I'm going to keep coming back to a bit.
[09:24.800 --> 09:28.300]  This is an early demo, just grabbing the user, throwing their tweets on a map.
[09:28.740 --> 09:30.540]  And what I did was I just, I asked the database.
[09:30.700 --> 09:32.160]  I was like, who's tweeting a lot in Toronto?
[09:33.220 --> 09:34.360]  What are the top 10 users?
[09:34.460 --> 09:35.180]  And then I went through them.
[09:35.280 --> 09:37.380]  And there were a couple bots, which was kind of interesting.
[09:37.380 --> 09:40.460]  Like a fire service was telling you where the fires were, which was kind of neat, I thought.
[09:40.740 --> 09:42.940]  But then this guy came up, and he makes a perfect example.
[09:43.080 --> 09:47.300]  You can see right on the screen right now a couple of clusters just by looking at it, right?
[09:48.780 --> 09:50.540]  There's three clusters that are pretty obvious.
[09:50.700 --> 09:55.220]  And they all have different amounts of tweets, so you can judge how much time the person's spending at each of these locations.
[09:55.640 --> 10:01.940]  And then if you look at the text of what they're saying, so the text that I've highlighted here, this guy's just finished work.
[10:02.040 --> 10:04.520]  So you have a sense that this tweet was probably at work in the parking lot.
[10:04.520 --> 10:07.400]  And there are some other tweets that corroborate that story.
[10:08.080 --> 10:10.100]  So this is kind of scary to look at.
[10:10.200 --> 10:12.040]  If somebody was just like, here's a map of where you live, right?
[10:12.080 --> 10:13.080]  You'd be like, well, that's weird.
[10:13.500 --> 10:16.680]  But I think we can do a little better with technology.
[10:17.020 --> 10:19.920]  So I thought I would teach the computer to do this analysis on its own.
[10:21.140 --> 10:24.060]  So I spent a while trying to learn about clustering algorithms.
[10:24.240 --> 10:25.500]  And it turns out it's a complicated field.
[10:26.060 --> 10:27.540]  And it's an MP-hard problem.
[10:28.040 --> 10:29.560]  And there are various heuristics.
[10:29.560 --> 10:31.340]  And I spent, you know, days reading Wikipedia.
[10:31.520 --> 10:32.780]  And Wikipedia is just not coding.
[10:33.100 --> 10:34.100]  It's just not.
[10:34.520 --> 10:35.660]  So one night I was talking to my friend.
[10:35.780 --> 10:37.080]  And I was like, I have this demo tomorrow.
[10:37.640 --> 10:38.980]  And I'm not done clustering.
[10:39.300 --> 10:40.220]  And it's an MP-hard.
[10:40.520 --> 10:41.600]  And I don't know which algorithm to use.
[10:41.680 --> 10:42.580]  And he was like, just come over.
[10:43.180 --> 10:44.280]  And we'll get this done.
[10:44.380 --> 10:45.280]  So I went to his place.
[10:45.320 --> 10:46.500]  And we just hacked away that night.
[10:46.860 --> 10:49.820]  And my buddy Riz came up with this solution right here.
[10:50.140 --> 10:52.220]  Which is, again, I said this is an MP-hard problem.
[10:52.260 --> 10:53.320]  This is the brute force solution.
[10:54.860 --> 10:55.480]  But it works.
[10:55.820 --> 10:57.120]  Which is all that matters.
[10:57.960 --> 10:59.160]  So again, it's not efficient.
[10:59.500 --> 11:00.760]  But it gets the job done.
[11:01.220 --> 11:04.640]  And there will be a lot of clusters with just one tweet, according to this algorithm.
[11:04.780 --> 11:06.600]  But I just ignore them because they're uninteresting.
[11:07.180 --> 11:08.940]  So you can see the results here on the map.
[11:09.440 --> 11:10.500]  So it's pretty successful.
[11:10.760 --> 11:12.200]  A bunch of clusters have been highlighted.
[11:12.380 --> 11:13.460]  The big one in the middle there.
[11:15.100 --> 11:16.180]  And it's not perfect.
[11:16.640 --> 11:20.460]  You can see there's some overlap and some bugs to sort out.
[11:20.740 --> 11:23.420]  But for the most part, for the most part, it works.
[11:23.480 --> 11:24.120]  And it works well enough.
[11:25.940 --> 11:29.220]  So the next step, I thought, is like, so we have these clusters.
[11:29.340 --> 11:29.820]  And that's scary.
[11:29.920 --> 11:32.340]  It can be like, here's where you spend the most of your time automatically.
[11:33.180 --> 11:35.780]  But how can we make this even creepier for people?
[11:35.860 --> 11:37.160]  Even more visceral.
[11:37.300 --> 11:39.320]  So they're like, wow, I really need to think about what I'm doing.
[11:40.080 --> 11:43.180]  And the next step, then, is to automatically classify the locations.
[11:45.500 --> 11:53.020]  So, again, I pulled up Wikipedia's documents on document classification and read about filtering and probable models and AIs.
[11:53.240 --> 11:54.100]  And I was like, you know what?
[11:54.100 --> 11:54.520]  Forget this.
[11:54.660 --> 12:00.260]  I just pulled up some of the big clusters that I knew, that I could tell by inspection where somebody was at home.
[12:00.680 --> 12:01.840]  And I just looked for keywords.
[12:02.100 --> 12:04.420]  So you can see, these are some examples.
[12:04.560 --> 12:05.880]  Somebody was at home with their dog a lot.
[12:05.920 --> 12:08.320]  So I thought maybe people only tweet about their dog when they're home.
[12:08.680 --> 12:10.180]  People tweeting about sleep in bed.
[12:10.360 --> 12:12.200]  That's the sort of thing that people do when they're at home.
[12:12.320 --> 12:13.340]  Or about being awake at night.
[12:14.240 --> 12:18.400]  And there are other types of locations that you can identify by keyword, too, right?
[12:18.640 --> 12:20.360]  So, at so-and-so's house.
[12:20.460 --> 12:22.800]  I mean, it doesn't get easier than that, right?
[12:24.120 --> 12:25.400]  And watching TV, right?
[12:25.460 --> 12:26.560]  Nobody watches TV at work.
[12:26.560 --> 12:27.520]  You watch TV at home.
[12:27.880 --> 12:29.260]  Unless it's the World Cup, I guess.
[12:31.420 --> 12:38.480]  So, what I'm doing now is, again, when I try to get a user's report, I have all the cluster data available.
[12:38.640 --> 12:44.740]  I scan through each cluster, look for these keywords that I've manually identified, and then I just assign a score to each cluster.
[12:44.880 --> 12:46.420]  And one cluster tends to win.
[12:48.720 --> 12:55.240]  So if you look at the tweets in this cluster, where there are hundreds of tweets, it's quite clearly where this person lives.
[12:55.440 --> 12:58.680]  And if you zoom in, you can pinpoint it pretty accurately.
[12:58.960 --> 13:00.840]  And you could reverse geocode that if you wanted.
[13:00.900 --> 13:03.320]  I'm not at this point, because that's not really my goal.
[13:03.880 --> 13:05.300]  But it's pretty scary, right?
[13:05.460 --> 13:07.160]  And so this person isn't consciously...
[13:07.160 --> 13:08.840]  He's not like, hey, come to my...
[13:08.840 --> 13:09.680]  You know, I'm at home.
[13:09.680 --> 13:10.240]  It's great.
[13:10.380 --> 13:11.600]  And taking pictures of their house or something.
[13:11.700 --> 13:15.520]  They're just tweeting about casual, mundane things that are going on in their life.
[13:15.780 --> 13:20.360]  And based on that, it's enough to extrapolate the type of location.
[13:21.520 --> 13:28.960]  And I haven't done it, but this could be easily extended, as I showed you, to workplaces and to friends' houses and to...
[13:28.960 --> 13:30.100]  I don't know where else do people go.
[13:30.540 --> 13:30.900]  Anything.
[13:31.360 --> 13:33.200]  But I thought this is an example from Toronto.
[13:33.580 --> 13:36.360]  And most people here aren't going to be from Toronto.
[13:36.480 --> 13:37.940]  So that's maybe kind of boring for you.
[13:38.240 --> 13:39.320]  So I thought we were in New York.
[13:39.520 --> 13:40.240]  Let's look at New York.
[13:40.820 --> 13:44.860]  So I got here Thursday night, set up in a coffee shop, and just let it map.
[13:45.040 --> 13:48.800]  And I have it mapping in real time, so I just let it collect tweets for a while.
[13:49.320 --> 13:51.100]  And this is about an hour's worth of tweets.
[13:53.980 --> 13:56.260]  And you can see that it's pretty busy here in New York.
[13:56.400 --> 13:57.000]  Lots of people are tweeting.
[13:57.140 --> 13:57.680]  Lots going on.
[13:59.960 --> 14:05.900]  And so what I did is I just kind of clicked at them, and I'd get a username, and I'd be like, oh, this person looks like they might tweet a lot.
[14:05.980 --> 14:07.120]  And I'd check out their Twitter page.
[14:07.740 --> 14:08.900]  Maybe they did, maybe they didn't.
[14:09.100 --> 14:14.300]  And then I just did a database query again, you know, aggregated everybody's tweets.
[14:14.300 --> 14:17.600]  And I was like, who are the most prolific New York tweeters that are geotagging?
[14:18.220 --> 14:20.760]  And so one example came up, and I thought I'd just run with it.
[14:21.080 --> 14:26.220]  So I put together this example on Thursday in just a little bit of time.
[14:26.320 --> 14:29.400]  So this is this person's Twitter tag, or Twitter page.
[14:29.540 --> 14:30.780]  And you can see a bunch of tweets there.
[14:30.940 --> 14:36.400]  And what's interesting, I find, about this example is that on this Twitter page, this person has thousands of tweets.
[14:36.780 --> 14:38.840]  None of these tweets here are geotagged.
[14:40.220 --> 14:43.560]  So only a small fraction of this person's tweets are geotagged.
[14:43.780 --> 14:45.280]  Not a significant number at all.
[14:45.460 --> 14:49.780]  And that's still enough that we can extrapolate lots of details about them.
[14:50.780 --> 14:52.280]  So I ran it through the system.
[14:52.880 --> 14:55.500]  I just plotted it, and you can kind of visually see...
[14:55.500 --> 14:58.320]  If you zoomed out, you'd see that this is a particularly popular cluster.
[14:59.700 --> 15:03.560]  The system works it into areas automatically.
[15:04.260 --> 15:05.960]  And then I decided to inspect the tweets.
[15:05.960 --> 15:08.920]  I was like, what type of location is this, just by hand?
[15:09.240 --> 15:10.580]  And I don't know if you can read that.
[15:10.940 --> 15:13.900]  But it says, wide awake at an hour when I wish I were wide asleep.
[15:14.220 --> 15:15.040]  So that's...
[15:15.040 --> 15:17.800]  This guy's in bed, can't sleep, and is tweeting from home.
[15:18.680 --> 15:23.600]  So, again, this is an interesting example because the automatic classification failed, right?
[15:23.720 --> 15:24.920]  There's no little home icon there.
[15:26.300 --> 15:29.320]  So that's an illustration that there's room to improve the algorithm.
[15:29.640 --> 15:30.980]  And maybe I missed some keywords.
[15:31.160 --> 15:34.060]  It actually tried to take a house somewhere else for him that was wrong.
[15:34.800 --> 15:35.820]  So that's interesting.
[15:35.980 --> 15:37.120]  It means there's more work for me to do.
[15:37.660 --> 15:39.040]  And so another thing I did...
[15:39.040 --> 15:43.560]  I was sitting around Friday afternoon between talks, and I thought I'd try to make it more interesting.
[15:43.800 --> 15:50.800]  So I added keyword support, and I just searched for the next hope and looked for people who were tweeting about next hope.
[15:51.740 --> 15:54.160]  And one that came up that was my favorite was this one.
[15:54.820 --> 15:56.980]  Ben Jackson tells me why I shouldn't tweet this picture.
[15:57.500 --> 16:02.080]  That was a talk about geotags in pictures and learning too much about people.
[16:03.840 --> 16:05.600]  But then I just asked the database.
[16:05.760 --> 16:13.420]  I was like, all right, which of these new hope or next hope tweeters have lots of tweets that are in New York, San Francisco, or Toronto?
[16:14.340 --> 16:15.620]  And I came up with one.
[16:15.980 --> 16:19.520]  Michael, I don't know if you're in the room, but I'd love to talk with you after.
[16:22.020 --> 16:25.300]  So here are just some tweets kind of in the hotel pen area.
[16:25.840 --> 16:31.960]  And so if I zoomed out a bit, though, there were tweets all over the New York City greater area.
[16:32.140 --> 16:35.160]  And in particular, I know that Michael likes hanging out on the beach.
[16:35.940 --> 16:40.580]  And what I also know, the system did make a guess as to try to classify his house.
[16:40.860 --> 16:43.860]  And I'm not going to show it, because he may be in the room.
[16:44.440 --> 16:47.340]  But again, if you're here, Michael, I'd love to chat with you after and see if I'm right.
[16:47.800 --> 16:49.660]  And if not, see if I can improve it.
[16:53.000 --> 16:55.520]  So the result... I mean, the system, it works really well, right?
[16:55.760 --> 16:59.640]  Like, I showed you a case where it nailed it spot on.
[16:59.760 --> 17:03.220]  And again, if you zoom in, you can get right into the crescent, right to the house itself.
[17:04.600 --> 17:07.060]  And that's the beauty of data, right?
[17:07.120 --> 17:10.900]  Like I said, one geotag is just an anecdote, but many are data.
[17:11.480 --> 17:13.560]  And you can use that to even out the errors.
[17:13.780 --> 17:22.240]  Because if your GPS error is somewhat consistent and truly random, then the center of all those tweets is the actual true center over time, right?
[17:22.340 --> 17:26.920]  So even if you're off by 500 meters, if you tweet 1,000 times, that's enough to really pinpoint it down.
[17:27.960 --> 17:30.960]  And so you have to, like, keep in mind that I'm just some guy, right?
[17:31.080 --> 17:33.060]  Like, I've been trying to do classes.
[17:33.180 --> 17:34.520]  I've been working on a thesis.
[17:34.720 --> 17:36.380]  I've been going to concerts, apparently.
[17:36.380 --> 17:40.200]  And, like, just trying to find time to work on this as my project.
[17:41.000 --> 17:42.620]  And it works pretty reasonably.
[17:43.060 --> 17:45.960]  And so what could somebody with resources do, right?
[17:47.140 --> 17:51.860]  Paying multiple programmers, paying people to classify what different types of locations are, right?
[17:52.200 --> 17:54.660]  What could people with untold budgets do?
[17:55.060 --> 17:56.040]  No accountability.
[17:57.380 --> 17:59.580]  And it goes further than that.
[18:00.020 --> 18:03.420]  I did this all on data that people chose to make public, right?
[18:03.640 --> 18:09.140]  I mean, nobody forced them to post their geotagged tweets, although the dialogues encouraged them to.
[18:10.940 --> 18:13.260]  But other organizations have way more data.
[18:13.720 --> 18:14.880]  So if you...
[18:14.880 --> 18:16.260]  Just thinking about Twitter, right?
[18:16.500 --> 18:22.300]  If you tweet via SMS, your cell phone company knows where that came from and can tag that with the location.
[18:22.300 --> 18:25.400]  The NSA knows where that came from and can tag it with your location.
[18:25.780 --> 18:27.440]  But it goes past tweets, right?
[18:27.600 --> 18:28.120]  So it goes...
[18:28.120 --> 18:30.580]  I mean, what if you text message somebody?
[18:30.580 --> 18:33.540]  I don't have access to that, and I can't do this sort of analysis with it.
[18:33.720 --> 18:34.760]  But there's no reason...
[18:34.760 --> 18:38.020]  The NSA, we know they have wiretaps everywhere, right?
[18:38.200 --> 18:39.180]  It's just a fact.
[18:39.560 --> 18:41.480]  And there's nothing stopping them from...
[18:41.480 --> 18:42.800]  And they know your location.
[18:42.980 --> 18:51.980]  So there's nothing stopping them from building a similar database with all these other disparate pieces of information they have and just having that much more data to do the analysis with.
[18:52.300 --> 18:55.120]  And I think that's kind of terrifying.
[18:56.800 --> 18:58.400]  But again, what's the big deal?
[18:58.540 --> 19:00.440]  Like, where you live is public knowledge.
[19:00.860 --> 19:05.620]  Most of us are probably in the phone book, and there are credit reports on us, and there's all sorts of information.
[19:06.160 --> 19:07.520]  So why is this a big deal?
[19:08.680 --> 19:09.520]  A couple reasons.
[19:09.880 --> 19:15.300]  First, I mean, again, like I said, as you post more and more, it stops being trivial facts.
[19:15.580 --> 19:18.280]  It starts being something that you can work with and analyze.
[19:18.980 --> 19:21.880]  And more important, it's not just, you know, this is where you live.
[19:22.140 --> 19:23.920]  It's also what you do while you're at home.
[19:24.160 --> 19:27.540]  It's who you're with, because maybe you tag your friend as watching the show with you.
[19:28.960 --> 19:34.060]  Maybe you don't tag your friend, but they're tweeting at the same time, too, and they have a cluster of tweets at your house.
[19:34.380 --> 19:36.280]  Now we know that there's this connection there.
[19:38.020 --> 19:40.900]  Maybe you went to a hacker con, and you tweeted from there.
[19:41.120 --> 19:43.060]  And then there were overlapping tweets with other people.
[19:43.200 --> 19:45.060]  Now, suddenly, you're on lists, right?
[19:45.240 --> 19:47.300]  I mean, who knows what's happening with this?
[19:47.920 --> 19:50.420]  And it's not ephemeral, right?
[19:50.620 --> 19:52.500]  So a conversation takes...
[19:52.500 --> 19:59.520]  We could have a conversation, and then it's done, and maybe the person forgets what we talked about, or maybe they remember, but they don't tell anybody.
[19:59.520 --> 20:09.060]  But every tweet, and this is true in particular for Twitter, but likely for other data on the Internet, basically lives forever at this point.
[20:09.380 --> 20:12.940]  I mean, it's a fact that Google is archiving all of Twitter.
[20:13.120 --> 20:14.260]  It's one of their projects.
[20:14.480 --> 20:16.860]  The Library of Congress is archiving all of Twitter.
[20:17.120 --> 20:20.080]  I'm archiving all of these geotag tweets, right?
[20:20.540 --> 20:22.960]  Like, it's not just big organizations.
[20:23.460 --> 20:23.940]  It's easy.
[20:24.180 --> 20:27.220]  And so big organizations are certainly doing it.
[20:27.900 --> 20:31.040]  And another thing that's interesting is finding where somebody lives.
[20:31.200 --> 20:33.780]  Like, it's not interesting if you know somebody's full name to find where they live.
[20:33.900 --> 20:34.500]  That's pretty easy.
[20:34.840 --> 20:36.840]  But Twitter, we don't know their real names, right?
[20:37.280 --> 20:37.660]  Necessarily.
[20:37.880 --> 20:42.980]  Somebody posting, thinking they're anonymous, but geotagging is giving away their location.
[20:43.120 --> 20:45.260]  And based on their location, you can find out their real name.
[20:45.400 --> 20:47.040]  You can find out things about their real life.
[20:47.860 --> 20:49.420]  And, I mean, that's scary.
[20:51.820 --> 20:57.080]  So I'm going to give you a real example of sort of the bigger picture why this is a problem.
[20:57.220 --> 20:59.720]  Rather than just the, we know where this guy lives, right?
[21:01.000 --> 21:02.500]  So we found his house.
[21:03.160 --> 21:08.380]  We could reverse geocode it and get an address if we really wanted to.
[21:08.880 --> 21:09.720]  And guess what?
[21:10.500 --> 21:11.820]  He lives with his brother.
[21:12.420 --> 21:13.680]  So I'm assuming he lives at home.
[21:15.200 --> 21:17.020]  So now we have this connection, right?
[21:17.240 --> 21:19.020]  It's not just him that we have information on.
[21:19.280 --> 21:20.380]  It's also his brother.
[21:21.820 --> 21:24.200]  But it goes further than that because he tweets about his brother.
[21:24.200 --> 21:25.360]  He doesn't really like his brother.
[21:28.940 --> 21:30.340]  He really doesn't like his brother.
[21:30.440 --> 21:32.100]  And we know his brother breaks the law.
[21:34.860 --> 21:40.940]  So if the police had a quota of drug busts for the month that they need to do, they now have somebody to find, right?
[21:41.480 --> 21:43.100]  And it goes further than that too, right?
[21:43.200 --> 21:51.740]  Because even if the police don't care about this, and hopefully they don't, maybe his brother in 20 years decides to run for government.
[21:52.160 --> 21:52.820]  I don't know.
[21:54.340 --> 21:59.080]  And again, this isn't really a big deal, but when you're running for office, the smallest thing can burn you, right?
[21:59.420 --> 22:05.720]  And so an enterprising journalist could look through the archive of Twitter, could look for tweets at this location, and bam, has some...
[22:05.720 --> 22:08.940]  or maybe it's not a journalist, maybe it's the other political party, right?
[22:09.340 --> 22:10.840]  And so then you can say, oh, he's a pothead.
[22:11.080 --> 22:15.380]  Would you want a pothead making your children's decisions or something like that, right?
[22:15.540 --> 22:16.780]  And just think about the children.
[22:17.000 --> 22:17.680]  There you go, right?
[22:17.780 --> 22:19.180]  There's a smear campaign right there.
[22:21.420 --> 22:22.560]  So that's what worries me.
[22:22.900 --> 22:24.780]  It's when you put it all together.
[22:26.340 --> 22:27.880]  So what more do I want to do?
[22:29.580 --> 22:31.760]  Technically, I want to get more tweets.
[22:31.940 --> 22:33.040]  Right now I'm just looking at three cities.
[22:33.180 --> 22:41.520]  I'd love to do it for the world or have users that aren't in those cities at least be able to say, give me my data and just retrieve it asynchronously.
[22:41.820 --> 22:45.920]  And that means I need better storage because, like I said, there are a lot of tweets.
[22:46.160 --> 22:49.200]  And I'm on shared hosting and they're going to get very upset if I keep this up.
[22:51.360 --> 22:52.680]  I'd like to improve the clustering.
[22:52.820 --> 22:56.260]  I showed you that there are sometimes overlaps and little issues like that.
[22:56.360 --> 22:58.720]  And with better clustering, we could just do better analysis.
[22:59.500 --> 23:00.900]  And I'd like to do better classification.
[23:00.960 --> 23:02.780]  Like I said, I showed you three examples.
[23:02.960 --> 23:04.280]  One where I know it got it right.
[23:04.460 --> 23:06.200]  One where we know it got it wrong.
[23:06.740 --> 23:08.140]  And one where I'm not sure.
[23:08.800 --> 23:10.540]  Again, Michael, come talk to me after.
[23:11.900 --> 23:14.000]  So I want to use a better model.
[23:14.800 --> 23:16.680]  I'd basically want to use spam filtering.
[23:16.820 --> 23:23.200]  Spam filters can tell within a reasonable error if a message is spam based on not too much information.
[23:23.460 --> 23:26.920]  And I think you could train something to do the same for these different types of locations.
[23:28.540 --> 23:30.180]  And some time-based analysis.
[23:30.500 --> 23:32.960]  So this is a map of when I tend to tweet.
[23:34.940 --> 23:37.060]  So typically noon to the afternoon.
[23:37.260 --> 23:38.880]  So you might get a sense that I'm a late riser.
[23:39.320 --> 23:41.920]  That I sleep at night so that I don't work the night shift.
[23:42.100 --> 23:43.600]  You know when I'm asleep, probably.
[23:43.680 --> 23:44.780]  Because I'm not tweeting when I'm asleep.
[23:45.460 --> 23:48.660]  And so this sort of analysis you can factor into locations too, right?
[23:48.840 --> 23:50.180]  And just give more data.
[23:50.300 --> 23:53.980]  If every Thursday I tweet something like, hey, I'm at the park playing Frisbee.
[23:54.260 --> 23:55.920]  You know where I go every Thursday, right?
[23:56.120 --> 24:01.860]  And if you're out to rob me, which somebody suggested was a possibility, then you know I'm not at home.
[24:02.040 --> 24:03.500]  Or that I'm probably not at home.
[24:04.540 --> 24:10.060]  And finally, I want to do that analysis I was talking about where you look for overlapping clusters from different users.
[24:10.060 --> 24:13.840]  So you can try to make inferences about who knows who and go a bit further.
[24:14.700 --> 24:16.200]  And I'd like to go public with this.
[24:16.320 --> 24:26.460]  I mean, again, the whole point is to draw attention, to put this in people's faces, raise awareness, and at least have people be able to make an informed choice whether they want to choose to share geotags or tweets or not.
[24:27.200 --> 24:29.640]  And so I'm kind of trying to decide what I want to do with it right now.
[24:29.740 --> 24:38.100]  And I'd love to get your thoughts on this afterwards, whether just at the microphone or after the session over a beer, just what the best way to present it is.
[24:38.640 --> 24:42.960]  Originally, I was just going to throw all the data out there and let you look up every user and be like, whoa, this is scary.
[24:43.240 --> 24:45.940]  But then I got to thinking, I mean, this is all public data.
[24:46.120 --> 24:49.300]  None of it is like, this is all stuff people chose to share.
[24:49.740 --> 24:52.680]  I've been retrieving it from Twitter, following the terms of service.
[24:52.860 --> 25:02.260]  There's nothing unethical about the collection of data, but I'm worried that there's maybe something unethical about sharing all the conclusions with everybody, despite the fact that they could generate the same report themselves.
[25:02.760 --> 25:04.960]  So I'm kind of torn between what I want to do with that.
[25:07.100 --> 25:11.260]  So like I said, I was going to creep you out with what's possible to do with geotags.
[25:11.580 --> 25:16.380]  And if you'd like to see it live, again, it's running, and I can just come up, talk to me after.
[25:16.480 --> 25:19.300]  We'll go somewhere else where I don't have to worry about my computer so much.
[25:20.140 --> 25:21.120]  And we'll look at it.
[25:21.700 --> 25:26.000]  So, but I wanted to share some good news too, because people are choosing to geotag, right?
[25:26.180 --> 25:27.740]  Like, nobody's forcing them to.
[25:27.880 --> 25:28.600]  What are they getting out of it?
[25:28.640 --> 25:29.660]  What are we getting out of it?
[25:32.600 --> 25:33.640]  So here's an example.
[25:33.780 --> 25:34.640]  This isn't geotagging.
[25:34.700 --> 25:36.940]  This is geolocational data, though, that you're choosing to share.
[25:36.960 --> 25:40.260]  If you bring up Google Maps, it tells Google where you are and how fast you're going.
[25:40.520 --> 25:43.220]  And it aggregates that, and it can tell you how long it's going to take you to get to work.
[25:43.760 --> 25:44.700]  That's awesome, right?
[25:44.860 --> 25:45.800]  I mean, that's like Wikipedia.
[25:46.700 --> 25:50.600]  They're channeling people's idle time into something awesome and for the public benefit.
[25:51.120 --> 25:52.300]  And that's great, right?
[25:52.440 --> 25:53.880]  Like, who doesn't want to get to work on time?
[25:55.420 --> 25:59.800]  And another example that came up recently in Toronto, there was a G20 summit in Toronto.
[26:00.360 --> 26:04.520]  And I don't know how much press it got outside of Toronto, but inside Toronto, it was a huge deal.
[26:05.620 --> 26:08.360]  We spent a ton of money, over a billion dollars, on security.
[26:08.660 --> 26:11.640]  Biggest arrests, mass arrests in Canadian history, as far as I know.
[26:12.000 --> 26:13.000]  A thousand people arrested.
[26:14.580 --> 26:21.760]  New surveillance cameras installed, riots, cop cars on fire, tear gas, all that jazz that comes along with the G20 summit.
[26:22.440 --> 26:27.340]  And what was interesting was trying to follow it and learn what was going on, Twitter was one of the best ways to do it.
[26:27.540 --> 26:33.940]  And I'm going to give you two examples of how Twitter was a powerful tool with geotags during the G20.
[26:34.620 --> 26:35.740]  So the first of all...
[26:35.740 --> 26:36.200]  Okay, I lied.
[26:36.200 --> 26:36.680]  It's not Twitter.
[26:36.800 --> 26:37.180]  It's Flickr.
[26:38.640 --> 26:43.020]  But like I mentioned, they installed a whole bunch of new surveillance cameras in Toronto for the G20.
[26:43.020 --> 26:44.720]  And they promised us they'd be temporary.
[26:46.140 --> 26:46.600]  Yeah.
[26:47.400 --> 26:48.360]  Exactly, exactly.
[26:48.700 --> 26:56.440]  So a professor at the University of Toronto, Andrew Clement, was skeptical, as I'm sure everyone is, but decided to do something about it.
[26:56.500 --> 26:57.200]  So he grabbed...
[26:57.200 --> 27:05.100]  He bought a new phone that could geotag photos, got the list of where they were installing temporary cameras, and walked it and took pictures of them all.
[27:05.380 --> 27:07.660]  And then he posted them publicly on Flickr with geotags.
[27:08.080 --> 27:18.480]  So by doing this, it means that after the summit, anybody who was interested and also skeptical could just wander around, go to the dot on the map, look for whatever the picture was of, and say, hey, the camera's still there.
[27:18.580 --> 27:19.380]  The camera's not there.
[27:20.100 --> 27:24.840]  And that's a pretty powerful use of using the surveillance tool to turn surveillance back, right?
[27:25.000 --> 27:25.620]  Counter surveillance.
[27:26.040 --> 27:31.440]  And as far as I know, the last I'd heard, about half of the cameras were gone, which is surprisingly good news.
[27:31.600 --> 27:33.360]  And I'm not sure what's happening with the rest.
[27:34.780 --> 27:41.520]  So another example of where geotagging was really helpful during the G20 summit was there were two perspectives of the events.
[27:42.920 --> 27:45.080]  And watching them was quite interesting.
[27:45.280 --> 27:48.080]  There was the mainstream media which really loved this picture.
[27:48.600 --> 27:49.960]  Really loved pictures like this.
[27:50.120 --> 27:50.940]  I mean, that's an awesome...
[27:50.940 --> 27:53.940]  There were some with protesters or protesters jumping on the cars.
[27:54.860 --> 27:57.180]  Burning cars make really good TV, it turns out.
[27:58.440 --> 27:59.760]  But that's not the full story.
[28:00.960 --> 28:12.220]  And so by following the hashtag on Twitter, by looking at geotagged tweets on Twitter as they popped up, you saw pictures coming in that had different perspectives, you saw videos, you got commentary.
[28:14.720 --> 28:18.060]  So, for example, here is...
[28:18.060 --> 28:22.460]  This is one user's worth of tweets during the G20, the clustered.
[28:23.160 --> 28:25.200]  So you can see they kind of wandered...
[28:25.200 --> 28:27.720]  This is downtown Toronto where the G20 was, just west of it.
[28:28.600 --> 28:33.560]  And you can see that there's a yellow cluster, just means that there's more tweets there than the bluey-green ones.
[28:33.680 --> 28:40.960]  And so you can see that there's one kind of off to the side and you might wonder why did this person tweet so much from this spot, Queen of Spadina.
[28:41.220 --> 28:45.620]  So if you were to Google that or search for it on Twitter, the very first hit that you would get would be this.
[29:17.680 --> 29:20.640]  So that's a different story than the big picture's painting, right?
[29:21.120 --> 29:25.920]  I mean, the big picture makes it look like there's a riot going on and the cops are struggling to maintain order.
[29:26.500 --> 29:35.440]  Whereas the video that was user-generated and posted to Twitter makes it seem more like there were people just hanging out singing the national anthem who were getting assaulted by police.
[29:36.120 --> 29:39.220]  And that's very different stories in fact, right?
[29:39.360 --> 29:43.460]  And it's a perspective that for some reason the mainstream media wasn't really picking up on at first.
[29:44.140 --> 29:49.060]  But as time went on and people were tweeting about it and talking about it, we're forced to a little more.
[29:49.580 --> 29:58.560]  So it was an interesting case of not only getting both perspectives but the sort of ground-up Twitter perspective having an effect on mainstream coverage.
[29:59.300 --> 30:02.780]  And so one final good example I'm going to give you is exploration.
[30:03.240 --> 30:09.700]  I showed you a geotag photo on Flickr and how that was being used for counter-surveillance.
[30:10.160 --> 30:19.740]  But there's another use and this is somebody I interviewed said that when they go somewhere new that they've never been before they like to pull up a map of the place and then they like to look at the pictures that are there.
[30:20.240 --> 30:24.720]  And that having seen the pictures of the place makes them feel more comfortable and safer when they actually go there.
[30:25.520 --> 30:30.720]  So I think that's a pretty compelling good benefit assuming it's not a false sense of security of course.
[30:32.000 --> 30:33.880]  So maybe you're going to New York for the first time.
[30:34.020 --> 30:37.060]  You're going to a con and you're wondering what hotel should I stay at?
[30:37.660 --> 30:40.880]  And so you look around at the map and you're like I wonder what the hotel pen looks like.
[30:40.960 --> 30:41.700]  Does it have good views?
[30:41.920 --> 30:43.060]  So you can even pull that up.
[30:44.520 --> 30:45.940]  And you might rethink your decision.
[30:46.120 --> 30:47.320]  No, it does not have good views.
[30:50.540 --> 31:04.440]  So to put it all together then I just went over a couple benefits to geotagging and I gave you some scariness earlier about what geotagging at the large could result in for an individual person.
[31:04.820 --> 31:10.960]  So for me and the point of the project and I'm trying to make is that it's about informed choice.
[31:12.400 --> 31:14.920]  People have a need to connect, right?
[31:15.040 --> 31:16.640]  People want to participate in public.
[31:16.640 --> 31:19.240]  There's no reason we all have to be in this room together.
[31:19.340 --> 31:21.140]  We could just be watching this stream online.
[31:21.920 --> 31:23.200]  But people like to be known.
[31:23.360 --> 31:24.260]  People like to meet people.
[31:24.400 --> 31:25.120]  People like to connect.
[31:25.280 --> 31:26.300]  Even over the Internet.
[31:26.540 --> 31:27.080]  Even in reality.
[31:27.220 --> 31:28.140]  Some people use Yelp.
[31:28.480 --> 31:36.480]  I've heard a story about people sitting two different people sitting in a coffee shop who checked in on Foursquare and based on both checking in at the same location decided to chat.
[31:37.340 --> 31:39.000]  Like those people were in the same location.
[31:39.100 --> 31:40.020]  They could see each other.
[31:40.740 --> 31:45.540]  But it wasn't until they used technology to mediate and break the ice that they started chatting.
[31:45.760 --> 31:47.220]  And that, I mean seems a little strange.
[31:47.540 --> 31:48.580]  But that's good, right?
[31:48.780 --> 31:51.160]  I mean, ultimately we want to connect and it's good when people do.
[31:52.240 --> 31:55.440]  So it's kind of there's this balance or trade-off.
[31:55.660 --> 32:00.280]  And it's really hard to quantify the long-term risk of sharing too much information.
[32:00.560 --> 32:03.360]  It's really hard to get a good sense for it.
[32:03.360 --> 32:17.000]  And my hope by doing this by taking people's tweets clustering them and classifying them and showing them hey, guess what is that people at least get a better sense of balance when they're making that decision.
[32:19.580 --> 32:22.400]  But even informed choice isn't perfect.
[32:22.860 --> 32:32.600]  And this came up in an interview I did and this person is a very technical person is a social media person and this is what they said when I asked them about why they were geotagging tweets.
[32:36.780 --> 32:41.940]  Until you mentioned it just there I totally forgot that my tweets were being geotagged as well.
[32:44.600 --> 32:54.580]  So really like he went on to say that when he first got the phone and it's always when people first get the phones it asks them and they made the decision to share but totally forgot.
[32:55.060 --> 33:00.000]  So tweeting with not even considering that they might be giving away personal information.
[33:00.080 --> 33:07.820]  So tweeting from home I mean like that's a problem I think that people are giving away this information without being aware of it.
[33:07.980 --> 33:09.240]  And so what can we do?
[33:09.620 --> 33:11.780]  I mean I think improving the apps would be a step right?
[33:12.640 --> 33:15.120]  Most of the apps ask once and then never tell you again.
[33:15.600 --> 33:21.320]  So at least periodically showing you what's going on would make you think about it right?
[33:21.480 --> 33:25.900]  And would at least trigger that memory that oh yeah yeah I did decide to geotag my tweets.
[33:26.400 --> 33:28.560]  Because unless you look for it I guess you just don't know.
[33:30.380 --> 33:31.380]  So that's what I'd like to see.
[33:31.440 --> 33:38.320]  I'd like to see an improvement in the apps see more reminders for people and have people look at what they're doing and think about it.
[33:38.420 --> 33:41.120]  But I get the feeling that it's actually going to get worse before it gets better.
[33:44.440 --> 33:57.540]  I don't know about you guys but Facebook I think there are going to be big problems when Facebook turns on geotagging which it looks like is pending because I think this was back in May when they originally announced they wanted to do it.
[33:58.700 --> 34:00.000]  And there are a number of reasons.
[34:00.240 --> 34:07.240]  First of all most people's Facebook accounts are tied to their real identity so nobody even has to try to figure out who is this person in reality.
[34:07.440 --> 34:08.040]  It's just there.
[34:09.120 --> 34:10.420]  Facebook never deletes anything.
[34:10.700 --> 34:12.300]  They never delete anything.
[34:12.400 --> 34:14.320]  If you delete a photo on Facebook it's not deleted.
[34:14.900 --> 34:18.140]  They just check like there's a Boolean in there that says deleted.
[34:18.460 --> 34:20.660]  That's what happens when you delete a photo on Facebook.
[34:21.180 --> 34:24.380]  And they even keep track of your click stream, right?
[34:24.440 --> 34:25.820]  They know whose photos you're looking at.
[34:25.880 --> 34:27.560]  They know whose wall user you're looking at.
[34:27.760 --> 34:29.260]  They just don't delete anything.
[34:29.520 --> 34:32.320]  And so now they're going to have geodata on top of that.
[34:32.580 --> 34:44.800]  And they know your real life friends graph already but now they can infer people who you've maybe Maybe deliberately didn't tell Facebook you now, based on updating your status from the same location.
[34:45.580 --> 34:46.460]  And it goes worse than that.
[34:48.000 --> 34:50.600]  Raise your hand if you think Facebook is an ethical company.
[34:54.020 --> 34:55.060]  Right, right, exactly.
[34:55.260 --> 34:57.220]  So I'll give you an example of why I asked that.
[34:58.060 --> 35:00.660]  There's a Blackberry app for Facebook that I use.
[35:00.820 --> 35:08.480]  And you can optionally allow it to sync your contacts on your Blackberry with Facebook, so it will automatically download the photos of people and put them in or whatever.
[35:09.540 --> 35:14.040]  So you'd think it would connect to Facebook, get your list of Facebook friends, and sync that on the phone.
[35:14.600 --> 35:15.380]  You'd be wrong.
[35:15.700 --> 35:24.820]  What it actually does is it takes all of the contacts on your Blackberry, uploads them to Facebook, and then Facebook comes back and says, oh yeah, this person's on Facebook too, but we're going to remember the rest of them.
[35:25.540 --> 35:28.900]  Or maybe they say that they're not keeping it all, but they may be, right?
[35:28.980 --> 35:33.220]  So even if you deliberately didn't befriend somebody on Facebook, but you have their contact in your cell phone.
[35:33.540 --> 35:34.720]  Now Facebook knows that.
[35:35.540 --> 35:37.820]  And they're not upfront about that.
[35:37.980 --> 35:41.440]  And that's the problem with Facebook is that they're not upfront in general, right?
[35:41.580 --> 35:45.080]  I mean, every time they change the privacy settings, they don't ask for permission.
[35:45.360 --> 35:46.700]  They don't say, hey, guess what?
[35:46.760 --> 35:47.520]  This change is coming.
[35:48.320 --> 35:49.640]  Click here to turn it on.
[35:50.000 --> 35:52.460]  They're just like, hey, by the way, all statuses are now public.
[35:52.680 --> 35:53.640]  You can turn it off.
[35:53.760 --> 35:54.640]  You can figure the site out.
[35:55.180 --> 36:03.220]  And who would be surprised if that's what they do with the next iPhone, the next version of the iPhone app is just turn geotag on by default.
[36:03.520 --> 36:07.200]  And then you can't make an informed choice because most people aren't even aware that it's happening.
[36:08.500 --> 36:10.800]  So I'm not the only person to worry about these sort of issues.
[36:11.720 --> 36:15.940]  Some other ones that are interesting, pleaserobme.com.
[36:16.060 --> 36:17.540]  I think everybody's probably seen it.
[36:17.700 --> 36:23.840]  They kind of did the rounds a while back, just taking your foursquare check-ins and telling people where you're not, which is at home.
[36:25.420 --> 36:31.180]  Your open book is interesting because, again, like I said, Facebook made its statuses public without necessarily informing people well.
[36:31.440 --> 36:33.480]  Or at least people didn't get informed about it.
[36:33.660 --> 36:38.240]  And so people are updating their status about the status of their DUI conviction and stuff like that.
[36:38.340 --> 36:38.960]  It's terrifying.
[36:39.680 --> 36:41.360]  I can stalk you was presented yesterday.
[36:41.600 --> 36:42.100]  It's awesome.
[36:42.240 --> 36:44.900]  It's doing the same with just geotagged photos that you post.
[36:45.160 --> 36:49.640]  Because, again, iPhones tend to want to add location data to photos.
[36:50.240 --> 36:56.360]  And you can infer a lot of people... or you can do the same thing I've been doing, but just with their photos.
[36:58.540 --> 37:06.900]  And, yeah, so to summarize, then, I mean, I think the key point is that, like I mentioned earlier, one tweet is harmless.
[37:06.960 --> 37:08.540]  One geotag is mostly harmless.
[37:08.900 --> 37:13.860]  But once you start generating amounts of data, then people can start making inferences.
[37:14.260 --> 37:16.100]  And you need to be aware of that.
[37:16.160 --> 37:17.320]  And people need to be aware of that.
[37:18.480 --> 37:19.900]  But it's not all bad news.
[37:20.840 --> 37:23.540]  Society reaps a benefit from geotags.
[37:24.740 --> 37:26.780]  Both... sometimes the users get a benefit.
[37:26.960 --> 37:31.880]  I had a... I interviewed one guy who said... I asked him what value he'd ever gotten from Foursquare.
[37:32.360 --> 37:41.920]  And one of his stories was that he woke up one morning, hung over, and looked through where he checked in from last night and was like, hey, I didn't know I went to that bar.
[37:42.020 --> 37:42.540]  That's awesome.
[37:42.980 --> 37:43.940]  And so...
[37:44.900 --> 37:47.260]  So there's a personal archive value right there, right?
[37:47.400 --> 37:48.660]  It told him where he'd been.
[37:48.800 --> 37:50.000]  And that's kind of cool.
[37:50.320 --> 37:53.480]  Maybe... maybe not great, but cool.
[37:54.940 --> 37:55.480]  And, yeah.
[37:55.800 --> 38:02.200]  And so my final point, then, that I want to leave you with is two years ago at The Last HOPE, I went to a talk on WikiScanner by Virgil Griffith.
[38:02.540 --> 38:06.040]  And he was talking about how there's all this data on the net.
[38:06.360 --> 38:07.380]  He looked at Wikipedia.
[38:08.040 --> 38:09.440]  I've been looking at Twitter.
[38:10.580 --> 38:14.860]  And there are tons of other sources of data that can be aggregated and analysis run.
[38:15.180 --> 38:25.160]  And it's my hope that the people in this room, in two years, will be doing a presentation about some interesting data set that they've mined from the Internet and presenting some conclusions for me to see then.
[38:25.660 --> 38:26.740]  So, thanks.
[38:26.920 --> 38:27.460]  Any questions?
[38:29.540 --> 38:30.020]  Lights?
[38:38.490 --> 38:39.630]  I can't see anything.
[38:39.630 --> 38:43.090]  I guess one thing I would mention... I think there's a microphone.
[38:43.610 --> 38:44.270]  Ah, here we go.
[38:45.250 --> 38:46.390]  What am I doing on battery?
[38:46.750 --> 38:57.210]  One thing I would mention is that the latest version of the iPhone software does show an arrow in the upper right-hand corner whenever the GPS is feeding the data to an app.
[38:57.490 --> 38:57.810]  Okay.
[39:01.320 --> 39:10.600]  Hey, I don't know if you've studied this at all, but basically, you know how now a lot of people are sending out fake Google searches along with the real Google searches to scramble that.
[39:10.780 --> 39:18.740]  Have you seen anyone doing that with geolocation, saying, one day they're in Turkey, one day they're in New York, but clearly every day by their tweets they're at home just to scramble stuff like what you're doing?
[39:18.940 --> 39:20.760]  I haven't seen that come up in the data set.
[39:20.800 --> 39:29.640]  Actually, one of the slides that I took out, I was going to say, step one, make site that scares people about geotagging, step two, question mark, step three, profit.
[39:29.880 --> 39:34.060]  And then step two is going to be make an app that lets people lie about where they are and sell it.
[39:34.480 --> 39:40.520]  So I'm sure that we'll start to see people selling Twitter apps that let you just pick a location on a map as to where you're tweeting from.
[39:41.600 --> 39:45.480]  I'm sure help people generate alibis, if nothing else.
[39:47.600 --> 39:48.520]  Any other questions?
[39:52.800 --> 39:53.440]  Okey-dokey.
[39:53.740 --> 39:58.880]  Like I said, I'm happy to chat after or demo it after over beer.
[40:00.940 --> 40:02.920]  Just one thing, where are you going for beer?
[40:05.920 --> 40:06.480]  I'll tweet.
[40:07.100 --> 40:08.680]  No, I don't.
[40:08.760 --> 40:10.100]  Just come up after and we'll go somewhere.