[00:00.000 --> 00:00.540]  Wow.
[00:01.300 --> 00:03.620]  I guess it can only go uphill from there.
[00:05.440 --> 00:07.300]  So, greetings and thanks.
[00:07.660 --> 00:08.480]  Can you hear me?
[00:08.740 --> 00:09.300]  Is this okay?
[00:09.560 --> 00:10.080]  Cool.
[00:10.780 --> 00:14.420]  Thanks to the HOPE organizers for having me to do this talk.
[00:14.640 --> 00:22.840]  Also to Greg for his great talk prior to this one, which overlaps with mine in some nice ways.
[00:24.040 --> 00:31.120]  So, the title is Obfuscation and its Discontents, DIY Privacy from Card Swap to Browser Hack.
[00:34.480 --> 00:38.420]  I'm going to present obfuscation as an expressive technology.
[00:38.740 --> 00:40.760]  I'll say what that means in a second.
[00:42.400 --> 00:45.220]  Primarily, I'm going to do that through a series of examples.
[00:45.520 --> 00:50.600]  And these are also art-like projects, a bit similar to what Greg was doing.
[00:50.600 --> 00:57.840]  But first, I want to do a little historical detour as a way of approaching this idea.
[00:58.100 --> 01:00.840]  So, please bear with me here for a moment.
[01:01.440 --> 01:02.160]  Another moment.
[01:03.100 --> 01:20.380]  So, in a recent talk at Columbia, Eben Moglen traced the current trend toward a total surveillance state back to the very violent period of the Roman Empire in the first century BC, when control by Emperor Augustus was nearly absolute.
[01:20.980 --> 01:28.840]  There was some lip service paid to ideas like freedom, but there was little toleration of dissent.
[01:29.360 --> 01:34.040]  Because the Roman Empire was so vast, there was nowhere for dissenters to hide.
[01:34.320 --> 01:40.380]  It was, as Moglen cites, fatal to resist and impossible to fly.
[01:42.260 --> 01:51.200]  But he tells us that Augustus' vast power rested not primarily on military might, though that played a part, but rather on his control of communications.
[01:52.200 --> 01:55.400]  Specifically along the trading routes near and far from Rome.
[01:56.200 --> 02:03.340]  He says that in the administration of power, Augustus was the best informed human being in the history of the world.
[02:03.340 --> 02:08.020]  And that he used this power, in large part, to eradicate human freedom.
[02:10.100 --> 02:13.880]  Perhaps similarly to the situation in the U.S.
[02:13.920 --> 02:18.080]  after 9-11, when those in power in the U.S.
[02:18.200 --> 02:21.920]  abandoned what Moglen refers to as the morality of freedom.
[02:23.860 --> 02:29.800]  Augustus vastly increased the power of the Emperor in relation to the Senate, the Roman Senate.
[02:29.800 --> 02:40.360]  And this included, interestingly, appointing himself censor of Rome, which gave him the power, among others, to supervise public morality.
[02:42.720 --> 02:49.020]  And in this first century B.C., when Augustus took control, this was also the peak of Roman imperialism.
[02:49.020 --> 03:10.200]  By the end of his reign, his armies had conquered Spain and Portugal, what is today Spain and Portugal and Switzerland, Bavaria, Austria, Slovenia, Albania, Croatia, Hungary, Serbia, and even extended the borders of the Africa provinces to the east and south.
[03:11.540 --> 03:13.320]  That's not supposed to start yet.
[03:18.020 --> 03:26.900]  But in the years just before Augustus took control, there was a series of slave rebellions, called the Servile Wars.
[03:26.900 --> 03:46.680]  The third of which was somewhat long-lasting and documented in the well-known 1960 film called Spartacus, which was directed by none other than Stanley Kubrick, which I'll show a clip of in a moment.
[03:46.680 --> 04:04.360]  But the film, you sort of need to understand the political climate of the U.S. in the 40s and 50s, which was during the second Red Scare in Hollywood and the larger country.
[04:05.230 --> 04:16.680]  Okay, so this was the height of McCarthyism where artists and intellectuals thought to be associated with communism were hauled before Congress to inform on friends and colleagues.
[04:17.500 --> 04:32.940]  And in fact, the scriptwriter for the Spartacus film was Dalton Trumbo, one of the famous Hollywood Ten, a group of writers and directors who were called before the House Committee on Un-American Activities in 1947.
[04:33.740 --> 04:38.900]  But they all refused to cooperate and then were blacklisted.
[04:39.920 --> 04:47.020]  The scriptwriter Trumbo, who you see here, was then jailed, among others, for contempt of Congress.
[04:48.900 --> 05:06.320]  The author of the Spartacus novel in which the film was based, Howard Fast, was also called before the committee a bit later in 1950, when he refused to disclose the names of contributors to a fund for orphans of American veterans of the Spanish Civil War.
[05:06.600 --> 05:08.240]  And then he was also jailed.
[05:08.660 --> 05:21.620]  It was in jail that he began to work on the novel that would become Spartacus, which was then, of course, blacklisted by all the major publishing houses, and he was forced to self-publish it after his release.
[05:21.620 --> 05:37.920]  Nonetheless, the film was eventually made, and after the release, amid some controversy, producer and star Kirk Douglas announced publicly that the blacklisted Trumbo was the screenwriter.
[05:39.480 --> 05:52.340]  And then later, President Kennedy crossed the picket lines to see the film, which sort of signaled the gradual end of blacklisting in Hollywood.
[05:53.580 --> 05:57.520]  Probably due mostly to the massive profits the film generated.
[05:59.240 --> 06:14.720]  I think it was the largest grossing film ever at that time, and continued to hold that position for ten more years, at least, till it was displaced by airport, I believe.
[06:21.700 --> 06:40.580]  So we go from imperialism and totalitarian rule, enabled in some large part by signal intelligence, to slavery, McCarthyism, and the morality, or lack thereof, associated with freedom in the U.S.
[06:43.330 --> 06:52.580]  So, we now can come to this clip from Spartacus, which I think provides a really nice illustration of obfuscation, which is the topic today.
[06:53.480 --> 06:55.420]  So, let's see if I can get this to play.
[07:09.040 --> 07:10.280]  Have we a count of prisoners?
[07:11.040 --> 07:12.820]  We haven't made the final count, sir.
[07:12.820 --> 07:22.100]  I bring a message from your master, Marcus Licinius Crossus, commander of Italy.
[07:22.680 --> 07:29.260]  By command of his most merciful excellency, your lives are to be spared.
[07:32.670 --> 07:36.430]  Slaves you were, and slaves you remain.
[07:36.430 --> 07:51.450]  But the terrible penalty of crucifixion has been set aside on the single condition that you identify the body or the living person of the slave called Spartacus.
[08:04.150 --> 08:05.820]  I'm Spartacus!
[08:05.960 --> 08:06.980]  I'm Spartacus!
[08:08.440 --> 08:10.040]  I'm Spartacus!
[08:10.380 --> 08:11.980]  I'm Spartacus!
[08:12.100 --> 08:13.300]  I'm Spartacus!
[08:13.480 --> 08:14.580]  I'm Spartacus!
[08:14.600 --> 08:16.200]  I'm Spartacus!
[08:44.540 --> 08:51.040]  Oh, very subtle soundtrack to not manipulate your emotions at all.
[08:51.640 --> 08:51.740]  Okay.
[08:53.280 --> 08:57.660]  So here's the definition of obfuscation that I'm going to use.
[08:57.660 --> 09:09.260]  The production, inclusion, addition, or communication of misleading, ambiguous, or false data in an effort to evade, distract, or confuse data gatherers.
[09:09.780 --> 09:10.500]  Okay.
[09:10.820 --> 09:15.460]  That's from Brunton and Nissenbaum in a 2011 paper.
[09:17.440 --> 09:35.340]  And now that we have that little historical background in place, we can look at some specific examples, many of which span an interesting line between technological solutions, political commentary, and art projects, an idea that I'll come back to.
[09:36.020 --> 09:37.400]  So here's the first one.
[09:37.460 --> 09:39.020]  I don't know if people know this work.
[09:39.020 --> 09:44.880]  It's called I Like What I See by Steve Kleiss in 2012.
[09:45.420 --> 09:54.660]  It's a recent, well, 2012 browser hack that uses obfuscation to confuse the collection of personal data on Facebook.
[09:55.820 --> 10:01.480]  So I Like What I See is a Chrome extension to automatically click all like links on Facebook.
[10:01.480 --> 10:08.840]  When you visit Facebook, click the thumbs up in the extension bar and start scrolling and liking, liking and scrolling.
[10:09.280 --> 10:11.980]  Every instance of the word like will be clicked.
[10:12.220 --> 10:13.140]  Don't worry.
[10:13.140 --> 10:16.660]  Facebook is a fun place full of all the stuff you like.
[10:17.760 --> 10:24.700]  And that's actually the markdown readme file from GitHub where I grabbed that.
[10:26.380 --> 10:42.760]  And while the language is tongue-in-cheek here, the problem explored is, of course, familiar to everyone and serious, specifically the aggregation of personal data by third-party service providers, in this case Facebook, into databases which can then be traded,
[10:43.020 --> 10:59.540]  sold, potentially stolen, used for all kinds of nefarious behavior, violating privacy in many cases and even assisting those who'd like to create a state of total surveillance similar to that which Augustus enjoyed in Rome.
[11:00.360 --> 11:10.640]  Snowden, of course, revealed for us how the NSA uses tracking cookies issued by service providers like Google to assist in their surveillance activities.
[11:13.080 --> 11:25.880]  And one of the things this piece does is highlight a tension between specific human values, between our desire to share with each other on the one hand and to maintain privacy on the other.
[11:26.860 --> 11:45.020]  Of course, people like Zuckerberg or Eric Schmidt want us to think that these two values are mutually exclusive, but we can quickly realize that there are no technical or logical reasons why this should be so, besides the massive profits generated by the companies telling us this.
[11:48.040 --> 12:11.100]  So this is a simple case, of course, but the larger argument and one of the assumptions of this talk is that technical artifact software can represent or embed specific political questions and agendas, and they are ways of... these could be ways of arguing about who gets a voice in determining future directions for society.
[12:12.000 --> 12:14.460]  Let's skip this slide there.
[12:15.280 --> 12:25.100]  So in addition to talking about how I like what I see pollutes Facebook's logs, we can also consider the arguments it makes as a cultural artifact.
[12:28.580 --> 12:36.740]  Again, referencing Greg's talk earlier, as we might with a piece of writing or an art installation that we view.
[12:37.040 --> 12:47.360]  Which is to say that some of these works, and I'll show several of them next, have both functional and expressive properties that we can consider.
[12:49.040 --> 12:55.860]  So before we do that, let's go back a few years to the project on which I like what I see was modeled.
[12:56.160 --> 13:01.500]  And this work was called Track Me Not, and was about privacy in the context of web search.
[13:01.880 --> 13:13.680]  It was released in 2006, soon after AOL released its anonymized search records, which were then quickly de-anonymized by researchers, resulting in a small scandal.
[13:14.420 --> 13:30.620]  AOL eventually apologized, but what this told the larger population, was that searches can reveal a great deal of personal and private information about people, most of whom weren't even aware that such data was being collected and stored.
[13:30.760 --> 13:38.900]  I'm not sure how many people remember this back in 2006, but it was a small, big deal.
[13:40.740 --> 13:54.840]  So what Track Me Not does is to notice where and when you search for things, say at Google or Bing or Baidu, and then to start sending fake, but realistic seeming queries to their servers at intervals around the original.
[13:55.460 --> 14:08.540]  Hiding your real interest, protecting yourself and potentially others, and polluting the engine's search logs.
[14:08.740 --> 14:23.580]  So after a simple prototype was released, there were several iterations, adding new features like dynamic query lists from live data sources, strategies to defeat timing and side-channel attacks.
[14:23.580 --> 14:40.340]  Most recently, there was a Department of Homeland Security feature added that you could opt into to include trigger words from this list in all your fake searches that Track Me Not is issuing.
[14:43.670 --> 14:48.550]  So beyond working is a tool to hide user interests and defeat targeted advertising.
[14:48.550 --> 14:51.910]  Track Me Not also has these expressive elements.
[14:51.910 --> 15:05.290]  The once rather disempowered user is provided with some small bit of a voice, a means of registering their dissatisfaction with those who run these massive and opaque technical systems.
[15:05.510 --> 15:17.510]  And they do that by literally scrawling their dissatisfaction over the search engine logs, which are arguably the most valuable asset that search engines keep.
[15:21.010 --> 15:38.110]  So besides generating lots of controversy, some of which I'll talk in a second, it was the start of two things in my own work, one being a type of project I like to call artware, which is a mix of software tool and art project, and the other being the use of this strategy we call obfuscation,
[15:38.130 --> 15:40.450]  which of course is not a new idea.
[15:40.450 --> 15:53.650]  This is a non-digital example of obfuscation dating back to World War II, when metalized material was dropped from planes in order to flood radar screens with false positives.
[15:57.730 --> 16:14.630]  Another non-digital example of obfuscation, again for privacy, this one was mentioned in the title, is loyalty card swapping, which was popular for a while as a way to confuse retail purchase databases.
[16:15.310 --> 16:35.630]  There was even a site called cardexchange.com that facilitated swaps between strangers, and one proponent sent out his UPC from his card on a sticker, this is the right side image above there, to any and all takers for their use.
[16:35.810 --> 16:45.190]  So if you sent him an email, he would send you back the sticker, and he was hoping to amass the greatest single shopping profile of all time.
[16:49.040 --> 17:07.840]  But even in these simple examples, we can find potential for harm, like the case above where a father found out from Target that his daughter was pregnant, or the court case cited by the Electronic Privacy Information Center, involving a customer who slipped and was injured in a supermarket,
[17:08.100 --> 17:19.920]  and then the market tried, in the court case, to use purchase data, to use his purchase data, specifically his alcohol purchases, to suggest he was likely impaired at the time of the fall.
[17:24.700 --> 17:33.540]  So here's another little software, obfuscating software example.
[17:34.180 --> 17:42.020]  Cash Cloak applies the concept of obfuscation to GPS data that's collected by mobile phone providers.
[17:42.020 --> 17:57.960]  So instead of sending only requests for data on the user's actual path, it sends requests for a number of different but plausible paths related to the current location, then presents the user only with the results they need for their actual path.
[17:59.140 --> 18:03.740]  And so that was a relatively recent piece, I'm not sure of the year.
[18:05.440 --> 18:12.800]  The Pirate Bay also uses obfuscation, or has at some point, to deter anti-privacy organizations.
[18:13.120 --> 18:25.340]  So what these organizations would do is to generate requests to trackers for peers, for a list of peers, and then log the IP addresses, and then send takedown notices.
[18:25.340 --> 18:42.360]  So versions of the Pirate Bay's tracker software were developed that automatically inserted a number of random peer IP addresses into the list, then thus providing all the users of the network with a sort of plausible deniability.
[18:43.260 --> 18:47.400]  You know, my IP is only in there because it was one of the fake ones.
[18:52.300 --> 18:56.260]  Another one called ScareMail.
[18:57.060 --> 19:01.260]  And this adds to all your emails when you enable it.
[19:01.620 --> 19:08.020]  Algorithmically generated narratives built primarily from the trigger words that are tracked by the NSA.
[19:08.320 --> 19:10.700]  And this puts this at the bottom of your actual email.
[19:12.520 --> 19:17.760]  Unfortunately, I don't have any examples of those narratives, but they're sometimes quite funny.
[19:19.900 --> 19:25.580]  That was a piece by Ben Grosser, I think, this year or last year.
[19:28.740 --> 19:31.220]  Okay, so now on to the discontents.
[19:32.020 --> 19:38.520]  And then I'll finish up with one very new project, which we're about to release hopefully later this month.
[19:39.440 --> 19:50.880]  So, as I mentioned, TrackMeNot initially generated quite a bit of controversy while the software was downloaded over one million times in the first year.
[19:51.160 --> 19:56.360]  Many critics, especially in the security community, were rather hostile to the approach.
[19:56.660 --> 20:01.340]  And these criticisms generally broke down into two camps.
[20:02.880 --> 20:10.920]  Functional critiques, arguing that it doesn't work to protect you or to dilute your profile.
[20:10.920 --> 20:15.340]  And ethical critiques, which argued that it was in some way immoral.
[20:18.670 --> 20:22.670]  So, I'll talk about the TMN case here, TrackMeNot case here.
[20:22.790 --> 20:28.070]  But similar arguments have been put forth against most of the other projects that use obfuscation.
[20:28.730 --> 20:30.170]  TrackMeNot was just the first.
[20:30.170 --> 20:42.910]  But these range from legal attacks to the types of side channel attacks that are listed above there.
[20:43.390 --> 20:51.290]  Usually being based on three different properties of the software.
[20:51.450 --> 20:55.170]  One being the types of queries it generates.
[20:55.170 --> 20:58.690]  The other being the timing of those queries that are generated.
[20:59.250 --> 21:01.250]  And then the third being side channel attacks.
[21:01.450 --> 21:03.050]  Information leaked by the browser.
[21:03.290 --> 21:04.170]  Incorrect headers.
[21:05.270 --> 21:12.110]  The fact that early versions didn't use the text suggestion bar to type in the queries.
[21:13.010 --> 21:14.990]  Click throughs on search results.
[21:15.150 --> 21:15.770]  Things like that.
[21:15.930 --> 21:16.770]  Active content.
[21:18.230 --> 21:23.170]  If a JavaScript file, for example, wasn't downloaded by the plugin.
[21:23.170 --> 21:28.290]  That would be a hint to the search engine that it wasn't an actual query.
[21:30.650 --> 21:41.070]  The larger question, of course, is whether such noise, when injected into tracking databases of various kinds, actually matters for the kind of data mining being done.
[21:41.070 --> 21:50.370]  And, of course, this is difficult to say without actually knowing what techniques are being used by search engines.
[21:50.650 --> 21:55.810]  And, of course, this info is carefully guarded and not released.
[21:57.130 --> 22:03.890]  But we do know that noise consumes significant resources for data miners.
[22:03.890 --> 22:13.590]  There's a recent quote stating that 80% of the time spent on data mining, sometimes more, is spent on cleaning, preparing the data, removing the noise, etc.
[22:15.870 --> 22:20.990]  It's a similar bit of research.
[22:22.590 --> 22:28.510]  Percent of project time spent on removing noise and data cleaning.
[22:31.170 --> 22:45.530]  So, even if some or perhaps all of the noise by different implementations can eventually be filtered out by search engines, one thing we would need to consider is the time and resources that they have to invest in order to do that.
[22:51.720 --> 23:04.320]  And so, if anyone wants to discuss the technical details of any of those implementations, happy to do so in the Q&A if I can get through these last two sections.
[23:07.380 --> 23:13.140]  But also, many of the articles were, sorry, many of the critiques were ethical ones.
[23:13.600 --> 23:18.040]  And they, I've broken down some of them here.
[23:23.160 --> 23:33.560]  Dishonesty, a sort of Kantian argument that it's a form of lying, which is misrepresenting yourself, which is always wrong.
[23:35.680 --> 23:56.480]  Free riding, that you're breaking the free web, or that others will have to pay the price for your avoidance of these, say, ads or marketing opportunities, which support so-called free content.
[23:57.540 --> 24:04.860]  And finally, waste, pollution, and system damage, which was very, many critiques focused on this.
[24:05.020 --> 24:10.100]  You're always, the program is wasting bandwidth, it's wasting storage.
[24:10.760 --> 24:21.020]  And then the sort of, accumulative critique that says, if everyone did this, the universe, the Internet wouldn't work anymore.
[24:22.040 --> 24:24.840]  Maybe that's what happened in this room earlier.
[24:28.320 --> 24:44.220]  And so, these were addressed in some depth, were addressed somewhat in the original paper we wrote about TrackMeNot, and addressed much more deeply in a political philosophy of obfuscation.
[24:44.220 --> 24:49.140]  And this was a 2011 paper by Brunton and Nissenbaum, again.
[24:50.360 --> 24:56.600]  Okay, so lastly, I'll talk about, I want to leave a little time for Q&A.
[24:56.800 --> 25:05.140]  I'll talk about a very recent project, which hasn't even been publicly released yet, and this is called Ad Nauseam.
[25:06.660 --> 25:12.040]  Also with Helen Nissenbaum, and the design you see here by Mushan Zeraviv.
[25:14.380 --> 25:20.620]  And here we're applying the concepts of obfuscation to the ad networks that track us over the web.
[25:22.720 --> 25:28.900]  So how we do this is by first blocking all the ads, similar to the way in which an ad blocker works.
[25:28.900 --> 25:33.420]  And then, in the background, clicking on all the ads.
[25:33.700 --> 25:40.660]  So every single ad on every page that you visit will be clicked in the background by the software.
[25:40.960 --> 25:44.060]  It will be visited in the background by the software.
[25:46.810 --> 25:55.990]  And so, it also presents you with a sort of collage of the ads you would have seen had you been seeing ads.
[26:00.650 --> 26:14.370]  And sort of presents, which I'll come back to, a picture of what your profile in the ad networks looks like, which most people don't have, or I at least, don't have a good idea of.
[26:15.010 --> 26:15.230]  Okay.
[26:16.350 --> 26:19.390]  So what are the objectives of this piece?
[26:20.270 --> 26:22.350]  Three things, or three and a half things.
[26:22.350 --> 26:30.410]  Protection, expression, subversion, and surveillance, as Greg mentioned earlier.
[26:31.670 --> 26:34.650]  So first, the protections are kind of obvious.
[26:35.590 --> 26:41.610]  Your true interests are hidden because you are clicking on everything.
[26:43.070 --> 26:51.590]  And your data profile is noise, is diluted with noise.
[26:56.540 --> 26:58.000]  So subversion.
[26:58.280 --> 27:03.440]  So this is talking, at least, partially about the system level, right?
[27:03.520 --> 27:10.280]  We're introducing distrust between advertisers and ad networks, right?
[27:10.280 --> 27:15.940]  Advertisers are paying ad networks to display ads.
[27:16.480 --> 27:20.380]  And yet, no humans, in this case, are seeing the ads.
[27:20.860 --> 27:25.340]  And so there's now tension between these two parties, which potentially...
[27:26.200 --> 27:29.520]  Or there's potentially tension, which there was not before.
[27:32.640 --> 27:35.680]  And lastly, expression, which I mentioned before.
[27:35.880 --> 27:56.840]  It amplifies the voices of those with little power in this system, who are able to, again, write their dissatisfaction across the logs and databases of the system that they're basically forced to use by navigating the web.
[27:58.840 --> 28:02.900]  And then, suveillance is this...
[28:02.900 --> 28:06.060]  That's actually the wrong slide, but...
[28:07.320 --> 28:11.900]  It's more like this slide, is this idea of watching the watchers.
[28:11.900 --> 28:18.200]  So here we get this sort of collage-like view of what the watcher...
[28:20.940 --> 28:22.800]  The profile that they have...
[28:22.800 --> 28:24.920]  That they imagine is ours.
[28:25.960 --> 28:26.040]  Okay.
[28:27.620 --> 28:28.820]  It's a sort of...
[28:28.820 --> 28:34.140]  I'm not quite sure whether that's a real case of suveillance, but it's interesting nonetheless.
[28:38.760 --> 28:39.440]  Okay.
[28:39.840 --> 28:44.300]  So obfuscation as a strategy, does this make sense?
[28:44.300 --> 28:47.380]  So here are some positive things to think about.
[28:47.540 --> 28:52.740]  It's a quick and dirty solution to potentially thorny problems.
[28:53.100 --> 29:04.860]  Many of these browser add-on hacks were quite quickly developed in a few hours or a couple of days by a single coder, and they can comment quickly on current events.
[29:06.180 --> 29:14.140]  They're often quite understandable to non-technical people, which I think is important.
[29:14.300 --> 29:17.260]  They tell an easy story that people understand.
[29:17.560 --> 29:28.940]  So in comparison to something like Tor, which is, of course, great for its purposes, non-technical users get these ideas immediately.
[29:28.940 --> 29:38.140]  They understand the basic workings because they're simple, and they can then be adopted much more widely.
[29:41.400 --> 29:57.420]  Power asymmetries, you know, these provide a voice for people without much power in these dynamics with large companies that they didn't have before.
[29:58.520 --> 30:16.380]  And, of course, there's the community element of it in the sense that you're often polluting a database, larger sort of metrics that can be derived from a database across users.
[30:26.560 --> 30:34.220]  And maybe obfuscation is only a technique or a strategy we'd want to use when others are not feasible.
[30:34.800 --> 30:39.140]  So some of the others that are suggested are listed here at the bottom.
[30:39.740 --> 30:41.020]  Opting out, of course.
[30:41.020 --> 30:59.400]  We may have the option to opt out of some of these systems, but this is becoming less and less of an option as governments and sort of necessary services start to work in this way.
[31:00.920 --> 31:09.000]  Corporate sort of self-policing is probably hopeless as there's little incentive for them to do better.
[31:09.000 --> 31:17.860]  Their core business is collecting and selling personal data that they... which they're gathering without any real informed consent.
[31:18.140 --> 31:20.580]  But it's making them a ton of money.
[31:22.220 --> 31:22.780]  Regulation.
[31:23.040 --> 31:27.980]  Government regulation is a potentially longer-term solution.
[31:28.200 --> 31:42.260]  But, again, depending on the countries, governments are closely linked to business and also have the added agenda of benefiting from surveillance possibilities.
[31:43.920 --> 31:55.700]  Technically, there are certainly longer-term solutions here as well, but they often involve things like standards and a range of different stakeholders, often with conflicting goals.
[31:55.700 --> 32:03.180]  The utter failure that was the do-not-track initiative being a good example of that.
[32:06.890 --> 32:12.470]  So, lastly, so open questions for thinking about obfuscation.
[32:12.870 --> 32:25.990]  Is it a viable, if short-term, technical strategy that we can use ethically along with other strategies like encryption, distributed social systems, opting out, et cetera?
[32:26.680 --> 32:31.950]  What kind of guarantees can obfuscation systems actually provide in terms of privacy?
[32:32.410 --> 32:39.630]  And what types of adversarial systems have been developed to thwart attempts at obfuscation?
[32:39.630 --> 32:48.890]  And we're working on some research to answer the second and third questions here.
[32:49.110 --> 33:01.270]  So, I'm going to stop there, and hopefully we leave 10 or 15 minutes for Q&A if people have things they'd like to ask.
[33:01.550 --> 33:03.350]  So, thanks a lot.
[33:18.040 --> 33:20.940]  Do you want to talk louder?
[33:21.240 --> 33:22.960]  I can repeat the question too.
[33:23.300 --> 33:32.340]  Well, I'm curious about how your obfuscator works on the text below.
[33:32.520 --> 33:42.760]  So, do you make all your requests that it's used on the page using a persistent connection or is it just a one-shoot and just trashes the data?
[33:43.040 --> 33:44.840]  Or what's it do with the information or requests?
[33:45.240 --> 33:46.360]  What's the request flow?
[33:46.940 --> 33:47.140]  Okay.
[33:47.360 --> 33:56.140]  So, the question was, in the case of ad nauseum, how does the system make requests to these advertising,
[34:00.100 --> 34:03.280]  to the ads, and then what do they do with the data that they get back?
[34:03.480 --> 34:03.900]  Is that...?
[34:03.900 --> 34:04.140]  Yeah.
[34:04.140 --> 34:04.780]  Okay.
[34:05.180 --> 34:12.160]  So, basically ad nauseum was built on top of ad block plus, right?
[34:12.320 --> 34:19.960]  So, that enabled us to identify all the ads on a page and then remove them, right?
[34:20.140 --> 34:20.660]  Hide them.
[34:21.280 --> 34:27.060]  Either by not requesting them or by requesting them, downloading them, and then making them invisible.
[34:27.800 --> 34:28.440]  Okay.
[34:28.600 --> 34:33.080]  So, now we have the URLs for all of these ads.
[34:35.580 --> 34:42.140]  And we load and we simulate a click on each of these in a hidden browser window.
[34:42.780 --> 34:43.380]  Okay.
[34:43.660 --> 35:01.440]  The hidden browser window then downloads the content that you would have seen if you had clicked on it in your browser and including resources like JavaScript and Flash and whatever else, images.
[35:02.180 --> 35:06.300]  And then, once it's downloaded, that stuff is all thrown away.
[35:10.260 --> 35:15.020]  Is there an understanding or anything for it given to the effect of a paper click model about it?
[35:15.660 --> 35:16.120]  Yeah.
[35:16.340 --> 35:17.360]  There is...
[35:18.080 --> 35:19.600]  I should have put that in the talk.
[35:19.720 --> 35:28.700]  There was some critiques early on in this project that this was a form of click fraud.
[35:29.000 --> 35:29.460]  Yes.
[35:31.080 --> 35:33.540]  That was what you're getting at.
[35:35.300 --> 35:45.460]  In our discussions with lawyers, it seems that, and I'm not very sure of the details, but it seems that click fraud implies some...
[35:48.640 --> 35:49.200]  Targeting.
[35:49.200 --> 35:49.740]  What did you say?
[35:49.800 --> 35:50.220]  Targeting.
[35:50.320 --> 35:51.120]  Click fraud.
[35:51.340 --> 35:51.740]  Yeah.
[35:52.040 --> 35:56.700]  It avoids targeting and also involves financial gain.
[35:57.440 --> 35:58.000]  Right?
[35:58.120 --> 35:58.840]  Fraud involves...
[36:01.900 --> 36:04.440]  I think that was the line of the argument.
[36:05.360 --> 36:12.580]  The result was that this wouldn't fall legally under the category of click fraud.
[36:12.740 --> 36:13.880]  But that doesn't mean...
[36:14.220 --> 36:17.800]  That certainly won't stop people from making that argument.
[36:18.800 --> 36:19.260]  All right.
[36:19.480 --> 36:19.740]  Thank you.
[36:22.660 --> 36:23.220]  So...
[36:23.220 --> 36:23.480]  Sorry.
[36:24.180 --> 36:24.880]  Go ahead.
[36:25.300 --> 36:37.020]  I was just wondering, and you just mentioned that, what made you initially decide instead of avoiding ads or avoiding...
[36:37.020 --> 36:50.890]  Like I was going to ask you about, now there's several search engines that supposedly act as like a proxy between you and Google and Bing, like privately or Startpage or DuckDuckGo.
[36:50.890 --> 36:51.730]  Right.
[36:52.110 --> 37:05.270]  What do you think, and where do you, about their efficacy, one and two, like which side of the coin do you think is more useful or helpful to the end user?
[37:05.470 --> 37:09.430]  Avoiding or just flooding as with ads?
[37:09.430 --> 37:16.650]  Like avoiding like AdBlock Plus or AdBlock Plus can also block all those Facebook like buttons.
[37:17.050 --> 37:22.910]  Or let me just click on everything and try to corrupt their database as much as possible.
[37:22.930 --> 37:23.190]  Right.
[37:23.430 --> 37:27.730]  I mean there's certainly two different strategies, both of which...
[37:27.730 --> 37:29.370]  Well, both of which I use.
[37:29.550 --> 37:30.730]  You can do both.
[37:30.730 --> 37:41.070]  So, the problem with these sites like DuckDuckGo and such is you're basically placing a sort of blind trust in another third-party.
[37:41.350 --> 37:51.970]  So, while I feel better about DuckDuckGo than I do about Google, I don't really have any guarantees that I should, you know, that that's accurate.
[37:53.610 --> 37:57.030]  But you can't actually do both as far as the ads are.
[37:57.150 --> 37:59.250]  You're either avoiding them or you're clicking on all of them.
[37:59.430 --> 38:05.770]  So, what was your desire to sort of not avoid them but just sort of click on all of them?
[38:05.890 --> 38:08.370]  Like what was that initial concept?
[38:08.830 --> 38:12.590]  I'm getting confused between the search engine case and the ad case.
[38:12.950 --> 38:14.210]  I was asking about both.
[38:14.390 --> 38:16.330]  But just you said I use both.
[38:16.890 --> 38:25.590]  I mean I use DuckDuckGo sometimes but, I mean in my experience, the search results you get from DuckDuckGo are often...
[38:25.590 --> 38:26.270]  Subpar.
[38:26.470 --> 38:27.070]  Subpar, yeah.
[38:28.690 --> 38:30.670]  Maybe that will change over time, hopefully.
[38:31.970 --> 38:36.330]  But, and then when I do search for Google, I'll use some kind of obfuscating...
[38:37.990 --> 38:44.630]  You know, there are also other ways of keeping your data out of a profile.
[38:44.630 --> 38:46.910]  So, you can make dummy accounts and et cetera.
[38:47.110 --> 38:52.790]  But with browser fingerprinting, many of those are not going to work very well at all.
[38:52.990 --> 38:55.630]  Do you use the user agent switching also?
[38:56.170 --> 38:56.310]  Yeah.
[38:58.530 --> 38:59.910]  Wait, let me... Sorry.
[39:00.730 --> 39:04.670]  I'm using public computers and you have to get an answer for that.
[39:04.830 --> 39:08.590]  But also the autofill option with fake data in the net.
[39:09.790 --> 39:10.230]  Autofill...
[39:10.230 --> 39:13.530]  So, I want to populate it, you know, a password to fill all that stuff.
[39:13.710 --> 39:17.910]  Turn that feature on, then you can probably turn it off.
[39:23.090 --> 39:28.210]  It's like an idea for a new, a new plugin that would do this.
[39:29.530 --> 39:31.130]  Submit forms on that, you think?
[39:31.430 --> 39:32.350]  I guess that'd be cool.
[39:32.670 --> 39:35.310]  I think such things have been at least attempted.
[39:37.390 --> 39:44.030]  Is there anything for use on public computers where you can't really install software such as your audience?
[39:45.430 --> 39:50.650]  Well, sometimes you can install software on public computers, things like Tails or whatever.
[39:50.830 --> 39:57.450]  But if you're on a public computer and you're not signed in, then it's not your profile anyway.
[39:57.970 --> 40:02.510]  So, it's not... Let me just check.
[40:02.950 --> 40:04.770]  I can't really see back there.
[40:06.410 --> 40:07.430]  Yeah, I have a question.
[40:07.590 --> 40:08.870]  Sure, I'll come back to you.
[40:10.410 --> 40:15.910]  You mentioned, you know, one of the top items was efficacy, right?
[40:16.010 --> 40:17.510]  The efficacy of obfuscation.
[40:17.690 --> 40:17.890]  Right.
[40:18.130 --> 40:23.670]  Do you know of any experimental evidence to describe the various...
[40:23.670 --> 40:24.130]  Yeah.
[40:24.650 --> 40:26.550]  So, as I said... Sorry, go ahead.
[40:27.530 --> 40:34.970]  So, as I said, I was... We're working on some of that now, but there's an excellent paper that exists.
[40:35.350 --> 40:37.110]  Let me see if I can find it here.
[40:37.570 --> 40:53.230]  In which six of these query obfuscators are evaluated against model adversaries with machine learning classifiers, trying to distinguish between real and fake, and they have varying degrees of success with each of them.
[40:53.650 --> 40:55.850]  Let me just see if I can find the reference here.
[40:55.850 --> 40:56.050]  Yeah.
[40:56.210 --> 40:58.930]  So, it's Balsa, Troncoso, and Diaz.
[40:59.030 --> 41:02.490]  And the paper is called Obfuscation-Based Private Web Search.
[41:02.950 --> 41:04.870]  And they look at six different...
[41:04.870 --> 41:06.050]  Six different...
[41:06.050 --> 41:11.250]  Track Me Not and five different subsequent implementations by other people.
[41:11.530 --> 41:11.990]  Great.
[41:12.310 --> 41:12.750]  Thank you.
[41:13.270 --> 41:14.530]  Could you say that again?
[41:14.770 --> 41:17.110]  The name of the paper?
[41:17.110 --> 41:19.810]  The paper is Obfuscation-Based Private Web Search.
[41:20.730 --> 41:25.030]  And the authors are Balsa, Troncoso, and Diaz.
[41:30.700 --> 41:31.340]  Hi.
[41:32.100 --> 41:40.440]  When you're using Ad Nauseam, I understand how it wouldn't be considered a legal issue if anyone taking advantage of the click-through generated.
[41:42.140 --> 41:46.000]  But doesn't the ad networks, aren't they still going to benefit from this?
[41:46.200 --> 41:48.920]  Aren't they even saying, gee, wow, our ads are working great.
[41:49.340 --> 41:50.760]  It's like crazy.
[41:50.920 --> 41:55.920]  And then, so the ad networks will say that...
[41:57.060 --> 41:57.580]  Yeah.
[41:58.640 --> 42:00.000]  You said the advertisers?
[42:00.280 --> 42:02.160]  The ad networks slash...
[42:02.160 --> 42:03.300]  The advertisers are paying.
[42:03.800 --> 42:04.000]  Right.
[42:04.000 --> 42:06.980]  The ad networks are going to be making quite a bit of money on...
[42:06.980 --> 42:07.100]  Right.
[42:07.260 --> 42:09.340]  So the ad networks are initially quite happy.
[42:09.460 --> 42:11.660]  Oh my gosh, we're getting all these new clicks.
[42:11.880 --> 42:19.240]  The advertisers who have to pay for all these new clicks, which no one has actually seen, are less happy.
[42:19.240 --> 42:31.200]  And that's what I was saying about introducing this, sorry, tensions between various parties in this system with the potential for subverting it in some way.
[42:31.800 --> 42:32.120]  Right?
[42:32.220 --> 42:37.060]  Because those two entities are very much on the same side at this point.
[42:37.120 --> 42:38.000]  So what do you think will happen?
[42:39.200 --> 42:40.480]  We'll see, I guess.
[42:40.840 --> 42:42.340]  We don't know if it's going to subvert it.
[42:42.540 --> 42:43.260]  Just something will happen.
[42:43.440 --> 42:44.640]  Yeah, something will happen.
[42:45.400 --> 42:52.160]  I mean, that's sort of the interesting thing about some of these more artistic kind of...
[42:52.160 --> 42:54.660]  Because I think of this as an art project, even though it's software.
[42:56.780 --> 43:00.040]  You know, it's not like we're trying to solve a specific problem.
[43:00.040 --> 43:02.820]  But part of it is we're trying to make a statement.
[43:03.140 --> 43:06.500]  We're trying to get people thinking about the issues.
[43:07.560 --> 43:15.720]  You know, maybe cause some press to be generated about tracking that wasn't generated before, et cetera.
[43:17.420 --> 43:17.900]  Question.
[43:20.240 --> 43:20.720]  So...
[43:21.780 --> 43:22.740]  ...interesting idea.
[43:23.240 --> 43:25.000]  What is it creating risk for the user?
[43:25.260 --> 43:26.720]  Like, because you're clicking on every ad?
[43:26.840 --> 43:27.800]  What if it's a malicious ad?
[43:27.920 --> 43:32.920]  Now I've actually...
[43:40.640 --> 43:42.200]  I mean, what...
[43:42.880 --> 43:49.700]  So that implies that you have a malicious ad on a page that you have already visited intentionally.
[43:53.080 --> 43:54.500]  So that's a good question.
[43:54.880 --> 43:55.460]  I...
[43:55.460 --> 43:59.500]  We've discussed this in the past, but I don't know what conclusions we came to.
[43:59.760 --> 44:11.860]  How much we can sandbox the downloading and potentially running of active content on the page that happens in this hidden browser tab.
[44:13.480 --> 44:14.840]  But that's a good question.
[44:18.460 --> 44:20.060]  So lately it seems...
[44:21.100 --> 44:21.620]  Next.
[44:21.820 --> 44:22.020]  Okay.
[44:23.880 --> 44:24.180]  Me?
[44:24.560 --> 44:25.080]  Yeah.
[44:25.320 --> 44:25.460]  Yeah.
[44:26.000 --> 44:26.840]  I can't see you at all.
[44:37.030 --> 44:38.850]  But you have to go to business.
[44:40.210 --> 44:40.790]  So...
[44:40.790 --> 44:44.470]  The problem is now that's generated millions of caused surges.
[44:44.850 --> 44:48.330]  And that would be a risk for a line and more ammunition.
[44:48.710 --> 44:49.970]  Any of that is...
[44:49.970 --> 44:52.110]  One or two of the surges happens in action.
[44:52.530 --> 44:52.550]  The
[44:56.280 --> 44:57.260]  potential...
[44:57.260 --> 44:58.400]  I would have asked if one or two of the surges happens in action.
[44:58.400 --> 45:01.280]  So if you're ready to...
[45:01.280 --> 45:10.520]  Do we...attracting the attention of law enforcement, then this would make it worse, is the argument potentially make it worse?
[45:12.140 --> 45:13.500]  I guess that's possible.
[45:13.680 --> 45:24.680]  The critique we get more often than that is that for the average person, suddenly you find, you know, because the queries are dynamically generated from current events, right?
[45:24.880 --> 45:29.140]  So things like, you know, pipe bomb will show up sometimes.
[45:29.140 --> 45:44.920]  And we'll get people who are looking for, you know, the newest cornbread recipe, you know, suddenly searching for pipe bomb and, you know, child porn or something.
[45:46.320 --> 45:53.980]  But what it also provides, assuming it works correctly, is plausible deniability in all those cases, right?
[45:54.060 --> 46:02.920]  So you might actually benefit from that because you could say that all these other queries were potentially also generated by TrackMeNot.
[46:10.490 --> 46:12.470]  How would who be able to determine?
[46:42.900 --> 46:43.500]  Yourself.
[46:43.500 --> 46:46.180]  Do all your fake searches show up in your browser history?
[46:46.380 --> 46:46.380]  Yeah.
[46:46.580 --> 46:49.600]  So you can...not in your browser history...yes.
[46:49.700 --> 46:52.320]  In your browser...in your search history on Google, they show up.
[46:52.680 --> 46:56.560]  But they...and they also show up in the logs for TrackMeNot.
[46:56.680 --> 46:58.420]  So you...and they actually show up in real time.
[46:58.420 --> 47:01.200]  So you can see them as they're happening in the interface.
[47:01.780 --> 47:06.840]  So in the status bar, you'll see the query that was just launched.
[47:07.220 --> 47:11.900]  And the search engine is going to...zero.
[47:15.320 --> 47:18.000]  Happy to talk about this more outside.
[47:18.720 --> 47:23.720]  I had my...you can grab me on Twitter if you want.
[47:28.560 --> 47:30.000]  I can post it.
[47:30.200 --> 47:33.980]  It also has a lot of the references for...for this stuff if you...if you want that.
[47:34.100 --> 47:35.040]  But I can post it.
[47:38.340 --> 47:42.040]  I'll...I'll...I'll tweet...with...with the link.
[47:42.180 --> 47:42.260]  Yeah.
[47:43.260 --> 47:44.800]  This is my Twitter.
[47:46.580 --> 47:47.140]  Oops.
[47:47.620 --> 47:48.000]  Gone.
[47:53.070 --> 47:54.090]  At...Daniel C. Howe.
[47:54.990 --> 47:56.190]  If you can...remember.
[47:56.190 --> 47:57.370]  There it is.
[47:58.410 --> 47:59.710]  Feel free to send me a note.
[48:00.050 --> 48:01.210]  Happy to talk about this more.