[Script Info]
Title: 
[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:00.31,0:00:10.24,Default,,0000,0000,0000,,{\i1}32c3 preroll music{\i0}
Dialogue: 0,0:00:10.24,0:00:13.92,Default,,0000,0000,0000,,Angel: I introduce Whitney Merrill.\NShe is an attorney in the US
Dialogue: 0,0:00:13.92,0:00:17.26,Default,,0000,0000,0000,,and she just recently, actually\Nlast week, graduated
Dialogue: 0,0:00:17.26,0:00:20.100,Default,,0000,0000,0000,,to her CS masters in Illinois.
Dialogue: 0,0:00:20.100,0:00:27.30,Default,,0000,0000,0000,,{\i1}applause{\i0}
Dialogue: 0,0:00:27.30,0:00:30.25,Default,,0000,0000,0000,,Angel: Without further ado:\N‘Predicting Crime In A Big Data World’.
Dialogue: 0,0:00:30.25,0:00:32.87,Default,,0000,0000,0000,,{\i1}cautious applause{\i0}
Dialogue: 0,0:00:32.87,0:00:36.92,Default,,0000,0000,0000,,Whitney Merrill: Hi everyone.\NThank you so much for coming.
Dialogue: 0,0:00:36.92,0:00:40.95,Default,,0000,0000,0000,,I know it´s been a exhausting Congress,\Nso I appreciate you guys coming
Dialogue: 0,0:00:40.95,0:00:45.30,Default,,0000,0000,0000,,to hear me talk about Big\NData and Crime Prediction.
Dialogue: 0,0:00:45.30,0:00:48.82,Default,,0000,0000,0000,,This is kind of a hobby of mine, I,
Dialogue: 0,0:00:48.82,0:00:53.03,Default,,0000,0000,0000,,in my last semester at Illinois,\Ndecided to poke around
Dialogue: 0,0:00:53.03,0:00:56.85,Default,,0000,0000,0000,,what´s currently happening, how these\Nalgorithms are being used and kind of
Dialogue: 0,0:00:56.85,0:01:00.39,Default,,0000,0000,0000,,figure out what kind of information can be\Ngathered. So, I have about 30 minutes
Dialogue: 0,0:01:00.39,0:01:04.63,Default,,0000,0000,0000,,with you guys. I´m gonna do a broad\Noverview of the types of programs.
Dialogue: 0,0:01:04.63,0:01:10.02,Default,,0000,0000,0000,,I´m gonna talk about what Predictive\NPolicing is, the data used,
Dialogue: 0,0:01:10.02,0:01:13.60,Default,,0000,0000,0000,,similar systems in other areas\Nwhere predictive algorithms are
Dialogue: 0,0:01:13.60,0:01:19.08,Default,,0000,0000,0000,,trying to better society,\Ncurrent uses in policing.
Dialogue: 0,0:01:19.08,0:01:22.12,Default,,0000,0000,0000,,I´m gonna talk a little bit about their\Neffectiveness and then give you
Dialogue: 0,0:01:22.12,0:01:26.41,Default,,0000,0000,0000,,some final thoughts. So, imagine,
Dialogue: 0,0:01:26.41,0:01:30.31,Default,,0000,0000,0000,,in the very near future a Police\Nofficer is walking down the street
Dialogue: 0,0:01:30.31,0:01:34.39,Default,,0000,0000,0000,,wearing a camera on her collar.\NIn her ear is a feed of information
Dialogue: 0,0:01:34.39,0:01:38.82,Default,,0000,0000,0000,,about the people and cars she passes\Nalerting her to individuals and cars
Dialogue: 0,0:01:38.82,0:01:43.26,Default,,0000,0000,0000,,that might fit a particular crime\Nor profile for a criminal.
Dialogue: 0,0:01:43.26,0:01:47.62,Default,,0000,0000,0000,,Early in the day she examined a\Nmap highlighting hotspots for crime.
Dialogue: 0,0:01:47.62,0:01:52.46,Default,,0000,0000,0000,,In the area she´s been set to patrol\Nthe predictive policing software
Dialogue: 0,0:01:52.46,0:01:57.59,Default,,0000,0000,0000,,indicates that there is an 82%\Nchance of burglary at 2 pm,
Dialogue: 0,0:01:57.59,0:02:01.54,Default,,0000,0000,0000,,and it´s currently 2:10 pm.\NAs she passes one individual
Dialogue: 0,0:02:01.54,0:02:05.55,Default,,0000,0000,0000,,her camera captures the\Nindividual´s face, runs it through
Dialogue: 0,0:02:05.55,0:02:10.40,Default,,0000,0000,0000,,a coordinated Police database - all of the\NPolice departments that use this database
Dialogue: 0,0:02:10.40,0:02:14.68,Default,,0000,0000,0000,,are sharing information. Facial\Nrecognition software indicates that
Dialogue: 0,0:02:14.68,0:02:19.58,Default,,0000,0000,0000,,the person is Bobby Burglar who was\Npreviously convicted of burglary,
Dialogue: 0,0:02:19.58,0:02:24.79,Default,,0000,0000,0000,,was recently released and is now currently\Non patrole. The voice in her ear whispers:
Dialogue: 0,0:02:24.79,0:02:29.97,Default,,0000,0000,0000,,50 percent likely to commit a crime.\NCan she stop and search him?
Dialogue: 0,0:02:29.97,0:02:32.97,Default,,0000,0000,0000,,Should she chat him up?\NShould see how he acts?
Dialogue: 0,0:02:32.97,0:02:37.15,Default,,0000,0000,0000,,Does she need additional information\Nto stop and detain him?
Dialogue: 0,0:02:37.15,0:02:40.90,Default,,0000,0000,0000,,And does it matter that he´s\Ncarrying a large duffle bag?
Dialogue: 0,0:02:40.90,0:02:45.58,Default,,0000,0000,0000,,Did the algorithm take this into account\Nor did it just look at his face?
Dialogue: 0,0:02:45.58,0:02:49.94,Default,,0000,0000,0000,,What information was being\Ncollected at the time the algorithm
Dialogue: 0,0:02:49.94,0:02:55.26,Default,,0000,0000,0000,,chose to say 50% to provide\Nthe final analysis?
Dialogue: 0,0:02:55.26,0:02:57.93,Default,,0000,0000,0000,,So, another thought I´m gonna\Nhave you guys think about as I go
Dialogue: 0,0:02:57.93,0:03:01.54,Default,,0000,0000,0000,,through this presentation, is this\Nquote that is more favorable
Dialogue: 0,0:03:01.54,0:03:05.87,Default,,0000,0000,0000,,towards Police algorithms, which is:\N“As people become data plots
Dialogue: 0,0:03:05.87,0:03:10.21,Default,,0000,0000,0000,,and probability scores, law enforcement\Nofficials and politicians alike
Dialogue: 0,0:03:10.21,0:03:16.52,Default,,0000,0000,0000,,can point and say: ‘Technology is void of\Nthe racist, profiling bias of humans.’”
Dialogue: 0,0:03:16.52,0:03:21.46,Default,,0000,0000,0000,,Is that true? Well, they probably\Nwill point and say that,
Dialogue: 0,0:03:21.46,0:03:24.86,Default,,0000,0000,0000,,but is it actually void of\Nracist, profiling humans?
Dialogue: 0,0:03:24.86,0:03:27.85,Default,,0000,0000,0000,,And I´m gonna talk about that as well.
Dialogue: 0,0:03:27.85,0:03:32.76,Default,,0000,0000,0000,,So, Predictive Policing explained.\NWho and what?
Dialogue: 0,0:03:32.76,0:03:35.62,Default,,0000,0000,0000,,First of all, Predictive Policing\Nactually isn´t new. All we´re doing
Dialogue: 0,0:03:35.62,0:03:41.47,Default,,0000,0000,0000,,is adding technology, doing better,\Nfaster aggregation of data.
Dialogue: 0,0:03:41.47,0:03:47.20,Default,,0000,0000,0000,,Analysts in Police departments have been\Ndoing this by hand for decades.
Dialogue: 0,0:03:47.20,0:03:50.95,Default,,0000,0000,0000,,These techniques are used to create\Nprofiles that accurately match
Dialogue: 0,0:03:50.95,0:03:55.53,Default,,0000,0000,0000,,likely offenders with specific past\Ncrimes. So, there´s individual targeting
Dialogue: 0,0:03:55.53,0:03:59.49,Default,,0000,0000,0000,,and then we have location-based\Ntargeting. The location-based,
Dialogue: 0,0:03:59.49,0:04:05.01,Default,,0000,0000,0000,,the goal is to help Police\Nforces deploy their resources
Dialogue: 0,0:04:05.01,0:04:10.23,Default,,0000,0000,0000,,in a correct manner, in an efficient\Nmanner. They can be as simple
Dialogue: 0,0:04:10.23,0:04:13.95,Default,,0000,0000,0000,,as recommending that general crime\Nmay happen in a particular area,
Dialogue: 0,0:04:13.95,0:04:19.11,Default,,0000,0000,0000,,or specifically, what type of crime will\Nhappen in a one-block-radius.
Dialogue: 0,0:04:19.11,0:04:24.05,Default,,0000,0000,0000,,They take into account the time\Nof day, the recent data collected
Dialogue: 0,0:04:24.05,0:04:30.04,Default,,0000,0000,0000,,and when in the year it´s happening\Nas well as weather etc.
Dialogue: 0,0:04:30.04,0:04:33.85,Default,,0000,0000,0000,,So, another really quick thing worth\Ngoing over, cause not everyone
Dialogue: 0,0:04:33.85,0:04:39.09,Default,,0000,0000,0000,,is familiar with machine learning.\NThis is a very basic breakdown
Dialogue: 0,0:04:39.09,0:04:43.07,Default,,0000,0000,0000,,of training an algorithm on a data set.
Dialogue: 0,0:04:43.07,0:04:46.24,Default,,0000,0000,0000,,You collect it from many different\Nsources, you put it all together,
Dialogue: 0,0:04:46.24,0:04:51.02,Default,,0000,0000,0000,,you clean it up, you split it into 3 sets:\Na training set, a validation set
Dialogue: 0,0:04:51.02,0:04:56.35,Default,,0000,0000,0000,,and a test set. The training set is\Nwhat is going to develop the rules
Dialogue: 0,0:04:56.35,0:05:01.38,Default,,0000,0000,0000,,in which it´s going to kind of\Ndetermine the final outcome.
Dialogue: 0,0:05:01.38,0:05:05.06,Default,,0000,0000,0000,,You´re gonna use a validation\Nset to optimize it and finally
Dialogue: 0,0:05:05.06,0:05:09.73,Default,,0000,0000,0000,,apply this to establish\Na confidence level.
Dialogue: 0,0:05:09.73,0:05:15.35,Default,,0000,0000,0000,,There you´ll set a support level where\Nyou say you need a certain amount of data
Dialogue: 0,0:05:15.35,0:05:19.94,Default,,0000,0000,0000,,to determine whether or not the\Nalgorithm has enough information
Dialogue: 0,0:05:19.94,0:05:24.19,Default,,0000,0000,0000,,to kind of make a prediction.\NSo, rules with a low support level
Dialogue: 0,0:05:24.19,0:05:28.76,Default,,0000,0000,0000,,are less likely to be statistically\Nsignificant and the confidence level
Dialogue: 0,0:05:28.76,0:05:34.10,Default,,0000,0000,0000,,in the end is basically if there´s\Nan 85% confidence level
Dialogue: 0,0:05:34.10,0:05:39.93,Default,,0000,0000,0000,,that means there´s an 85% chance that the\Nsuspect, e.g. meeting the rule in question,
Dialogue: 0,0:05:39.93,0:05:45.14,Default,,0000,0000,0000,,is engaged in criminal conduct.\NSo, what does this mean? Well,
Dialogue: 0,0:05:45.14,0:05:49.59,Default,,0000,0000,0000,,it encourages collection and hoarding\Nof data about crimes and individuals.
Dialogue: 0,0:05:49.59,0:05:52.72,Default,,0000,0000,0000,,Because you want as much information\Nas possible so that you detect
Dialogue: 0,0:05:52.72,0:05:56.03,Default,,0000,0000,0000,,even the less likely scenarios.
Dialogue: 0,0:05:56.03,0:05:59.73,Default,,0000,0000,0000,,Information sharing is also\Nencouraged because it´s easier,
Dialogue: 0,0:05:59.73,0:06:04.09,Default,,0000,0000,0000,,it´s done by third parties, or even\Nwhat are called fourth parties
Dialogue: 0,0:06:04.09,0:06:07.86,Default,,0000,0000,0000,,and shared amongst departments.\NAnd here, your criminal data again
Dialogue: 0,0:06:07.86,0:06:10.81,Default,,0000,0000,0000,,was being done by analysts in Police\Ndepartments for decades, but
Dialogue: 0,0:06:10.81,0:06:13.66,Default,,0000,0000,0000,,the information sharing and the amount\Nof information they could aggregate
Dialogue: 0,0:06:13.66,0:06:17.17,Default,,0000,0000,0000,,was just significantly more difficult. So,
Dialogue: 0,0:06:17.17,0:06:21.41,Default,,0000,0000,0000,,what are these Predictive Policing\Nalgorithms and software…
Dialogue: 0,0:06:21.41,0:06:24.58,Default,,0000,0000,0000,,what are they doing? Are they\Ndetermining guilt and innocence?
Dialogue: 0,0:06:24.58,0:06:29.29,Default,,0000,0000,0000,,And, unlike a thoughtcrime, they\Nare not saying this person is guilty,
Dialogue: 0,0:06:29.29,0:06:33.29,Default,,0000,0000,0000,,this person is innocent. It´s creating\Na probability of whether or not
Dialogue: 0,0:06:33.29,0:06:37.80,Default,,0000,0000,0000,,the person has likely committed\Na crime or will likely commit a crime.
Dialogue: 0,0:06:37.80,0:06:41.03,Default,,0000,0000,0000,,And it can only say something\Nto the future and the past.
Dialogue: 0,0:06:41.03,0:06:46.31,Default,,0000,0000,0000,,This here is a picture from\None particular piece of software
Dialogue: 0,0:06:46.31,0:06:50.20,Default,,0000,0000,0000,,provided by HunchLab; and patterns\Nemerge here from past crimes
Dialogue: 0,0:06:50.20,0:06:58.23,Default,,0000,0000,0000,,that can profile criminal types and\Nassociations, detect crime patterns etc.
Dialogue: 0,0:06:58.23,0:07:02.14,Default,,0000,0000,0000,,Generally in this types of algorithms\Nthey are using unsupervised data,
Dialogue: 0,0:07:02.14,0:07:05.48,Default,,0000,0000,0000,,that means someone is not going through\Nand saying true-false, good-bad, good-bad.
Dialogue: 0,0:07:05.48,0:07:10.78,Default,,0000,0000,0000,,There´s just 1) too much information and\N2) they´re trying to do clustering,
Dialogue: 0,0:07:10.78,0:07:15.28,Default,,0000,0000,0000,,determine the things that are similar.
Dialogue: 0,0:07:15.28,0:07:20.11,Default,,0000,0000,0000,,So, really quickly, I´m also gonna\Ntalk about the data that´s used.
Dialogue: 0,0:07:20.11,0:07:23.26,Default,,0000,0000,0000,,There are several different types:\NPersonal characteristics,
Dialogue: 0,0:07:23.26,0:07:28.17,Default,,0000,0000,0000,,demographic information, activities\Nof individuals, scientific data etc.
Dialogue: 0,0:07:28.17,0:07:32.69,Default,,0000,0000,0000,,This comes from all sorts of sources,\None that really shocked me, was,
Dialogue: 0,0:07:32.69,0:07:36.86,Default,,0000,0000,0000,,and I´ll talk about it a little bit in the\Nfuture, but, is the radiation detectors
Dialogue: 0,0:07:36.86,0:07:41.31,Default,,0000,0000,0000,,on New York City Police are\Nconstantly taking in data
Dialogue: 0,0:07:41.31,0:07:44.82,Default,,0000,0000,0000,,and it´s so sensitive, it can detect if\Nyou´ve had a recent medical treatment
Dialogue: 0,0:07:44.82,0:07:49.33,Default,,0000,0000,0000,,that involves radiation. Facial\Nrecognition and biometrics
Dialogue: 0,0:07:49.33,0:07:52.86,Default,,0000,0000,0000,,are clear here and the third-party\Ndoctrine – which basically says
Dialogue: 0,0:07:52.86,0:07:56.55,Default,,0000,0000,0000,,in the United States that you have no\Nreasonable expectation of privacy in data
Dialogue: 0,0:07:56.55,0:08:01.49,Default,,0000,0000,0000,,you share with third parties –\Nfacilitates easy collection
Dialogue: 0,0:08:01.49,0:08:05.72,Default,,0000,0000,0000,,for Police officers and Government\Nofficials because they can go
Dialogue: 0,0:08:05.72,0:08:11.08,Default,,0000,0000,0000,,and ask for the information\Nwithout any sort of warrant.
Dialogue: 0,0:08:11.08,0:08:16.26,Default,,0000,0000,0000,,For a really great overview: a friend of\Nmine, Dia, did a talk here at CCC
Dialogue: 0,0:08:16.26,0:08:21.28,Default,,0000,0000,0000,,on “The architecture of a street level\Npanopticon”. Does a really great overview
Dialogue: 0,0:08:21.28,0:08:25.20,Default,,0000,0000,0000,,of how this type of data is collected\Non the streets. Worth checking out
Dialogue: 0,0:08:25.20,0:08:29.49,Default,,0000,0000,0000,,´cause I´m gonna gloss over\Nkind of the types of data.
Dialogue: 0,0:08:29.49,0:08:33.45,Default,,0000,0000,0000,,There is in the United States what\Nthey call Multistate Anti-Terrorism
Dialogue: 0,0:08:33.45,0:08:38.28,Default,,0000,0000,0000,,Information Exchange Program which\Nuses everything from credit history,
Dialogue: 0,0:08:38.28,0:08:42.03,Default,,0000,0000,0000,,your concealed weapons permits,\Naircraft pilot licenses,
Dialogue: 0,0:08:42.03,0:08:46.80,Default,,0000,0000,0000,,fishing licences etc. that´s searchable\Nand shared amongst Police departments
Dialogue: 0,0:08:46.80,0:08:50.53,Default,,0000,0000,0000,,and Government officials and this is just\Nmore information. So, if they can collect
Dialogue: 0,0:08:50.53,0:08:57.69,Default,,0000,0000,0000,,it, they will aggregate it into a data\Nbase. So, what are the current uses?
Dialogue: 0,0:08:57.69,0:09:01.78,Default,,0000,0000,0000,,There are many, many different\Ncompanies currently
Dialogue: 0,0:09:01.78,0:09:04.95,Default,,0000,0000,0000,,making software and marketing\Nit to Police departments.
Dialogue: 0,0:09:04.95,0:09:08.47,Default,,0000,0000,0000,,All of them are slightly different, have\Ndifferent features, but currently
Dialogue: 0,0:09:08.47,0:09:12.26,Default,,0000,0000,0000,,it´s a competition to get clients,\NPolice departments etc.
Dialogue: 0,0:09:12.26,0:09:15.83,Default,,0000,0000,0000,,The more Police departments you have\Nthe more data sharing you can sell,
Dialogue: 0,0:09:15.83,0:09:21.09,Default,,0000,0000,0000,,saying: “Oh, by enrolling you’ll now have\Nx,y and z Police departments’ data
Dialogue: 0,0:09:21.09,0:09:27.04,Default,,0000,0000,0000,,to access” etc. These here\Nare Hitachi and HunchLab,
Dialogue: 0,0:09:27.04,0:09:31.35,Default,,0000,0000,0000,,they both are hotspot targeting,\Nit´s not individual targeting,
Dialogue: 0,0:09:31.35,0:09:35.14,Default,,0000,0000,0000,,those are a lot rarer. And it´s actually\Nbeing used in my home town,
Dialogue: 0,0:09:35.14,0:09:39.55,Default,,0000,0000,0000,,which I´ll talk about in a little bit.\NHere, the appropriate tactics
Dialogue: 0,0:09:39.55,0:09:44.18,Default,,0000,0000,0000,,are automatically displayed for officers\Nwhen they´re entering mission areas.
Dialogue: 0,0:09:44.18,0:09:47.92,Default,,0000,0000,0000,,So HunchLab will tell an officer:\N“Hey, you´re entering an area
Dialogue: 0,0:09:47.92,0:09:52.18,Default,,0000,0000,0000,,where there´s gonna be burglary that you\Nshould keep an eye out, be aware”.
Dialogue: 0,0:09:52.18,0:09:58.01,Default,,0000,0000,0000,,And this is updating in live time and\Nthey´re hoping it mitigates crime.
Dialogue: 0,0:09:58.01,0:10:01.24,Default,,0000,0000,0000,,Here are 2 other ones, the Domain\NAwareness System was created
Dialogue: 0,0:10:01.24,0:10:05.14,Default,,0000,0000,0000,,in New York City after 9/11\Nin conjunction with Microsoft.
Dialogue: 0,0:10:05.14,0:10:10.00,Default,,0000,0000,0000,,New York City actually makes\Nmoney selling it to other cities
Dialogue: 0,0:10:10.00,0:10:16.47,Default,,0000,0000,0000,,to use this. CCTV-cameras\Nare collected, they can…
Dialogue: 0,0:10:16.47,0:10:21.03,Default,,0000,0000,0000,,If they say there´s a man\Nwearing a red shirt,
Dialogue: 0,0:10:21.03,0:10:24.43,Default,,0000,0000,0000,,the software will look for people\Nwearing red shirts and at least
Dialogue: 0,0:10:24.43,0:10:28.14,Default,,0000,0000,0000,,alert Police departments to\Npeople that meet this description
Dialogue: 0,0:10:28.14,0:10:34.39,Default,,0000,0000,0000,,walking in public in New York\NCity. The other one is by IBM
Dialogue: 0,0:10:34.39,0:10:40.14,Default,,0000,0000,0000,,and there are quite a few, you know, it´s\Njust generally another hotspot targeting,
Dialogue: 0,0:10:40.14,0:10:45.84,Default,,0000,0000,0000,,each have a few different features.\NWorth mentioning, too, is the Heat List.
Dialogue: 0,0:10:45.84,0:10:50.77,Default,,0000,0000,0000,,This targeted individuals. I’m from the\Ncity of Chicago. I grew up in the city.
Dialogue: 0,0:10:50.77,0:10:55.15,Default,,0000,0000,0000,,There are currently 420 names, when\Nthis came out about a year ago,
Dialogue: 0,0:10:55.15,0:10:59.92,Default,,0000,0000,0000,,of individuals who are 500 times more\Nlikely than average to be involved
Dialogue: 0,0:10:59.92,0:11:05.23,Default,,0000,0000,0000,,in violence. Individual names, passed\Naround to each Police officer in Chicago.
Dialogue: 0,0:11:05.23,0:11:10.03,Default,,0000,0000,0000,,They consider the rap sheet,\Ndisturbance calls, social network etc.
Dialogue: 0,0:11:10.03,0:11:15.54,Default,,0000,0000,0000,,But one of the main things they considered\Nin placing mainly young black individuals
Dialogue: 0,0:11:15.54,0:11:19.28,Default,,0000,0000,0000,,on this list were known acquaintances\Nand their arrest histories.
Dialogue: 0,0:11:19.28,0:11:23.28,Default,,0000,0000,0000,,So if kids went to school or young\Nteenagers went to school
Dialogue: 0,0:11:23.28,0:11:27.88,Default,,0000,0000,0000,,with several people in a gang – and that\Nindividual may not even be involved
Dialogue: 0,0:11:27.88,0:11:32.16,Default,,0000,0000,0000,,in a gang – they’re more likely to\Nappear on the list. The list has been
Dialogue: 0,0:11:32.16,0:11:36.83,Default,,0000,0000,0000,,heavily criticized for being racist,\Nfor not giving these children
Dialogue: 0,0:11:36.83,0:11:40.66,Default,,0000,0000,0000,,or young individuals on the list\Na chance to change their history
Dialogue: 0,0:11:40.66,0:11:44.51,Default,,0000,0000,0000,,because it’s being decided for them.\NThey’re being told: “You are likely
Dialogue: 0,0:11:44.51,0:11:49.85,Default,,0000,0000,0000,,to be a criminal, and we’re gonna\Nwatch you”. Officers in Chicago
Dialogue: 0,0:11:49.85,0:11:53.55,Default,,0000,0000,0000,,visited these individuals would do knock\Nand announce with a knock on the door
Dialogue: 0,0:11:53.55,0:11:58.03,Default,,0000,0000,0000,,and say: “Hi, I’m here, like just\Nchecking up what are you up to”.
Dialogue: 0,0:11:58.03,0:12:02.48,Default,,0000,0000,0000,,Which you don’t need any special\Nsuspicion to do. But it’s, you know,
Dialogue: 0,0:12:02.48,0:12:06.86,Default,,0000,0000,0000,,kind of a harassment that\Nmight cause a feedback,
Dialogue: 0,0:12:06.86,0:12:11.31,Default,,0000,0000,0000,,back into the data collected.
Dialogue: 0,0:12:11.31,0:12:15.21,Default,,0000,0000,0000,,This is PRECOBS. It’s currently\Nused here in Hamburg.
Dialogue: 0,0:12:15.21,0:12:19.10,Default,,0000,0000,0000,,They actually went to Chicago and\Nvisited the Chicago Police Department
Dialogue: 0,0:12:19.10,0:12:24.17,Default,,0000,0000,0000,,to learn about Predictive Policing\Ntactics in Chicago to implement it
Dialogue: 0,0:12:24.17,0:12:29.73,Default,,0000,0000,0000,,throughout Germany, Hamburg and Berlin.
Dialogue: 0,0:12:29.73,0:12:33.62,Default,,0000,0000,0000,,It’s used to generally\Nforecast repeat-offenses.
Dialogue: 0,0:12:33.62,0:12:39.93,Default,,0000,0000,0000,,Again, when training data sets you need\Nenough data points to predict crime.
Dialogue: 0,0:12:39.93,0:12:43.70,Default,,0000,0000,0000,,So crimes that are less likely to\Nhappen or happen very rarely:
Dialogue: 0,0:12:43.70,0:12:48.12,Default,,0000,0000,0000,,much harder to predict. Crimes that\Naren’t reported: much harder to predict.
Dialogue: 0,0:12:48.12,0:12:52.48,Default,,0000,0000,0000,,So a lot of these software…\Nlike pieces of software
Dialogue: 0,0:12:52.48,0:12:58.29,Default,,0000,0000,0000,,rely on algorithms that are hoping\Nthat there’s a same sort of picture,
Dialogue: 0,0:12:58.29,0:13:03.07,Default,,0000,0000,0000,,that they can predict: where and when\Nand what type of crime will happen.
Dialogue: 0,0:13:03.07,0:13:06.89,Default,,0000,0000,0000,,PRECOBS is actually a term with a plan
Dialogue: 0,0:13:06.89,0:13:11.24,Default,,0000,0000,0000,,– the movie ‘Minority Report’, if you’re\Nfamiliar with it, it’s the 3 psychics
Dialogue: 0,0:13:11.24,0:13:15.37,Default,,0000,0000,0000,,who predict crimes\Nbefore they happen.
Dialogue: 0,0:13:15.37,0:13:19.15,Default,,0000,0000,0000,,So there’re other, similar systems\Nin the world that are being used
Dialogue: 0,0:13:19.15,0:13:22.95,Default,,0000,0000,0000,,to predict whether or not\Nsomething will happen.
Dialogue: 0,0:13:22.95,0:13:27.36,Default,,0000,0000,0000,,The first one is ‘Disease and Diagnosis’.\NThey found that algorithms are actually
Dialogue: 0,0:13:27.36,0:13:33.81,Default,,0000,0000,0000,,more likely than doctors to predict\Nwhat disease an individual has.
Dialogue: 0,0:13:33.81,0:13:39.48,Default,,0000,0000,0000,,It’s kind of shocking. The other is\N‘Security Clearance’ in the US.
Dialogue: 0,0:13:39.48,0:13:44.24,Default,,0000,0000,0000,,It allows access to classified documents.\NThere’s no automatic access in the US.
Dialogue: 0,0:13:44.24,0:13:48.75,Default,,0000,0000,0000,,So every person who wants to see\Nsome sort of secret cleared document
Dialogue: 0,0:13:48.75,0:13:53.09,Default,,0000,0000,0000,,must go through this process.\NAnd it’s vetting individuals.
Dialogue: 0,0:13:53.09,0:13:56.69,Default,,0000,0000,0000,,So it’s an opt-in process. But here\Nthey’re trying to predict who will
Dialogue: 0,0:13:56.69,0:14:00.55,Default,,0000,0000,0000,,disclose information, who will\Nbreak the clearance system;
Dialogue: 0,0:14:00.55,0:14:05.81,Default,,0000,0000,0000,,and predict there… Here, the error rate,\Nthey’re probably much more comfortable
Dialogue: 0,0:14:05.81,0:14:09.36,Default,,0000,0000,0000,,with a high error rate. Because they\Nhave so many people competing
Dialogue: 0,0:14:09.36,0:14:13.70,Default,,0000,0000,0000,,for a particular job, to get\Nclearance, that if they’re wrong,
Dialogue: 0,0:14:13.70,0:14:18.00,Default,,0000,0000,0000,,that somebody probably won’t disclose\Ninformation, they don’t care,
Dialogue: 0,0:14:18.00,0:14:22.32,Default,,0000,0000,0000,,they just rather eliminate\Nthem than take the risk.
Dialogue: 0,0:14:22.32,0:14:27.51,Default,,0000,0000,0000,,So I’m an attorney in the US. I have\Nthis urge to talk about US law.
Dialogue: 0,0:14:27.51,0:14:32.09,Default,,0000,0000,0000,,It also seems to impact a lot\Nof people internationally.
Dialogue: 0,0:14:32.09,0:14:36.36,Default,,0000,0000,0000,,Here we’re talking about the targeting\Nof individuals, not hotspots.
Dialogue: 0,0:14:36.36,0:14:40.81,Default,,0000,0000,0000,,So targeting of individuals is\Nnot as widespread, currently.
Dialogue: 0,0:14:40.81,0:14:45.58,Default,,0000,0000,0000,,However it’s happening in Chicago;
Dialogue: 0,0:14:45.58,0:14:49.26,Default,,0000,0000,0000,,and other cities are considering\Nimplementing programs and there are grants
Dialogue: 0,0:14:49.26,0:14:53.73,Default,,0000,0000,0000,,right now to encourage\NPolice departments
Dialogue: 0,0:14:53.73,0:14:57.11,Default,,0000,0000,0000,,to figure out target lists.
Dialogue: 0,0:14:57.11,0:15:00.70,Default,,0000,0000,0000,,So in the US suspicion is based on\Nthe totality of the circumstances.
Dialogue: 0,0:15:00.70,0:15:04.73,Default,,0000,0000,0000,,That’s the whole picture. The Police\Nofficer, the individual must look
Dialogue: 0,0:15:04.73,0:15:08.27,Default,,0000,0000,0000,,at the whole picture of what’s happening\Nbefore they can detain an individual.
Dialogue: 0,0:15:08.27,0:15:11.92,Default,,0000,0000,0000,,It’s supposed to be a balanced\Nassessment of relative weights, meaning
Dialogue: 0,0:15:11.92,0:15:16.40,Default,,0000,0000,0000,,– you know – if you know that the\Nperson is a pastor maybe then
Dialogue: 0,0:15:16.40,0:15:21.72,Default,,0000,0000,0000,,pacing in front of a liquor\Nstore, is not as suspicious
Dialogue: 0,0:15:21.72,0:15:26.37,Default,,0000,0000,0000,,as somebody who’s been convicted\Nof 3 burglaries. It has to be ‘based
Dialogue: 0,0:15:26.37,0:15:31.43,Default,,0000,0000,0000,,on specific and articulable facts’. And\Nthe Police officers can use experience
Dialogue: 0,0:15:31.43,0:15:37.47,Default,,0000,0000,0000,,and common sense to determine\Nwhether or not their suspicion…
Dialogue: 0,0:15:37.47,0:15:42.92,Default,,0000,0000,0000,,Large amounts of networked data generally\Ncan provide individualized suspicion.
Dialogue: 0,0:15:42.92,0:15:48.41,Default,,0000,0000,0000,,The principal components here… the\Nevents leading up to the stop-and-search
Dialogue: 0,0:15:48.41,0:15:52.32,Default,,0000,0000,0000,,– what is the person doing right before\Nthey’re detained as well as the use
Dialogue: 0,0:15:52.32,0:15:57.71,Default,,0000,0000,0000,,of historical facts known about that\Nindividual, the crime, the area
Dialogue: 0,0:15:57.71,0:16:02.33,Default,,0000,0000,0000,,in which it’s happening etc.\NSo it can rely on both things.
Dialogue: 0,0:16:02.33,0:16:06.82,Default,,0000,0000,0000,,No court in the US has really put out\Na percentage as what Probable Cause
Dialogue: 0,0:16:06.82,0:16:11.09,Default,,0000,0000,0000,,and Reasonable Suspicion. So ‘Probable\NCause’ you need to get a warrant
Dialogue: 0,0:16:11.09,0:16:14.64,Default,,0000,0000,0000,,to search and seize an individual.\N‘Reasonable Suspicion’ is needed
Dialogue: 0,0:16:14.64,0:16:20.33,Default,,0000,0000,0000,,to do stop-and-frisk in the US – stop\Nan individual and question them.
Dialogue: 0,0:16:20.33,0:16:24.10,Default,,0000,0000,0000,,And this is a little bit different than\Nwhat they call ‘Consensual Encounters’,
Dialogue: 0,0:16:24.10,0:16:27.68,Default,,0000,0000,0000,,where a Police officer goes up to you and\Nchats you up. ‘Reasonable Suspicion’
Dialogue: 0,0:16:27.68,0:16:32.03,Default,,0000,0000,0000,,– you’re actually detained. But I had\Na law professor who basically said:
Dialogue: 0,0:16:32.03,0:16:35.73,Default,,0000,0000,0000,,“30%..45% seem like a really good number
Dialogue: 0,0:16:35.73,0:16:39.29,Default,,0000,0000,0000,,just to show how low it really is”.You\Ndon’t even need to be 50% sure
Dialogue: 0,0:16:39.29,0:16:42.18,Default,,0000,0000,0000,,that somebody has committed a crime.
Dialogue: 0,0:16:42.18,0:16:47.46,Default,,0000,0000,0000,,So, officers can draw from their own\Nexperience to determine ‘Probable Cause’.
Dialogue: 0,0:16:47.46,0:16:51.35,Default,,0000,0000,0000,,And the UK has a similar\N‘Reasonable Suspicion’ standard
Dialogue: 0,0:16:51.35,0:16:55.01,Default,,0000,0000,0000,,which depend on the circumstances\Nof each case. So,
Dialogue: 0,0:16:55.01,0:16:58.82,Default,,0000,0000,0000,,I’m not as familiar with UK law but I\Nbelieve even that some of the analysis-run
Dialogue: 0,0:16:58.82,0:17:03.48,Default,,0000,0000,0000,,‘Reasonable Suspicion’ is similar.
Dialogue: 0,0:17:03.48,0:17:07.34,Default,,0000,0000,0000,,Is this like a black box?\NSo, I threw this slide in
Dialogue: 0,0:17:07.34,0:17:10.96,Default,,0000,0000,0000,,for those who are interested\Nin comparing this US law.
Dialogue: 0,0:17:10.96,0:17:15.28,Default,,0000,0000,0000,,Generally a dog sniff in the US\Nfalls under a particular set
Dialogue: 0,0:17:15.28,0:17:20.14,Default,,0000,0000,0000,,of legal history which is: a\Ndog can go up, sniff for dogs,
Dialogue: 0,0:17:20.14,0:17:24.22,Default,,0000,0000,0000,,alert and that is completely okay.
Dialogue: 0,0:17:24.22,0:17:28.10,Default,,0000,0000,0000,,And the Police officers can use that\Ndata to detain and further search
Dialogue: 0,0:17:28.10,0:17:33.52,Default,,0000,0000,0000,,an individual. So is an algorithm similar\Nto the dog which is kind of a black box?
Dialogue: 0,0:17:33.52,0:17:37.03,Default,,0000,0000,0000,,Information goes out, it’s processed,\Ninformation comes out and
Dialogue: 0,0:17:37.03,0:17:42.72,Default,,0000,0000,0000,,a prediction is made.\NPolice rely on the ‘Good Faith’
Dialogue: 0,0:17:42.72,0:17:48.78,Default,,0000,0000,0000,,in ‘Totality of the Circumstances’\Nto make their decision. So there’s
Dialogue: 0,0:17:48.78,0:17:53.97,Default,,0000,0000,0000,,really no… if they’re\Nrelying on the algorithm
Dialogue: 0,0:17:53.97,0:17:57.23,Default,,0000,0000,0000,,and think in that situation that\Neverything’s okay we might reach
Dialogue: 0,0:17:57.23,0:18:01.98,Default,,0000,0000,0000,,a level of ‘Reasonable Suspicion’ where\Nthe individual can now pat down
Dialogue: 0,0:18:01.98,0:18:08.47,Default,,0000,0000,0000,,the person he’s decided on the street\Nor the algorithm has alerted to. So,
Dialogue: 0,0:18:08.47,0:18:13.22,Default,,0000,0000,0000,,the big question is, you know, “Could the\Nofficer consult predictive software apps
Dialogue: 0,0:18:13.22,0:18:18.61,Default,,0000,0000,0000,,in any individual analysis. Could he\Nsay: ‘60% likely to commit a crime’”.
Dialogue: 0,0:18:18.61,0:18:24.18,Default,,0000,0000,0000,,In my hypo: Does that\Nmean that the person
Dialogue: 0,0:18:24.18,0:18:29.16,Default,,0000,0000,0000,,without looking at anything\Nelse detain that individual.
Dialogue: 0,0:18:29.16,0:18:33.81,Default,,0000,0000,0000,,And the answer is “Probably not”. One:\Npredictive Policing algorithms just
Dialogue: 0,0:18:33.81,0:18:37.77,Default,,0000,0000,0000,,can not take in the Totality of the\NCircumstances. They have to be
Dialogue: 0,0:18:37.77,0:18:42.69,Default,,0000,0000,0000,,frequently updated, there are\Nthings that are happening that
Dialogue: 0,0:18:42.69,0:18:46.06,Default,,0000,0000,0000,,the algorithm possibly could\Nnot have taken into account.
Dialogue: 0,0:18:46.06,0:18:48.59,Default,,0000,0000,0000,,The problem here is\Nthat the algorithm itself,
Dialogue: 0,0:18:48.59,0:18:51.78,Default,,0000,0000,0000,,the prediction itself becomes part\Nof Totality of the Circumstances,
Dialogue: 0,0:18:51.78,0:18:56.33,Default,,0000,0000,0000,,which I’m going to talk\Nabout a little bit more later.
Dialogue: 0,0:18:56.33,0:19:00.66,Default,,0000,0000,0000,,But officers have to have Reasonable\NSuspicion before the stop occurs.
Dialogue: 0,0:19:00.66,0:19:04.66,Default,,0000,0000,0000,,Retroactive justification\Nis not sufficient. So,
Dialogue: 0,0:19:04.66,0:19:08.79,Default,,0000,0000,0000,,the algorithm can’t just say:\N“60% likely, you detain the individual
Dialogue: 0,0:19:08.79,0:19:12.13,Default,,0000,0000,0000,,and then figure out why you’ve\Ndetained the person”. It has to be
Dialogue: 0,0:19:12.13,0:19:16.57,Default,,0000,0000,0000,,before the detention actually happens.\NAnd the suspicion must relate
Dialogue: 0,0:19:16.57,0:19:19.99,Default,,0000,0000,0000,,to current criminal activity. The\Nperson must be doing something
Dialogue: 0,0:19:19.99,0:19:24.70,Default,,0000,0000,0000,,to indicate criminal activity. Just\Nthe fact that an algorithm says,
Dialogue: 0,0:19:24.70,0:19:29.44,Default,,0000,0000,0000,,based on these facts: “60%”,\Nor even without articulating
Dialogue: 0,0:19:29.44,0:19:33.89,Default,,0000,0000,0000,,why the algorithm has\Nchosen that, isn’t enough.
Dialogue: 0,0:19:33.89,0:19:38.38,Default,,0000,0000,0000,,Maybe you can see a gun\Nshaped bulge in the pocket etc.
Dialogue: 0,0:19:38.38,0:19:43.16,Default,,0000,0000,0000,,So, effectiveness… the\NTotality of the Circumstances,
Dialogue: 0,0:19:43.16,0:19:46.72,Default,,0000,0000,0000,,can the algorithms keep up?\NGenerally, probably not.
Dialogue: 0,0:19:46.72,0:19:50.56,Default,,0000,0000,0000,,Missing data, not capable of\Nprocessing this data in real time.
Dialogue: 0,0:19:50.56,0:19:54.82,Default,,0000,0000,0000,,There’s no idea… the\Nalgorithm doesn’t know,
Dialogue: 0,0:19:54.82,0:19:58.95,Default,,0000,0000,0000,,and the Police officer probably\Ndoesn’t know the all of the facts.
Dialogue: 0,0:19:58.95,0:20:03.26,Default,,0000,0000,0000,,So the Police officer can take\Nthe algorithm into consideration
Dialogue: 0,0:20:03.26,0:20:08.13,Default,,0000,0000,0000,,but the problem here is: Did the algorithm\Nknow that the individual was active
Dialogue: 0,0:20:08.13,0:20:12.67,Default,,0000,0000,0000,,in the community, or was a politician, or
Dialogue: 0,0:20:12.67,0:20:17.45,Default,,0000,0000,0000,,that was a personal friend of the officer\Netc. It can’t just be relied upon.
Dialogue: 0,0:20:17.45,0:20:22.64,Default,,0000,0000,0000,,What if the algorithm did take into\Naccount that the individual was a Pastor?
Dialogue: 0,0:20:22.64,0:20:26.18,Default,,0000,0000,0000,,Now that information is counted twice\Nand the balancing for the Totality
Dialogue: 0,0:20:26.18,0:20:34.32,Default,,0000,0000,0000,,of the Circumstances is off. Humans\Nhere must be the final decider.
Dialogue: 0,0:20:34.32,0:20:38.04,Default,,0000,0000,0000,,What are the problems?\NWell, there’s bad underlying data,
Dialogue: 0,0:20:38.04,0:20:41.97,Default,,0000,0000,0000,,there’s no transparency into\Nwhat kind of data is being used,
Dialogue: 0,0:20:41.97,0:20:45.72,Default,,0000,0000,0000,,how it was collected, how old it\Nis, how often it’s been updated,
Dialogue: 0,0:20:45.72,0:20:51.01,Default,,0000,0000,0000,,whether or not it’s been verified. There\Ncould just be noise in the training data.
Dialogue: 0,0:20:51.01,0:20:57.24,Default,,0000,0000,0000,,Honestly, the data is biased. It was\Ncollected by individuals in the US;
Dialogue: 0,0:20:57.24,0:21:01.02,Default,,0000,0000,0000,,generally there’ve been\Nseveral studies done that
Dialogue: 0,0:21:01.02,0:21:05.27,Default,,0000,0000,0000,,black, young individuals are\Nstopped more often than whites.
Dialogue: 0,0:21:05.27,0:21:09.80,Default,,0000,0000,0000,,And this is going to\Ncause a collection bias.
Dialogue: 0,0:21:09.80,0:21:14.55,Default,,0000,0000,0000,,It’s gonna be drastically disproportionate\Nto the makeup of the population of cities;
Dialogue: 0,0:21:14.55,0:21:19.44,Default,,0000,0000,0000,,and as more data has been collected on\Nminorities, refugees in poor neighborhoods
Dialogue: 0,0:21:19.44,0:21:23.64,Default,,0000,0000,0000,,it’s gonna feed back in and of course only\Nhave data on those groups and provide
Dialogue: 0,0:21:23.64,0:21:26.41,Default,,0000,0000,0000,,feedback and say:\N“More crime is likely to
Dialogue: 0,0:21:26.41,0:21:27.77,Default,,0000,0000,0000,,happen because that’s where the data
Dialogue: 0,0:21:27.77,0:21:32.25,Default,,0000,0000,0000,,was collected”. So, what’s\Nan acceptable error rate, well,
Dialogue: 0,0:21:32.25,0:21:37.50,Default,,0000,0000,0000,,depends on the burden of proof. Harm\Nis different for an opt-in system.
Dialogue: 0,0:21:37.50,0:21:40.84,Default,,0000,0000,0000,,You know, what’s my harm if I don’t\Nget clearance, or I don’t get the job;
Dialogue: 0,0:21:40.84,0:21:45.16,Default,,0000,0000,0000,,but I’m opting in, I’m asking to\Nbeing considered for employment.
Dialogue: 0,0:21:45.16,0:21:49.08,Default,,0000,0000,0000,,In the US, what’s an error? If you\Nsearch and find nothing, if you think
Dialogue: 0,0:21:49.08,0:21:53.63,Default,,0000,0000,0000,,you have Reasonable Suspicion\Nbased on good faith,
Dialogue: 0,0:21:53.63,0:21:57.06,Default,,0000,0000,0000,,both on the algorithm and what\Nyou witness, the US says that it’s
Dialogue: 0,0:21:57.06,0:22:00.62,Default,,0000,0000,0000,,no 4th Amendment violation,\Neven if nothing has happened.
Dialogue: 0,0:22:00.62,0:22:05.97,Default,,0000,0000,0000,,It’s very low error\Nfalse-positive rate here.
Dialogue: 0,0:22:05.97,0:22:09.14,Default,,0000,0000,0000,,In Big Data, generally, and\Nmachine-learning it’s great!
Dialogue: 0,0:22:09.14,0:22:13.55,Default,,0000,0000,0000,,Like 1% error is fantastic! But that’s\Npretty large for the number of individuals
Dialogue: 0,0:22:13.55,0:22:17.93,Default,,0000,0000,0000,,stopped each day. Or who might\Nbe subject to these algorithms.
Dialogue: 0,0:22:17.93,0:22:21.95,Default,,0000,0000,0000,,Because even though there’re only\N400 individuals on the list in Chicago
Dialogue: 0,0:22:21.95,0:22:25.21,Default,,0000,0000,0000,,those individuals have been\Nlisted basically as targets
Dialogue: 0,0:22:25.21,0:22:28.87,Default,,0000,0000,0000,,by the Chicago Police Department.
Dialogue: 0,0:22:28.87,0:22:33.70,Default,,0000,0000,0000,,Other problems include database errors.\NExclusion of evidence in the US
Dialogue: 0,0:22:33.70,0:22:37.17,Default,,0000,0000,0000,,only happens when there’s gross\Nnegligence or systematic misconduct.
Dialogue: 0,0:22:37.17,0:22:42.15,Default,,0000,0000,0000,,That’s very difficult to prove, especially\Nwhen a lot of people view these algorithms
Dialogue: 0,0:22:42.15,0:22:47.36,Default,,0000,0000,0000,,as a big box. Data goes in,\Npredictions come out, everyone’s happy.
Dialogue: 0,0:22:47.36,0:22:53.10,Default,,0000,0000,0000,,You rely and trust on the\Nquality of IBM, HunchLab etc.
Dialogue: 0,0:22:53.10,0:22:56.73,Default,,0000,0000,0000,,to provide good software.
Dialogue: 0,0:22:56.73,0:23:01.00,Default,,0000,0000,0000,,Finally, some more concerns I have\Ninclude feedback loop auditing
Dialogue: 0,0:23:01.00,0:23:04.81,Default,,0000,0000,0000,,and access to data and algorithms\Nand the prediction thresholds.
Dialogue: 0,0:23:04.81,0:23:09.97,Default,,0000,0000,0000,,How certain must a prediction be\N– before it’s reported to the Police –
Dialogue: 0,0:23:09.97,0:23:13.23,Default,,0000,0000,0000,,that the person might commit a\Ncrime. Or that crime might happen
Dialogue: 0,0:23:13.23,0:23:18.46,Default,,0000,0000,0000,,in the individual area. If Reasonable\NSuspicion is as low as 35%,
Dialogue: 0,0:23:18.46,0:23:23.74,Default,,0000,0000,0000,,and reasonable Suspicion in the US has\Nbeen held at: That guy drives a car
Dialogue: 0,0:23:23.74,0:23:28.35,Default,,0000,0000,0000,,that drug dealers like to drive,\Nand he’s in the DEA database
Dialogue: 0,0:23:28.35,0:23:36.55,Default,,0000,0000,0000,,as a possible drug dealer. That was\Nenough to stop and search him.
Dialogue: 0,0:23:36.55,0:23:40.09,Default,,0000,0000,0000,,So, are there Positives? Well, PredPol,
Dialogue: 0,0:23:40.09,0:23:44.80,Default,,0000,0000,0000,,which is one of the services that\Nprovides Predictive Policing software,
Dialogue: 0,0:23:44.80,0:23:49.65,Default,,0000,0000,0000,,says: “Since these cities have\Nimplemented there’s been dropping crime”.
Dialogue: 0,0:23:49.65,0:23:54.03,Default,,0000,0000,0000,,In L.A. 13% reduction in\Ncrime, in one division.
Dialogue: 0,0:23:54.03,0:23:57.51,Default,,0000,0000,0000,,There was even one day where\Nthey had no crime reported.
Dialogue: 0,0:23:57.51,0:24:04.55,Default,,0000,0000,0000,,Santa Cruz – 25..29% reduction,\N-9% in assaults etc.
Dialogue: 0,0:24:04.55,0:24:10.03,Default,,0000,0000,0000,,One: these are Police departments\Nself-reporting these successes for…
Dialogue: 0,0:24:10.03,0:24:14.67,Default,,0000,0000,0000,,you know, take it for what it is\Nand reiterated by the people
Dialogue: 0,0:24:14.67,0:24:20.51,Default,,0000,0000,0000,,selling the software. But perhaps\Nit is actually reducing crime.
Dialogue: 0,0:24:20.51,0:24:24.39,Default,,0000,0000,0000,,It’s kind of hard to tell because\Nthere’s a feedback loop.
Dialogue: 0,0:24:24.39,0:24:29.20,Default,,0000,0000,0000,,Do we know that crime is really being\Nreduced? Will it affect the data
Dialogue: 0,0:24:29.20,0:24:33.17,Default,,0000,0000,0000,,that is collected in the future? It’s\Nreally hard to know. Because
Dialogue: 0,0:24:33.17,0:24:38.33,Default,,0000,0000,0000,,if you send the Police officers into\Na community it’s more likely
Dialogue: 0,0:24:38.33,0:24:42.58,Default,,0000,0000,0000,,that they’re going to affect that\Ncommunity and that data collection.
Dialogue: 0,0:24:42.58,0:24:46.94,Default,,0000,0000,0000,,Will more crimes happen because they\Nfeel like the Police are harassing them?
Dialogue: 0,0:24:46.94,0:24:52.02,Default,,0000,0000,0000,,It’s very likely and it’s a problem here.
Dialogue: 0,0:24:52.02,0:24:56.93,Default,,0000,0000,0000,,So, some final thoughts. Predictive\NPolicing programs are not going anywhere.
Dialogue: 0,0:24:56.93,0:25:01.43,Default,,0000,0000,0000,,They’re only in their wheelstart.
Dialogue: 0,0:25:01.43,0:25:06.03,Default,,0000,0000,0000,,And I think that more analysis, more\Ntransparency, more access to data
Dialogue: 0,0:25:06.03,0:25:10.56,Default,,0000,0000,0000,,needs to happen around these algorithms.\NThere needs to be regulation.
Dialogue: 0,0:25:10.56,0:25:16.00,Default,,0000,0000,0000,,Currently, a very successful way in which
Dialogue: 0,0:25:16.00,0:25:19.31,Default,,0000,0000,0000,,these companies get data is they\Nbuy from Third Party sources
Dialogue: 0,0:25:19.31,0:25:24.59,Default,,0000,0000,0000,,and then sell it to Police departments. So\Nperhaps PredPol might get information
Dialogue: 0,0:25:24.59,0:25:28.78,Default,,0000,0000,0000,,from Google, Facebook, Social Media\Naccounts; aggregate data themselves,
Dialogue: 0,0:25:28.78,0:25:31.89,Default,,0000,0000,0000,,and then turn around and sell it to\NPolice departments or provide access
Dialogue: 0,0:25:31.89,0:25:36.11,Default,,0000,0000,0000,,to Police departments. And generally, the\NCourts are gonna have to begin to work out
Dialogue: 0,0:25:36.11,0:25:40.21,Default,,0000,0000,0000,,how to handle this type of data.\NThere’s not case law,
Dialogue: 0,0:25:40.21,0:25:45.16,Default,,0000,0000,0000,,at least in the US, that really knows\Nhow to handle predictive algorithms
Dialogue: 0,0:25:45.16,0:25:48.90,Default,,0000,0000,0000,,in determining what the analysis says.\NAnd so there really needs to be
Dialogue: 0,0:25:48.90,0:25:52.60,Default,,0000,0000,0000,,a lot more research and\Nthought put into this.
Dialogue: 0,0:25:52.60,0:25:56.48,Default,,0000,0000,0000,,And one of the big things in order\Nfor this to actually be useful:
Dialogue: 0,0:25:56.48,0:26:01.59,Default,,0000,0000,0000,,if this is a tactic that had been used\Nby Police departments for decades,
Dialogue: 0,0:26:01.59,0:26:04.42,Default,,0000,0000,0000,,we need to eliminate the bias in\Nthe data sets. Because right now
Dialogue: 0,0:26:04.42,0:26:09.09,Default,,0000,0000,0000,,all that it’s doing is facilitating and\Ncontinuing bias, set in the database.
Dialogue: 0,0:26:09.09,0:26:12.61,Default,,0000,0000,0000,,And it’s incredibly difficult.\NIt’s data collected by humans.
Dialogue: 0,0:26:12.61,0:26:17.78,Default,,0000,0000,0000,,And it causes initial selection bias.\NWhich is gonna have to stop
Dialogue: 0,0:26:17.78,0:26:21.38,Default,,0000,0000,0000,,for it to be successful.
Dialogue: 0,0:26:21.38,0:26:25.93,Default,,0000,0000,0000,,And perhaps these systems can cause\Nimplicit bias or confirmation bias,
Dialogue: 0,0:26:25.93,0:26:29.03,Default,,0000,0000,0000,,e.g. Police are going to believe\Nwhat they’ve been told.
Dialogue: 0,0:26:29.03,0:26:33.17,Default,,0000,0000,0000,,So if a Police officer goes\Non duty to an area
Dialogue: 0,0:26:33.17,0:26:36.66,Default,,0000,0000,0000,,and an algorithm says: “You’re\N70% likely to find a burglar
Dialogue: 0,0:26:36.66,0:26:40.84,Default,,0000,0000,0000,,in this area”. Are they gonna find\Na burglar because they’ve been told:
Dialogue: 0,0:26:40.84,0:26:45.93,Default,,0000,0000,0000,,“You might find a burglar”?\NAnd finally the US border.
Dialogue: 0,0:26:45.93,0:26:49.80,Default,,0000,0000,0000,,There is no 4th Amendment\Nprotection at the US border.
Dialogue: 0,0:26:49.80,0:26:53.74,Default,,0000,0000,0000,,It’s an exception to the warrant\Nrequirement. This means
Dialogue: 0,0:26:53.74,0:26:58.74,Default,,0000,0000,0000,,no suspicion is needed to commit\Na search. So this data is gonna go into
Dialogue: 0,0:26:58.74,0:27:03.68,Default,,0000,0000,0000,,a way to examine you when\Nyou cross the border.
Dialogue: 0,0:27:03.68,0:27:09.96,Default,,0000,0000,0000,,And aggregate data can be used to\Nrefuse you entry into the US etc.
Dialogue: 0,0:27:09.96,0:27:13.69,Default,,0000,0000,0000,,And I think that’s pretty much it.\NAnd so a few minutes for questions.
Dialogue: 0,0:27:13.69,0:27:24.49,Default,,0000,0000,0000,,{\i1}applause{\i0}\NThank you!
Dialogue: 0,0:27:24.49,0:27:27.46,Default,,0000,0000,0000,,Herald: Thanks a lot for your talk,\NWhitney. We have about 4 minutes left
Dialogue: 0,0:27:27.46,0:27:31.80,Default,,0000,0000,0000,,for questions. So please line up at\Nthe microphones and remember to
Dialogue: 0,0:27:31.80,0:27:37.74,Default,,0000,0000,0000,,make short and easy questions.
Dialogue: 0,0:27:37.74,0:27:42.06,Default,,0000,0000,0000,,Microphone No.2, please.
Dialogue: 0,0:27:42.06,0:27:53.74,Default,,0000,0000,0000,,Question: Just a comment: if I want\Nto run a crime organization, like,
Dialogue: 0,0:27:53.74,0:27:57.76,Default,,0000,0000,0000,,I would target the PRECOBS\Nhere in Hamburg, maybe.
Dialogue: 0,0:27:57.76,0:28:01.17,Default,,0000,0000,0000,,So I can take the crime to the scenes
Dialogue: 0,0:28:01.17,0:28:05.70,Default,,0000,0000,0000,,where the PRECOBS doesn’t suspect.
Dialogue: 0,0:28:05.70,0:28:08.94,Default,,0000,0000,0000,,Whitney: Possibly. And I think this is\Na big problem in getting availability
Dialogue: 0,0:28:08.94,0:28:13.41,Default,,0000,0000,0000,,of data; in that there’s a good argument\Nfor Police departments to say:
Dialogue: 0,0:28:13.41,0:28:16.59,Default,,0000,0000,0000,,“We don’t want to tell you what\Nour tactics are for Policing,
Dialogue: 0,0:28:16.59,0:28:19.49,Default,,0000,0000,0000,,because it might move crime”.
Dialogue: 0,0:28:19.49,0:28:23.13,Default,,0000,0000,0000,,Herald: Do we have questions from\Nthe internet? Yes, then please,
Dialogue: 0,0:28:23.13,0:28:26.58,Default,,0000,0000,0000,,one question from the internet.
Dialogue: 0,0:28:26.58,0:28:29.77,Default,,0000,0000,0000,,Signal Angel: Is there evidence that data\Nlike the use of encrypted messaging
Dialogue: 0,0:28:29.77,0:28:35.71,Default,,0000,0000,0000,,systems, encrypted emails, VPN, TOR,\Nwith automated request to the ISP,
Dialogue: 0,0:28:35.71,0:28:41.98,Default,,0000,0000,0000,,are used to obtain real names and\Ncollected to contribute to the scoring?
Dialogue: 0,0:28:41.98,0:28:45.58,Default,,0000,0000,0000,,Whitney: I’m not sure if that’s\Nbeing taken into account
Dialogue: 0,0:28:45.58,0:28:49.53,Default,,0000,0000,0000,,by Predictive Policing algorithms,\Nor by the software being used.
Dialogue: 0,0:28:49.53,0:28:55.16,Default,,0000,0000,0000,,I know that Police departments do\Ntake those things into consideration.
Dialogue: 0,0:28:55.16,0:29:00.63,Default,,0000,0000,0000,,And considering that in the US\NTotality of the Circumstances is
Dialogue: 0,0:29:00.63,0:29:04.98,Default,,0000,0000,0000,,how you evaluate suspicion. They are gonna\Ntake all of those things into account
Dialogue: 0,0:29:04.98,0:29:09.15,Default,,0000,0000,0000,,and they actually kind of\Nhave to take into account.
Dialogue: 0,0:29:09.15,0:29:11.83,Default,,0000,0000,0000,,Herald: Okay, microphone No.1, please.
Dialogue: 0,0:29:11.83,0:29:16.79,Default,,0000,0000,0000,,Question: In your example you mentioned\Ndisease tracking, e.g. Google Flu Trends
Dialogue: 0,0:29:16.79,0:29:21.87,Default,,0000,0000,0000,,is a good example of preventive Predictive\NPolicing. Are there any examples
Dialogue: 0,0:29:21.87,0:29:27.63,Default,,0000,0000,0000,,where – instead of increasing Policing\Nin the lives of communities –
Dialogue: 0,0:29:27.63,0:29:34.26,Default,,0000,0000,0000,,where sociologists or social workers\Nare called to use predictive tools,
Dialogue: 0,0:29:34.26,0:29:36.21,Default,,0000,0000,0000,,instead of more criminalization?
Dialogue: 0,0:29:36.21,0:29:41.36,Default,,0000,0000,0000,,Whitney: I’m not aware if that’s…\Nif Police departments are sending
Dialogue: 0,0:29:41.36,0:29:45.25,Default,,0000,0000,0000,,social workers instead of Police officers.\NBut that wouldn’t surprise me because
Dialogue: 0,0:29:45.25,0:29:50.06,Default,,0000,0000,0000,,algorithms are being used to suspect child\Nabuse. And in the US they’re gonna send
Dialogue: 0,0:29:50.06,0:29:53.23,Default,,0000,0000,0000,,a social worker in regard. So I would\Nnot be surprised if that’s also being
Dialogue: 0,0:29:53.23,0:29:56.89,Default,,0000,0000,0000,,considered. Since that’s\Npart of the resources.
Dialogue: 0,0:29:56.89,0:29:59.03,Default,,0000,0000,0000,,Herald: OK, so if you have\Na really short question, then
Dialogue: 0,0:29:59.03,0:30:01.47,Default,,0000,0000,0000,,microphone No.2, please.\NLast question.
Dialogue: 0,0:30:01.47,0:30:08.44,Default,,0000,0000,0000,,Question: Okay, thank you for the\Ntalk. This talk as well as few others
Dialogue: 0,0:30:08.44,0:30:13.71,Default,,0000,0000,0000,,brought the thought in the debate\Nabout the fine-tuning that is required
Dialogue: 0,0:30:13.71,0:30:19.79,Default,,0000,0000,0000,,between false positives and\Npreventing crimes or terror.
Dialogue: 0,0:30:19.79,0:30:24.25,Default,,0000,0000,0000,,Now, it’s a different situation\Nif the Policeman is predicting,
Dialogue: 0,0:30:24.25,0:30:28.35,Default,,0000,0000,0000,,or a system is predicting somebody’s\Nstealing a paper from someone;
Dialogue: 0,0:30:28.35,0:30:32.23,Default,,0000,0000,0000,,or someone is creating a terror attack.
Dialogue: 0,0:30:32.23,0:30:38.03,Default,,0000,0000,0000,,And the justification to prevent it
Dialogue: 0,0:30:38.03,0:30:42.98,Default,,0000,0000,0000,,under the expense of false positive\Nis different in these cases.
Dialogue: 0,0:30:42.98,0:30:49.08,Default,,0000,0000,0000,,How do you make sure that the decision\Nor the fine-tuning is not going to be
Dialogue: 0,0:30:49.08,0:30:53.57,Default,,0000,0000,0000,,deep down in the algorithm\Nand by the programmers,
Dialogue: 0,0:30:53.57,0:30:58.65,Default,,0000,0000,0000,,but rather by the customer\N– the Policemen or the authorities?
Dialogue: 0,0:30:58.65,0:31:02.72,Default,,0000,0000,0000,,Whitney: I can imagine that Police\Nofficers are using common sense in that,
Dialogue: 0,0:31:02.72,0:31:06.22,Default,,0000,0000,0000,,and their knowledge about the situation\Nand even what they’re being told
Dialogue: 0,0:31:06.22,0:31:10.45,Default,,0000,0000,0000,,by the algorithm. You hope\Nthat they’re gonna take…
Dialogue: 0,0:31:10.45,0:31:13.79,Default,,0000,0000,0000,,they probably are gonna take\Nterrorism to a different level
Dialogue: 0,0:31:13.79,0:31:17.26,Default,,0000,0000,0000,,than a common burglary or\Na stealing of a piece of paper
Dialogue: 0,0:31:17.26,0:31:21.76,Default,,0000,0000,0000,,or a non-violent crime.\NAnd that fine-tuning
Dialogue: 0,0:31:21.76,0:31:26.16,Default,,0000,0000,0000,,is probably on a Police department
Dialogue: 0,0:31:26.16,0:31:29.39,Default,,0000,0000,0000,,by Police department basis.
Dialogue: 0,0:31:29.39,0:31:32.09,Default,,0000,0000,0000,,Herald: Thank you! This was Whitney\NMerrill, give a warm round of applause, please!!
Dialogue: 0,0:31:32.09,0:31:40.49,Default,,0000,0000,0000,,Whitney: Thank you!\N{\i1}applause{\i0}
Dialogue: 0,0:31:40.49,0:31:42.51,Default,,0000,0000,0000,,{\i1}postroll music{\i0}
Dialogue: 0,0:31:42.51,0:31:51.50,Default,,0000,0000,0000,,{\i1}Subtitles created by c3subtitles.de\Nin the year 2016. Join and help us!{\i0}