[Script Info] Title: [Events] Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text Dialogue: 0,0:00:01.04,0:00:04.18,Default,,0000,0000,0000,,Hello, I'm Joy, a poet of code, Dialogue: 0,0:00:04.20,0:00:09.19,Default,,0000,0000,0000,,on a mission to stop\Nan unseen force that's rising, Dialogue: 0,0:00:09.22,0:00:12.07,Default,,0000,0000,0000,,a force that I called "the coded gaze," Dialogue: 0,0:00:12.10,0:00:15.40,Default,,0000,0000,0000,,my term for algorithmic bias. Dialogue: 0,0:00:15.43,0:00:19.73,Default,,0000,0000,0000,,Algorithmic bias, like human bias,\Nresults in unfairness. Dialogue: 0,0:00:19.75,0:00:25.78,Default,,0000,0000,0000,,However, algorithms, like viruses,\Ncan spread bias on a massive scale Dialogue: 0,0:00:25.80,0:00:27.38,Default,,0000,0000,0000,,at a rapid pace. Dialogue: 0,0:00:27.94,0:00:32.33,Default,,0000,0000,0000,,Algorithmic bias can also lead\Nto exclusionary experiences Dialogue: 0,0:00:32.35,0:00:34.48,Default,,0000,0000,0000,,and discriminatory practices. Dialogue: 0,0:00:34.51,0:00:36.57,Default,,0000,0000,0000,,Let me show you what I mean. Dialogue: 0,0:00:36.98,0:00:39.42,Default,,0000,0000,0000,,(Video) Joy Boulamwini: Hi, camera.\NI've got a face. Dialogue: 0,0:00:40.16,0:00:42.03,Default,,0000,0000,0000,,Can you see my face? Dialogue: 0,0:00:42.05,0:00:43.68,Default,,0000,0000,0000,,No-glasses face? Dialogue: 0,0:00:43.70,0:00:45.92,Default,,0000,0000,0000,,You can see her face. Dialogue: 0,0:00:46.24,0:00:48.48,Default,,0000,0000,0000,,What about my face? Dialogue: 0,0:00:51.89,0:00:55.64,Default,,0000,0000,0000,,I've got a mask. Can you see my mask? Dialogue: 0,0:00:56.47,0:00:58.84,Default,,0000,0000,0000,,Joy Boulamwini: So how did this happen? Dialogue: 0,0:00:58.86,0:01:02.00,Default,,0000,0000,0000,,Why am I sitting in front of a computer Dialogue: 0,0:01:02.03,0:01:03.45,Default,,0000,0000,0000,,in a white mask, Dialogue: 0,0:01:03.48,0:01:07.13,Default,,0000,0000,0000,,trying to be detected by a cheap webcam? Dialogue: 0,0:01:07.15,0:01:09.44,Default,,0000,0000,0000,,Well, when I'm not fighting the coded gaze Dialogue: 0,0:01:09.46,0:01:10.98,Default,,0000,0000,0000,,as a poet of code, Dialogue: 0,0:01:11.01,0:01:14.28,Default,,0000,0000,0000,,I'm a graduate student\Nat the MIT Media Lab, Dialogue: 0,0:01:14.30,0:01:19.22,Default,,0000,0000,0000,,and there I have the opportunity to work\Non all sorts of whimsical projects, Dialogue: 0,0:01:19.25,0:01:21.27,Default,,0000,0000,0000,,including the Aspire Mirror, Dialogue: 0,0:01:21.30,0:01:26.43,Default,,0000,0000,0000,,a project I did so I could project\Ndigital masks onto my reflection. Dialogue: 0,0:01:26.46,0:01:28.80,Default,,0000,0000,0000,,So in the morning, if I wanted\Nto feel powerful, Dialogue: 0,0:01:28.83,0:01:30.26,Default,,0000,0000,0000,,I could put on a lion. Dialogue: 0,0:01:30.29,0:01:33.78,Default,,0000,0000,0000,,If I wanted to be uplifted,\NI might have a quote. Dialogue: 0,0:01:33.81,0:01:36.80,Default,,0000,0000,0000,,So I used generic\Nfacial recognition software Dialogue: 0,0:01:36.82,0:01:38.17,Default,,0000,0000,0000,,to build the system, Dialogue: 0,0:01:38.20,0:01:43.30,Default,,0000,0000,0000,,but found it was really hard to test it\Nunless I wore a white mask. Dialogue: 0,0:01:44.28,0:01:48.63,Default,,0000,0000,0000,,Unfortunately, I've run\Ninto this issue before. Dialogue: 0,0:01:48.65,0:01:52.96,Default,,0000,0000,0000,,When I was an undergraduate\Nat Georgia Tech studying computer science, Dialogue: 0,0:01:52.98,0:01:55.03,Default,,0000,0000,0000,,I used to work on social robots, Dialogue: 0,0:01:55.06,0:01:58.84,Default,,0000,0000,0000,,and one of my tasks was to get a robot\Nto play peek-a-boo, Dialogue: 0,0:01:58.86,0:02:00.54,Default,,0000,0000,0000,,a simple turn-taking game Dialogue: 0,0:02:00.57,0:02:04.89,Default,,0000,0000,0000,,where partners cover their face\Nand then uncover it saying, "Peek-a-boo!" Dialogue: 0,0:02:04.91,0:02:09.34,Default,,0000,0000,0000,,The problem is, peek-a-boo\Ndoesn't really work if I can't see you, Dialogue: 0,0:02:09.36,0:02:11.86,Default,,0000,0000,0000,,and my robot couldn't see me. Dialogue: 0,0:02:11.89,0:02:15.84,Default,,0000,0000,0000,,But I borrowed my roommate's face\Nto get the project done, Dialogue: 0,0:02:15.86,0:02:17.24,Default,,0000,0000,0000,,submitted the assignment, Dialogue: 0,0:02:17.26,0:02:21.02,Default,,0000,0000,0000,,and figured, you know what,\Nsomebody else will solve this problem. Dialogue: 0,0:02:21.67,0:02:23.67,Default,,0000,0000,0000,,Not too long after, Dialogue: 0,0:02:23.70,0:02:27.86,Default,,0000,0000,0000,,I was in Hong Kong\Nfor an entrepreneurship competition. Dialogue: 0,0:02:28.34,0:02:31.03,Default,,0000,0000,0000,,The organizers decided\Nto take participants Dialogue: 0,0:02:31.06,0:02:33.43,Default,,0000,0000,0000,,on a tour of local start-ups. Dialogue: 0,0:02:33.45,0:02:36.17,Default,,0000,0000,0000,,One of the start-ups had a social robot, Dialogue: 0,0:02:36.19,0:02:38.10,Default,,0000,0000,0000,,and they decided to do a demo. Dialogue: 0,0:02:38.13,0:02:41.11,Default,,0000,0000,0000,,The demo worked on everybody\Nuntil it got to me, Dialogue: 0,0:02:41.13,0:02:43.06,Default,,0000,0000,0000,,and you can probably guess it. Dialogue: 0,0:02:43.08,0:02:46.04,Default,,0000,0000,0000,,It couldn't detect my face. Dialogue: 0,0:02:46.07,0:02:48.58,Default,,0000,0000,0000,,I asked the developers what was going on, Dialogue: 0,0:02:48.60,0:02:54.14,Default,,0000,0000,0000,,and it turned out we had used the same\Ngeneric facial recognition software. Dialogue: 0,0:02:54.16,0:02:55.81,Default,,0000,0000,0000,,Halfway around the world, Dialogue: 0,0:02:55.83,0:02:59.69,Default,,0000,0000,0000,,I learned that algorithmic bias\Ncan travel as quickly Dialogue: 0,0:02:59.71,0:03:02.88,Default,,0000,0000,0000,,as it takes to download\Nsome files off of the internet. Dialogue: 0,0:03:03.74,0:03:06.82,Default,,0000,0000,0000,,So what's going on?\NWhy isn't my face being detected? Dialogue: 0,0:03:06.84,0:03:10.20,Default,,0000,0000,0000,,Well, we have to look at\Nhow we give machines sight. Dialogue: 0,0:03:10.22,0:03:13.63,Default,,0000,0000,0000,,Computer vision uses\Nmachine learning techniques Dialogue: 0,0:03:13.66,0:03:15.54,Default,,0000,0000,0000,,to do facial recognition. Dialogue: 0,0:03:15.56,0:03:19.46,Default,,0000,0000,0000,,So how this works is, you create\Na training set with examples of faces. Dialogue: 0,0:03:19.48,0:03:22.30,Default,,0000,0000,0000,,This is a face. This is a face.\NThis is not a face. Dialogue: 0,0:03:22.32,0:03:26.84,Default,,0000,0000,0000,,And over time, you can teach a computer\Nhow to recognize other faces. Dialogue: 0,0:03:26.87,0:03:30.86,Default,,0000,0000,0000,,However, if the training sets\Naren't really that diverse, Dialogue: 0,0:03:30.88,0:03:34.23,Default,,0000,0000,0000,,any face that deviates too much\Nfrom the established norm Dialogue: 0,0:03:34.25,0:03:35.90,Default,,0000,0000,0000,,will be harder to detect, Dialogue: 0,0:03:35.93,0:03:37.89,Default,,0000,0000,0000,,which is what was happening to me. Dialogue: 0,0:03:37.91,0:03:40.30,Default,,0000,0000,0000,,But don't worry -- there's some good news. Dialogue: 0,0:03:40.32,0:03:43.09,Default,,0000,0000,0000,,Training sets don't just\Nmaterialize out of nowhere. Dialogue: 0,0:03:43.12,0:03:44.90,Default,,0000,0000,0000,,We actually can create them. Dialogue: 0,0:03:44.93,0:03:49.10,Default,,0000,0000,0000,,So there's an opportunity to create\Nfull-spectrum training sets Dialogue: 0,0:03:49.13,0:03:52.95,Default,,0000,0000,0000,,that reflect a richer\Nportrait of humanity. Dialogue: 0,0:03:52.98,0:03:55.20,Default,,0000,0000,0000,,Now you've seen in my examples Dialogue: 0,0:03:55.22,0:03:56.99,Default,,0000,0000,0000,,how social robots Dialogue: 0,0:03:57.01,0:04:01.62,Default,,0000,0000,0000,,was how I found out about exclusion\Nwith algorithmic bias. Dialogue: 0,0:04:01.65,0:04:06.46,Default,,0000,0000,0000,,But algorithmic bias can also lead\Nto discriminatory practices. Dialogue: 0,0:04:07.44,0:04:08.89,Default,,0000,0000,0000,,Across the US, Dialogue: 0,0:04:08.91,0:04:13.11,Default,,0000,0000,0000,,police departments are starting to use\Nfacial recognition software Dialogue: 0,0:04:13.14,0:04:15.60,Default,,0000,0000,0000,,in their crime-fighting arsenal. Dialogue: 0,0:04:15.62,0:04:17.63,Default,,0000,0000,0000,,Georgetown Law published a report Dialogue: 0,0:04:17.66,0:04:24.42,Default,,0000,0000,0000,,showing that one in two adults\Nin the US -- that's 117 million people -- Dialogue: 0,0:04:24.44,0:04:27.98,Default,,0000,0000,0000,,have their faces\Nin facial recognition networks. Dialogue: 0,0:04:28.00,0:04:32.55,Default,,0000,0000,0000,,Police departments can currently look\Nat these networks unregulated, Dialogue: 0,0:04:32.58,0:04:36.86,Default,,0000,0000,0000,,using algorithms that have not\Nbeen audited for accuracy. Dialogue: 0,0:04:36.89,0:04:40.75,Default,,0000,0000,0000,,Yet we know facial recognition\Nis not fail proof, Dialogue: 0,0:04:40.78,0:04:44.95,Default,,0000,0000,0000,,and labeling faces consistently\Nremains a challenge. Dialogue: 0,0:04:44.98,0:04:46.74,Default,,0000,0000,0000,,You might have seen this on Facebook. Dialogue: 0,0:04:46.76,0:04:49.75,Default,,0000,0000,0000,,My friends and I laugh all the time\Nwhen we see other people Dialogue: 0,0:04:49.78,0:04:52.23,Default,,0000,0000,0000,,mislabeled in our photos. Dialogue: 0,0:04:52.26,0:04:57.85,Default,,0000,0000,0000,,But misidentifying a suspected criminal\Nis no laughing matter, Dialogue: 0,0:04:57.87,0:05:00.70,Default,,0000,0000,0000,,nor is breaching civil liberties. Dialogue: 0,0:05:00.72,0:05:03.93,Default,,0000,0000,0000,,Machine learning is being used\Nfor facial recognition, Dialogue: 0,0:05:03.95,0:05:08.46,Default,,0000,0000,0000,,but it's also extending beyond the realm\Nof computer vision. Dialogue: 0,0:05:09.27,0:05:13.28,Default,,0000,0000,0000,,In her book, "Weapons\Nof Math Destruction," Dialogue: 0,0:05:13.31,0:05:19.99,Default,,0000,0000,0000,,data scientist Cathy O'Neil\Ntalks about the rising new WMDs -- Dialogue: 0,0:05:20.01,0:05:24.36,Default,,0000,0000,0000,,widespread, mysterious\Nand destructive algorithms Dialogue: 0,0:05:24.39,0:05:27.35,Default,,0000,0000,0000,,that are increasingly being used\Nto make decisions Dialogue: 0,0:05:27.38,0:05:30.55,Default,,0000,0000,0000,,that impact more aspects of our lives. Dialogue: 0,0:05:30.58,0:05:32.45,Default,,0000,0000,0000,,So who gets hired or fired? Dialogue: 0,0:05:32.47,0:05:34.58,Default,,0000,0000,0000,,Do you get that loan?\NDo you get insurance? Dialogue: 0,0:05:34.61,0:05:38.11,Default,,0000,0000,0000,,Are you admitted into the college\Nyou wanted to get into? Dialogue: 0,0:05:38.13,0:05:41.64,Default,,0000,0000,0000,,Do you and I pay the same price\Nfor the same product Dialogue: 0,0:05:41.67,0:05:44.11,Default,,0000,0000,0000,,purchased on the same platform? Dialogue: 0,0:05:44.13,0:05:47.89,Default,,0000,0000,0000,,Law enforcement is also starting\Nto use machine learning Dialogue: 0,0:05:47.92,0:05:50.20,Default,,0000,0000,0000,,for predictive policing. Dialogue: 0,0:05:50.23,0:05:53.72,Default,,0000,0000,0000,,Some judges use machine-generated\Nrisk scores to determine Dialogue: 0,0:05:53.75,0:05:58.15,Default,,0000,0000,0000,,how long an individual\Nis going to spend in prison. Dialogue: 0,0:05:58.17,0:06:00.63,Default,,0000,0000,0000,,So we really have to think\Nabout these decisions. Dialogue: 0,0:06:00.65,0:06:01.83,Default,,0000,0000,0000,,Are they fair? Dialogue: 0,0:06:01.86,0:06:04.75,Default,,0000,0000,0000,,And we've seen that algorithmic bias Dialogue: 0,0:06:04.77,0:06:08.14,Default,,0000,0000,0000,,doesn't necessarily always\Nlead to fair outcomes. Dialogue: 0,0:06:08.17,0:06:10.13,Default,,0000,0000,0000,,So what can we do about it? Dialogue: 0,0:06:10.16,0:06:13.84,Default,,0000,0000,0000,,Well, we can start thinking about\Nhow we create more inclusive code Dialogue: 0,0:06:13.86,0:06:16.85,Default,,0000,0000,0000,,and employ inclusive coding practices. Dialogue: 0,0:06:16.88,0:06:19.18,Default,,0000,0000,0000,,It really starts with people. Dialogue: 0,0:06:19.71,0:06:21.67,Default,,0000,0000,0000,,So who codes matters. Dialogue: 0,0:06:21.69,0:06:25.81,Default,,0000,0000,0000,,Are we creating full-spectrum teams\Nwith diverse individuals Dialogue: 0,0:06:25.84,0:06:28.25,Default,,0000,0000,0000,,who can check each other's blind spots? Dialogue: 0,0:06:28.27,0:06:31.82,Default,,0000,0000,0000,,On the technical side,\Nhow we code matters. Dialogue: 0,0:06:31.84,0:06:35.49,Default,,0000,0000,0000,,Are we factoring in fairness\Nas we're developing systems? Dialogue: 0,0:06:35.52,0:06:38.43,Default,,0000,0000,0000,,And finally, why we code matters. Dialogue: 0,0:06:38.78,0:06:43.87,Default,,0000,0000,0000,,We've used tools of computational creation\Nto unlock immense wealth. Dialogue: 0,0:06:43.89,0:06:48.34,Default,,0000,0000,0000,,We now have the opportunity\Nto unlock even greater equality Dialogue: 0,0:06:48.36,0:06:51.29,Default,,0000,0000,0000,,if we make social change a priority Dialogue: 0,0:06:51.32,0:06:53.49,Default,,0000,0000,0000,,and not an afterthought. Dialogue: 0,0:06:54.01,0:06:58.53,Default,,0000,0000,0000,,And so these are the three tenets\Nthat will make up the "incoding" movement. Dialogue: 0,0:06:58.55,0:07:00.21,Default,,0000,0000,0000,,Who codes matters, Dialogue: 0,0:07:00.23,0:07:01.77,Default,,0000,0000,0000,,how we code matters Dialogue: 0,0:07:01.80,0:07:03.82,Default,,0000,0000,0000,,and why we code matters. Dialogue: 0,0:07:03.84,0:07:06.94,Default,,0000,0000,0000,,So to go towards incoding,\Nwe can start thinking about Dialogue: 0,0:07:06.97,0:07:10.13,Default,,0000,0000,0000,,building platforms that can identify bias Dialogue: 0,0:07:10.16,0:07:13.23,Default,,0000,0000,0000,,by collecting people's experiences\Nlike the ones I shared, Dialogue: 0,0:07:13.26,0:07:16.33,Default,,0000,0000,0000,,but also auditing existing software. Dialogue: 0,0:07:16.35,0:07:20.12,Default,,0000,0000,0000,,We can also start to create\Nmore inclusive training sets. Dialogue: 0,0:07:20.14,0:07:22.94,Default,,0000,0000,0000,,Imagine a "Selfies for Inclusion" campaign Dialogue: 0,0:07:22.97,0:07:26.62,Default,,0000,0000,0000,,where you and I can help\Ndevelopers test and create Dialogue: 0,0:07:26.65,0:07:28.74,Default,,0000,0000,0000,,more inclusive training sets. Dialogue: 0,0:07:29.30,0:07:32.13,Default,,0000,0000,0000,,And we can also start thinking\Nmore conscientiously Dialogue: 0,0:07:32.15,0:07:37.54,Default,,0000,0000,0000,,about the social impact\Nof the technology that we're developing. Dialogue: 0,0:07:37.57,0:07:39.96,Default,,0000,0000,0000,,To get the incoding movement started, Dialogue: 0,0:07:39.99,0:07:42.83,Default,,0000,0000,0000,,I've launched the Algorithmic\NJustice League, Dialogue: 0,0:07:42.86,0:07:48.73,Default,,0000,0000,0000,,where anyone who cares about fairness\Ncan help fight the coded gaze. Dialogue: 0,0:07:48.75,0:07:52.05,Default,,0000,0000,0000,,On codedgaze.com, you can report bias, Dialogue: 0,0:07:52.07,0:07:54.52,Default,,0000,0000,0000,,request audits, become a tester Dialogue: 0,0:07:54.54,0:07:57.31,Default,,0000,0000,0000,,and join the ongoing conversation, Dialogue: 0,0:07:57.34,0:07:59.62,Default,,0000,0000,0000,,#codedgaze. Dialogue: 0,0:08:00.74,0:08:03.23,Default,,0000,0000,0000,,So I invite you to join me Dialogue: 0,0:08:03.25,0:08:06.97,Default,,0000,0000,0000,,in creating a world where technology\Nworks for all of us, Dialogue: 0,0:08:06.100,0:08:08.89,Default,,0000,0000,0000,,not just some of us, Dialogue: 0,0:08:08.92,0:08:13.50,Default,,0000,0000,0000,,a world where we value inclusion\Nand center social change. Dialogue: 0,0:08:13.53,0:08:14.70,Default,,0000,0000,0000,,Thank you. Dialogue: 0,0:08:14.73,0:08:18.100,Default,,0000,0000,0000,,(Applause) Dialogue: 0,0:08:20.87,0:08:23.73,Default,,0000,0000,0000,,But I have one question: Dialogue: 0,0:08:23.75,0:08:25.81,Default,,0000,0000,0000,,Will you join me in the fight? Dialogue: 0,0:08:25.83,0:08:27.12,Default,,0000,0000,0000,,(Laughter) Dialogue: 0,0:08:27.14,0:08:30.83,Default,,0000,0000,0000,,(Applause)