Frances Haugen: 'Facebook keeps lying to the public' | AFP — Transcript

Full transcript

1. 0:00Facebook has established a  history of,
2. 0:02and I hate to be so damning,
3. 0:05but they keep lying to the public.
4. 0:13For every single whistleblower, whistleblowing is
5. 0:15never plan A or B or even like C or D. It's like E, F or G, because you know
6. 0:24whistleblowing is scary.
7. 0:26Like no one wants to  blow the whistle,
8. 0:29it's just at some point you start living with a secret
9. 0:32that you think could impact the lives of millions of people.
10. 0:35And there's like at some point you can't live with that pain anymore
11. 0:39and so you have to do something.
12. 0:41Journalist: But did it happen very soon after  you entered the company or did it take months to
13. 0:46slowly realize what was going on?
14. 0:49How would you describe it and was there some moment
15. 0:51where you decided  yeah it cannot continue like that.
16. 0:56I was very shocked at how bad it  was almost immediately.
17. 1:00I came in, you know, having seen the harms of misinformation in the United States
18. 1:04and I had no idea how much worse it was outside the United States.
19. 1:08I'm an American, I live in a bubble.
20. 1:15But I didn't really lose faith in Facebook until they dissolved civic integrity.
21. 1:22So there used to be a part of the company that was focused on making sure Facebook
22. 1:25was a constructive force in the world, in society.
23. 1:29You know a positive force in elections, that kind of thing.
24. 1:32And they dissolved it right after the US 2020 election
25. 1:36and I don't think I'm the only one  who kind of lost faith at that point.
26. 1:40I didn't start collecting documents right away,
27. 1:44I had to figure out how to do things, because I had to be careful.
28. 1:47I had to think for a long time.
29. 1:50But other people also left in high numbers,
30. 1:53like when I left, the pod that I was part of was probably 75% people from civic integrity
31. 1:58-- this is like seven or eight product managers, program managers --
32. 2:02and all of us left for either other places in the company
33. 2:05or left the company  entirely over a six-week span.
34. 2:09And so you know I don't think I was the only one who was discouraged
35. 2:13by civic integrity being dissolved.
36. 2:16Journalist: Is the fact that those conversations are happening
37. 2:18enough to justify the risks that you took?
38. 2:21Or do you feel like you're  not going to rest until there is legal change
39. 2:24and then that will be enough or is  it something else that you're looking for?
40. 2:28During my time at Facebook I became aware that…
41. 2:31So, I haven't worked at one social media company, I've worked at four social media companies.
42. 2:34And I have specifically worked as an algorithmic product person
43. 2:38at multiple companies and what  I saw at Facebook was substantially worse
44. 2:43than I had seen at other places.
45. 2:45And because Facebook  bought the right to be the internet
46. 2:50in some of the most fragile places in the world,
47. 2:52so they subsidized your internet if you used Facebook,
48. 2:55but if you wanted to use the open and free internet you had to pay
49. 2:58and so now for the majority of languages in the world 80 to 90% of the content is on Facebook.
50. 3:06When Facebook did that they took on a higher level responsibility,
51. 3:09because there isn't an alternative now
52. 3:12and what I saw was, I think there's a million,
53. 3:15maybe millions, maybe tens of millions of lives on the line,
54. 3:18because of Facebook's tendency to give the most reach to the most extreme content.
55. 3:24And so compared to any risk I could face
56. 3:27that's just not worth a  million lives, it's not worth 10 million lives
57. 3:30and so yeah I want to keep pushing.
58. 3:33I want to start a youth movement, I want to give, I want people not to feel powerless.
59. 3:40I want young people not to feel powerless with regard to big platforms.
60. 3:44There's lots of things that they could do.
61. 3:47I want to work on how do we teach about social networks,
62. 3:51so right now if you  want to take a lab if you want to be a chemist
63. 3:54there are lots of lab classes you can take in college
64. 3:57where you blow stuff up you breathe and stuff you shouldn't breathe in,
65. 4:01but if you  want to be an algorithmic engineer, you want to be a data scientist
66. 4:04there aren't really lab classes.
67. 4:06And if we had simulated social networks, both academics could study them and be like:
68. 4:11“Oh interesting this is how we could disrupt the dynamics of the bad things on the network.”
69. 4:14But it also means that we could teach lab classes
70. 4:17where you know 19 year olds could come in and really analyze experiments, run experiments.
71. 4:23And I think that would change  people's perspectives
72. 4:26and it would give us a whole new generation of security professionals.
73. 4:30Because right now the only way you really can get this job
74. 4:33is by being trained inside one of these companies.
75. 4:36And so I think all of us will be safer if we have a broader ecosystem
76. 4:40of like a public muscle for accountability.
77. 4:43Journalist: There's a lot of elections coming  up around the world next year.
78. 4:46In South Korea, in France, I think there are more elections in the US coming up.
79. 4:51What kind of risks [are we facing with] the way that social media
80. 4:54and specifically Facebook the way that it operates right now?
81. 4:58I'll give you a really concrete example.
82. 4:59This is probably one of the things I'm most concerned on.
83. 5:02So we talked about the idea of  engagement-based ranking.
84. 5:05So Mark Zuckerberg in 2018 came out and said
85. 5:07engagement-based  ranking,
86. 5:09that's where you prioritize what content
87. 5:12goes in your news feed based on its ability to  provoke a reaction from you.
88. 5:16You know a comment, get a reshare.
89. 5:18He said engagement-based  ranking is dangerous because people are drawn
90. 5:24to interact with extreme content
91. 5:26even though when we asked them later: "did you like this content?" they say no.
92. 5:30And he said but don't worry, AI will save us.
93. 5:35The problem with a strategy like that
94. 5:36is engagement-based ranking has problems all over the product,
95. 5:40and one of the places where it is most acute for democracy is
96. 5:43when you run an ad, that same engagement-based ranking
97. 5:49is used to price the ad, so an ad that provokes  more comments, more reshares, more clicks,
98. 5:55is a cheaper ad than an ad that doesn't do that.
99. 6:00And as a result you end up subsidizing hate and division.
100. 6:03Because an ad that is hateful or divisive  is going to be five to ten times cheaper
101. 6:09than an ad that is compassionate or empathetic.
102. 6:12Because it's hard to provoke a reaction for empathy.
103. 6:15And so I'm advocating for flat rates on political  ads,
104. 6:18like there should be some cost per impression
105. 6:20and that's just what it is.
106. 6:22Because otherwise we end up giving a subsidy to people
107. 6:26who want to tear our society apart
108. 6:29and so that's an example of something where like that's going to unquestionably play a role
109. 6:32in both the French  election, the South Korean election.
110. 6:35Because if people know they  get a discount for going dark,
111. 6:39that's a problem.
112. 6:40Because we give five to ten times more reach effectively per dollar
113. 6:44for things that are going  to pull us apart, instead of pull us together.
114. 6:48The second question is around the weaponization of the platform.
115. 6:53So the number of threat intelligence people
116. 6:56that they have who are looking for networks  of influence operations is very small.
117. 7:01So there were only 17 people on that team when I left
118. 7:06and Facebook claims they have 200 people working on this.
119. 7:10Maybe, who knows? I want them  to articulate exactly what do those people do.
120. 7:15Because when I worked at Google they published
121. 7:18how many women they had in tech or in their tech staff
122. 7:21and they included secretaries for engineers
123. 7:24in their count of women who were technical staff.
124. 7:27So it's one of these things where you have to say like:
125. 7:28“what's the job of those 200 people, what are they actually doing?”
126. 7:31Because Facebook will  just dance with the data otherwise.
127. 7:34And so what that means is when you only have 17 people,
128. 7:37like when I worked in counter espionage we had like maybe six people maybe seven people
129. 7:42and at any given time we could only work on a third of the cases we had
130. 7:46because, you know, we just didn't  have staff and so we never built a detector
131. 7:52to proactively find more espionage
132. 7:54because we already couldn't deal with the cases that we had.
133. 7:57Journalist: What is the counterespionage team?
134. 8:02So there are nation states that use Facebook, but there's also private firms,
135. 8:07that use Facebook to do surveillance,
136. 8:10to compromise individual users.
137. 8:12For example a nation state might send spyware that will compromise someone's phone.
138. 8:17Because if you compromise a phone
139. 8:19you can compromise a community because you've got their address book.
140. 8:22Now you know a bunch of people,
141. 8:23you can get their pictures which can have geolocated information usually.
142. 8:27So now you know where do they congregate, what are they interested in.
143. 8:30Yeah it's very problematic.
144. 8:33They might be able to spy on their phone calls at that point.
145. 8:35And so we would monitor for teams that  would do these things
146. 8:39and sometimes they would be against other countries' militaries,
147. 8:42you know, they'd pretend to be a pretty girl and try to make friends with someone
148. 8:45and send them spyware.
149. 8:47And so it's one of these things where that's what our team did, it was look for those.
150. 8:53But in the case of information operations
151. 8:56they're looking for coordinated efforts
152. 8:57to send misinformation, disinformation sometimes to targeted populations.
153. 9:02So for example the team I worked on early in  my time at Facebook was called
154. 9:05“narrow cast misinformation.”
155. 9:08So there's viral misinformation  that goes out to lots of people
156. 9:11and there's narrow cast misinformation that goes to the targeted population.
157. 9:15And that project was started because Russia was targeting police officers,
158. 9:20African-American activists, environmentalists, gay activists
159. 9:24and that's because all those groups  are leaders in their communities respectively.
160. 9:28Also think about what that means.
161. 9:31Targeted misinformation to police officers like people with guns.
162. 9:34That's a real thing, they also target militaries.
163. 9:37So when you under fund people working on information operations, influence operations,
164. 9:45that has serious national security implications.
165. 9:47And so I'm a big proponent  that if Facebook wants to stay behind the wall
166. 9:52and do all of it themselves,
167. 9:54they should have to disclose which teams exist for threat intelligence
168. 9:59and how many people work on them and what their jobs are.
169. 10:02Because right now we know there's a counterterrorism team
170. 10:06and we know there's a counter-information operations team
171. 10:09and a counter-espionage team.
172. 10:11There's also human trafficking or human exploitation more generally.
173. 10:15But we don't know how many  people work on those teams
174. 10:18and when you only have maybe six or seven  people for espionage all around the world,
175. 10:23I don't think that's enough.
176. 10:25Because think about Russia and China, they have 10,000 people each
177. 10:29doing misinformation and espionage.
178. 10:33I don't know, it's a problem we need more transparency.
179. 10:37Journalist: That point that you just made some  people have taken a little bit further and said
180. 10:41well maybe we should be regulating algorithms  in some way. How do you feel about that?
181. 10:47So regulating algorithms. We can't think of  algorithms in isolation.
182. 10:53So algorithms work in concert with product changes.
183. 10:57An example: Facebook did a product change
184. 11:00where instead of having to re-share over and over again to different groups
185. 11:04they made a picker so you could re-share it up to 10 groups at a time.
186. 11:08And when they did it they never thought that it might increase misinformation.
187. 11:13But many months down the  road there was a surge of misinformation
188. 11:17and they traced it back to [the fact that] they had made it so much easier
189. 11:20to reshare into these large groups.
190. 11:24I think if we just come in and say:  “we are going to regulate algorithms”
191. 11:28that is so amorphous and I think it's  more powerful to say:
192. 11:32“Hey, Facebook you have a lot more transparency than we do,
193. 11:34you need to let us know what you're worrying about.”
194. 11:37But the other reality is we can learn a lot from  people's experiences of these products.
195. 11:42So one of my favorite misinformation researchers
196. 11:44is actually an anthropologist, she's not a data scientist
197. 11:48and the way she learns about new misinformation  techniques
198. 11:50is she literally sits down with young people with old people
199. 11:54and she says: “Show me  your news feed.
200. 11:56Can you explain it to me? What do you see?”
201. 11:57And so there's a lot that we can do to  understand algorithms
202. 12:01just from talking to people.
203. 12:04Journalist: The mechanism you described sounds  really good. Do you think it should be up to
204. 12:09companies to be like: yes, okay we will bend to  public pressure and pressure from experts and
205. 12:13do this, or should we be legally forcing companies  like Facebook to introduce this kind of mechanism?
206. 12:18I think we need to legally require it.
207. 12:19Because companies like Facebook, Facebook specifically,
208. 12:21I don't  know about other companies.
209. 12:22Facebook has established a history of -- I hate to like be so damning --
210. 12:28but they keep lying to the public.
211. 12:30Omissions and lies seem to be the pattern at Facebook,
212. 12:33and I think unless there is a legally mandated process
213. 12:36I don't think we can trust what the results of it will be.
214. 12:39Journalist: Here in Europe it's generally seen as Europe being fairly well ahead
215. 12:43in terms of thinking about how to regulate AI,
216. 12:45thinking about how to regulate tech in general,
217. 12:48but it also feels like as soon as we've  worked out how to regulate something in Europe
218. 12:52the situation has already moved on. Is that going to happen again?
219. 12:55So part of why I'm against the government coming in and saying:
220. 12:58“This is what you can do, this is what you can't do,” is that is exactly what happens.
221. 13:02Like they put a fence in and tech walks around the fence.
222. 13:05When you have a system that repeats every six months,
223. 13:08where you have to articulate the harms,
224. 13:09everyone else gets to articulate the harms and you have to say how you're gonna fix them...
225. 13:14Let's imagine the metaverse for just one second.
226. 13:23So one of the things I'm most scared about the metaverse
227. 13:26is when you put on your VR headset,
228. 13:29your avatar is a little more handsome, a  little more pretty than you are, you know,
229. 13:34better haircut, better clothes, nicer apartment.  So let's say you get to the end of your evening,
230. 13:39you come home from work, you put it on. Time to  brush your teeth and you look in the mirror and
231. 13:44you're just not as pretty or as handsome as you  were in VR and you know
232. 13:48maybe a little bit more weight, less muscle, your clothes are definitely not as nice.
233. 13:54What does that do to a person?
234. 13:55If that's what happens every night
235. 13:58and I think it's one of those things where like right now
236. 14:00I have not heard Facebook articulate  how they're gonna deal with that harm.
237. 14:03But if they had to do a risk assessment three months from now
238. 14:06because now they have this new product
239. 14:08and they had to maybe articulate it… But maybe they didn't even articulate it,
240. 14:10because they're too techno-optimistic, but  if someone else said hey what are you going
241. 14:14to do about the fact that people have massive  incentives not to take off these VR headsets?
242. 14:19What are you going to do about people hating themselves more because you're providing
243. 14:24a virtual version of them as nicer?
244. 14:26And I guarantee you that's not on  Facebook's radar right now.
245. 14:30But if they had to now sit down with a team and say: “Okay, what would we do about this?”
246. 14:35That would cut the current cycle.
247. 14:37So the current thing that happens is tech  platforms do these things and five years down the road
248. 14:41they get so bad that we can't ignore them anymore.
249. 14:43That we get enough anecdotal evidence in society that we can put pressures on the company.
250. 14:48But if you have a system like this: company, community, accountability, you cut that cycle
251. 14:53and we could have that conversation three months from now not five years from now.
252. 14:56And I think that's how we stay in lockstep with technology.
253. 14:59Journalist: So you want Facebook to engage right now on
254. 15:01a conversation on the metaverse  and what effect it could have on the people?
255. 15:06Yeah they should. They're about to  invest 10,000 engineers in this.
256. 15:10Isn't that a conversation we should have now?
257. 15:13I also think the fact that they can find 10,000 engineers to invest in the metaverse
258. 15:18and they can't find three thousand more engineers to focus on safety.
259. 15:22I think that's also  unacceptable but that's a different issue.
260. 15:27Another issue with the metaverse that  I'm really worried about
261. 15:29-- I would love to get that articulated probably by the community --
262. 15:32is so to do metaverse you have to put more microphones,
263. 15:36more sensors from a company that has shown they lie
264. 15:39whenever it's in their benefit, in your home.
265. 15:42And if you choose to not have them in your home,
266. 15:44I hope the company you work for doesn't choose for you.
267. 15:48Because let's say you work from home and one day your company goes:
268. 15:51“Hey, we're a metaverse company now.”
269. 15:54You could be forced to choose between your livelihood
270. 15:57and having a whole bunch of sensors from Facebook in your home.
271. 16:02Shouldn't we talk about that now before it becomes a giant product?
272. 16:06And if we had a system like this, so: company, community, accountability,
273. 16:09we would have this conversation now.
274. 16:12But if we don't have a framework like this, Facebook's not going to have that conversation.
275. 16:17Journalist: Knowing what we all know now as  a result of these documents,
276. 16:20what is your advice for everyday normal social media users
277. 16:24in terms of how they might change their behaviour?
278. 16:29I think it's good to be mindful of how much time  you spend on these things.
279. 16:32So a lot of phones now tell you how much time you spend on apps per day.
280. 16:36I think sometimes people might be surprised if they saw how much they did.
281. 16:39I think if you have kids, one of the top things you can do is
282. 16:42there's rate limiting software,
283. 16:44so there's software that  will give them a budget every day
284. 16:46of two hours or something.
285. 16:48Because the thing that's happening with both kids and adults,
286. 16:51is sometimes they develop self-soothing patterns
287. 16:54where you know they get  depressed, they get anxious
288. 16:57and the way they soothe is they just you know keep scrolling and dissociate, disconnect.
289. 17:03But if the thing that's causing you to be anxious or depressed is the content, that's a problem.
290. 17:09And Facebook knows that adults have this problem  they know that kids have this problem
291. 17:14and they also know that when kids try to talk to  their parents about their usage you know kids say:
292. 17:20“I know this makes me unhappy, I can't stop if  I leave I'll be ostracized.”
293. 17:24And because parents have never lived this experience
294. 17:27they sometimes  give advice that makes kids feel really alone
295. 17:30and this is all in Facebook's documents.
296. 17:32You know when a parent says: “why don't you just not use it?”
297. 17:35That can make the kid feel like it's their  fault that they're having these experiences
298. 17:39or they can feel just alone like they can't  tell their parents about it.
299. 17:42And so I think approaching young people particularly from a place of listening
300. 17:46and like trying to refrain from judgment
301. 17:50and come up with solutions together on how to support them and move forward.
302. 17:56Because, you know, there's  this thing called network effects where
303. 17:59for every additional person you  add to a product it becomes more valuable.
304. 18:03This is happening to our kids. Every additional  kid in the school that goes on Instagram
305. 18:07makes it more of a demand to have the other kids on
306. 18:12and so unless we do movements at schools
307. 18:14where we say this whole school we're making a choice to go off Instagram together,
308. 18:19we'll still SMS but we're not doing Instagram anymore, which is a real thing.
309. 18:23At some privileged schools in the [San Francisco] Bay Area, you know, places that executives send their kids to,
310. 18:28the norm established in the schools is we just don't do that,
311. 18:30we don't do social media.
312. 18:32And so these are all choices we can make but it's hard for us to make them individually
313. 18:37and so we have to, you know,  solve problems together not solve them alone.
314. 18:46Journalist: Facebook has  gotten over scandals before.
315. 18:49Does it feel like this time it’s different?
316. 18:52I do think there's an opportunity this time in  terms of before we didn't really have a broad swath.
317. 19:00One of the things my disclosures show is a pattern of repeated behavior
318. 19:05where over and over again Facebook chose profit over people.
319. 19:09And so it's hard to ignore it, it's not just like a scandal like:
320. 19:13“Oh, no yeah we learned again Facebook is bad.”
321. 19:15This time there's enough to say no there's  problems with how facebook is organized,
322. 19:21Facebook is not going to solve this on their own
323. 19:24and I've talked to a number of governmental regulators
324. 19:26who said that this disclosure just  changed the entire tone of the debate
325. 19:31and so my hope is that this time will be different.
326. 19:34And I have a really long view on it.
327. 19:35Part of why I want to found a social movement
328. 19:38is because maybe we do have to push for a couple years,
329. 19:41maybe we have to push for five years.
330. 19:44I'm not going to get, you know, distraught
331. 19:46if the story fades a little bit in  the coming weeks
332. 19:49because I have a long-term view.