Latest / Uncomfortable Conversations with Josh Szeps / "Facebook Whistleblower" Frances Haugen

Transcript

"Facebook Whistleblower" Frances Haugen · Uncomfortable Conversations with Josh Szeps

• 0:00
  Get A Humans, welcome to Safe Space for Dangerous Ideas.
• 0:06
  Here's a dangerous idea for you, all of the negative effects that you know accrute you
• 0:13
  from social media use.
• 0:16
  The algorithms that reinforce what you want to believe and demonize the things that you
• 0:20
  don't want to believe.
• 0:22
  The echo chambers that keep you liking and commenting and scrolling and clicking and
• 0:26
  sharing and arguing the recommendations that tailored just for an automaton like you
• 0:33
  to ingest into your algorithmic robot brain.
• 0:37
  The self-judgment that comes from overuse of social media, the anxiety, the bullying,
• 0:43
  the praying of perverts on underage users, that hollow panoply, that universe, that mosaic
• 0:49
  of negative consequences that constitute the vast psychological experiments that is social
• 0:55
  media use in the 21st century, all those downsides may not be the regrettable side effects of
• 1:04
  the core product of connecting people.
• 1:07
  Those negative effects may actually be part of the core product, fundamental to its
• 1:13
  addictiveness, fundamental to its business model and the people who run those social
• 1:18
  media companies know it.
• 1:22
  That is basically the conclusion to which today's guest came when she left Facebook and
• 1:26
  became a whistleblower.
• 1:28
  She collaborated with the Wall Street Journal a few years ago on a bombshell series of articles
• 1:32
  unveiling what Facebook knows about what it could do to prevent user harm and what it is
• 1:38
  failing to do.
• 1:40
  Mark Zuckerberg has been a busy little bunny on the witness stand in Washington DC lately.
• 1:46
  You may have heard amid all of the Trump chaos that there is a big court has going on.
• 1:50
  The US government is suing meta under antitrust for allegedly maintaining a monopoly.
• 1:57
  It could reshape the future of social media, depending on which way the case goes.
• 2:01
  It is one of the most consequential antitrust cases in tech history.
• 2:05
  It will impact privacy regulation.
• 2:06
  It will impact big tech.
• 2:07
  It will impact social media.
• 2:09
  Francis, today's guest, started working at Google, should probably working out at Google
• 2:15
  as well.
• 2:16
  They have gyms on site.
• 2:18
  She worked at Yelp.
• 2:19
  She worked at Pinterest and ended up at Facebook as a product manager and a data engineer.
• 2:24
  She herself helped to design some of the algorithms that initially figured out how to rank results,
• 2:32
  how to rank your feed, to put in front of you what is most likely to make you linger and
• 2:37
  share and like and click.
• 2:40
  She became disillusioned with that.
• 2:42
  As I say, she made global news as a whistleblower.
• 2:45
  I want to talk to her about ever since having met her about a year ago when I was on a television
• 2:51
  panel show in Australia with her, called Q&A.
• 2:55
  In the green room afterwards, I said, you've got to get on the show.
• 2:57
  It's taken this long just to line up our schedules to discuss with her.
• 3:01
  In an ideal world, how would she advocate?
• 3:04
  In fact, how does she advocate with them on profit that she now works with, reforming
• 3:07
  social media to make it more humane in the most simple and achievable ways?
• 3:13
  Please enjoy.
• 3:14
  This conversation with the one they only, Francis Hagen.
• 3:19
  There's still this kind of like grinding sense that the big tech companies are possibly
• 3:25
  large and like we have no control and blah, blah, blah.
• 3:28
  But if you actually look at the details of how even meta has changed over the last four
• 3:34
  years, it's pretty dramatic.
• 3:37
  Just a couple days ago, they announced that they are going to now detect when under 18
• 3:42
  year olds are using their platforms and lying about their age.
• 3:45
  I like, for example, examining the content that they can post, which they could have
• 3:48
  done a decade ago, but they chose not to because the service grows faster.
• 3:52
  When you know it's just ignore, you have eight year olds that are using your service.
• 3:56
  Just to clarify, when you say, analyze the content that they're posting, this has been
• 4:00
  a point that you've made for a long time, which is they say, well, we can't really verify
• 4:05
  people's ages.
• 4:06
  We have to take their word for it because otherwise we would have to implement some kind
• 4:09
  of a internal police state requiring every user to provide ID and that would be bad for
• 4:14
  whistle tours and so on.
• 4:16
  You've made the point that if you actually sat down with them and you set around a table
• 4:20
  and were like, okay, what are the 10 to 15 easiest things that we could do other than
• 4:24
  that to try to establish people's ages, they would easily come up with them.
• 4:28
  They just don't want to do those things.
• 4:30
  One of which would be track what young people are telling each other because quite frequently
• 4:36
  a 10-year-old will say that they're 13 and then we'll jump on and as soon as they've
• 4:39
  got their profile, we'll be talking to a bunch of other 10-year-olds about how they're
• 4:43
  10.
• 4:44
  Exactly.
• 4:45
  A staff that I think most parents would be shocked by is even two years ago, so we can
• 4:50
  assume this number is probably worse now.
• 4:52
  Even two years ago in 2022, this is like two and a half, 30% of seven to nine-year-olds
• 4:59
  were on some form of social media.
• 5:02
  We are not talking about should a 13-year-old, should a 14-year-old, should a 15-year-old
• 5:06
  be on social media.
• 5:08
  We need to be having a conversation of should a seven-year-old, should an eight-year-old,
• 5:10
  should a five-year-old be on social media.
• 5:12
  I think while we can have an honest, intellectual conversation, and I feel actually quite conflicted
• 5:18
  on it.
• 5:19
  I'm not a prohibitionist.
• 5:20
  I just want to see safer social media get built.
• 5:23
  I think most of us can agree that having eight-year-olds, changing their behavior to get
• 5:28
  more likes and go viral, is probably not developmentally appropriate for them.
• 5:33
  Just while we're on the question of age-gating, because I know this is a hot topic at the
• 5:37
  moment, and the Australia is pretty aggressive.
• 5:39
  Pregnant Australia.
• 5:40
  On this, yeah, yeah, and thank goodness for it since I have seven-year-olds, and I want
• 5:45
  that gate as close as tightly as possible until 16.
• 5:49
  But the question of ideas, I've never quite understood why this is a potential human rights
• 5:54
  violation or terrible for whistleblowers if you could.
• 5:58
  Couldn't you just have an exemption where if you don't want to upload an idea, you can
• 6:02
  just send an email and they'll allow you to create an honest account, but at least the
• 6:07
  additional bit of friction would probably discourage a lot of underage people from doing
• 6:11
  so.
• 6:12
  I think ideas are one of these things where there's a lot of potential ways that it goes
• 6:16
  just badly.
• 6:17
  Like, we assume perfect competence, we assume people following the rules, and that just
• 6:26
  can't be counted on for the wide number of social media platforms that exist.
• 6:30
  Like, there isn't one or two reasons.
• 6:33
  So when you think about the way that ideas are handled in well-functioning high-tech places
• 6:38
  like Australia and Scandinavia and places where when you're applying as a US citizen
• 6:46
  to come to Australia and to get the visa waiver, you can now, like, just hold your passport
• 6:51
  over your smartphone and it knows what it is and it detects all the information, there's
• 6:56
  like facial information.
• 6:57
  That's a pretty example.
• 6:58
  Stuff, there's like, there are heaps of examples where I need to verify my identity for
• 7:03
  some instant organization that is collaborative with my state, like the state government,
• 7:09
  for example, whether that's like the equivalent of the DMV or something like that or a Medicare
• 7:13
  or something.
• 7:15
  And it'll be a third party app, which has two factor authentication and stuff.
• 7:19
  And they can, so you don't need to give up your own sensitive information to the entity
• 7:24
  itself.
• 7:25
  But you outsource that.
• 7:26
  Couldn't you have a collaborative central social media clearing house that just verifies
• 7:31
  people's ages in as anonymous and secure a way as possible and then gives them a tick
• 7:36
  of approval or not depending on what age.
• 7:39
  So I think you could hypothetically do, like, so they often talk about this concept of zero
• 7:43
  knowledge proofs, where you have an outside organization and you have some kind of credibility
• 7:49
  or credential that you can carry on the internet without disclosing who your fundamental identity
• 7:55
  is.
• 7:56
  Because I think one of the things that it's sometimes just taking for granted for particularly
• 7:59
  in Australia is not all places are equally functional and that systems that when they're
• 8:03
  put in place often get used for counter purposes in places that are less, less functional.
• 8:08
  And before we get too far, I think one of the key things to also talk about is I think
• 8:12
  identities are a distraction, like people respond really emotionally when we start talking
• 8:17
  about IDs because there are, there are these kind of implications of, you know, the state
• 8:22
  is watching over me or whatever.
• 8:24
  I think the important question to ask is what extra do we actually get versus say looking
• 8:29
  at people's faces when they sign up, looking at the content they engage with, looking at
• 8:34
  simple things like how fast do they type, can they spell, how do they hold their phone?
• 8:40
  You know, I watched an interview with a kid, was it with a number of kids about roadblocks
• 8:45
  and labor exploitation, which I think is a fascinating topic.
• 8:48
  You have this $10 billion company that was built on the backs of child labor, like kids
• 8:52
  writing games and then having very, very unequal rules around monetizing those compared
• 8:56
  to say YouTube or any of the or TikTok.
• 9:00
  But the interview this kid is part of that documentary and the, you know, the, the, the
• 9:06
  journalist to just spend a few minutes talking his kid about how passionate he is about
• 9:09
  running his business where he makes these video games and makes $100 a month or something.
• 9:14
  And she's like, is everything you don't like about this platform?
• 9:17
  And this little 10 year old without missing a beat goes, he did when creepy adults talk
• 9:21
  to me.
• 9:23
  And she like was startled, right?
• 9:25
  Because in roadblocks, everyone is like a cartoon, like how do you know you're talking
• 9:28
  about creepy adults, right?
• 9:29
  You can't show your face even if you wanted to show your face.
• 9:32
  And he's like, they tie so fast, right?
• 9:37
  Like we have all these things where, you know, maybe adding ideas would get 0.1% more
• 9:42
  precision than we currently have.
• 9:44
  But if it means that we don't act, because people get scared and the platforms, these
• 9:48
  are fear monger, like we just don't use them.
• 9:51
  Right.
• 9:52
  Okay.
• 9:53
  So yes.
• 9:54
  And you're, I mean, of course, I'm speaking now that you say that I realize that I'm speaking
• 9:56
  in such an old fashioned, funny, dirty way in an era of AI to think that you would need
• 10:01
  a physical, you know, identification card that has this stamp of approval of some kind
• 10:07
  of authority.
• 10:08
  I feel like my grandmother holding carrying papers around war-torn Europe or something
• 10:12
  is showing papers.
• 10:13
  Obviously, AI is going to have more sophisticated ways of sorting out the characteristics of
• 10:18
  people who attend versus people who are 18.
• 10:21
  But we have gotten sidetracked by the youth thing because I am more interested in just
• 10:25
  your general snapshot.
• 10:26
  I mean, this is a vast cycle of experience that all of us are going through, not just
• 10:30
  young people, right?
• 10:31
  And I don't know whether or not we're on a better or worse trajectory than we might have
• 10:37
  worried that we would be 10 years ago.
• 10:38
  So let's just start there.
• 10:39
  Like, how would you?
• 10:40
  Sure.
• 10:41
  In other words, this is a problem that I think we all are aware of.
• 10:44
  But it's hard for us to articulate exactly what the problem is, but you spend your life
• 10:47
  articulating what the problem is.
• 10:49
  So can you define for us what the problem is?
• 10:54
  I was at a really basic level.
• 10:55
  It's that a handful of corporations and one is an outlier of badness in this way, which
• 11:02
  is meta because it's run by one person with these handful of corporations that are pretty
• 11:07
  fundamentally unaccountable to the countries they operate in.
• 11:10
  You know, they've done a wonderful job of selling complete chaos within the government
• 11:14
  in the United States to the point where, with even a 60-co-sponsors, we couldn't get
• 11:19
  our kids safety law passed last year, right?
• 11:21
  You know, I would say we probably have a much more toxic environment than even Australia
• 11:25
  does.
• 11:26
  And it doesn't matter because they've so much fun around it.
• 11:31
  But when we move to, like, I think the important question, so there's two things.
• 11:38
  So one is that question of unaccountability, the corporations that don't really have
• 11:43
  any checks and balances on them.
• 11:44
  We have no right to even basic transparency anymore because they shut down the few little
• 11:50
  windows that we had into these platforms over the course of the last couple years.
• 11:55
  That combined with the fact that they really intimately impact all of our lives.
• 11:59
  Like even if you, yourself, don't use social media, you interact with people every day
• 12:03
  who do use social media.
• 12:04
  I've just been a lot of time on it.
• 12:06
  You only need 3% of a population on the streets to have a revolution.
• 12:11
  If you have systems where people aren't picking out what you focus on, but AI's are picking
• 12:16
  out.
• 12:17
  You get into a situation where people can pull down these rabbit holes where they keep
• 12:20
  seeing more and more reinforcing content.
• 12:23
  That's about needing disorder.
• 12:25
  That's about paranoia about the government.
• 12:28
  That's about racial hatred.
• 12:29
  And it's one of these challenges where when we have no accountability, we have no ability
• 12:34
  to see if the problem's getting worse or like what is the level of severity, which means
• 12:37
  we don't actually get to choose to use these products.
• 12:41
  It's a really serious situation.
• 12:43
  And just to channel the inevitable concerns or pushbacks of conservatives when they
• 12:48
  hear you articulate the problem in that way, I would also remain alert to the fact that
• 12:55
  the echo chamber effect is not just at the 3% of the extremes, but we are all in an ambient,
• 13:02
  I guess there's a ring of concentric circles of greater or lesser influence and capture
• 13:07
  by our own echo chamber.
• 13:09
  100%.
• 13:10
  And thus they become blind spots that include, including in the media where, as you may
• 13:15
  have seen during covert, for example, or over the trans debate, there'll be consensus
• 13:19
  that develop as a result of the soup that you're swimming in, the kind of social media
• 13:25
  that you're consuming, the kinds of penalties for being a dissenter from a mainstream orthodoxy
• 13:32
  that conspire to limit your thinking and make you perhaps more serious or more blinded
• 13:38
  to other points of view.
• 13:39
  And certainly the left was vulnerable to that over the past five years as well.
• 13:43
  100%.
• 13:44
  But I think one of the ways to think about it a little bit is even very, very basic things
• 13:49
  which would allow people to consent to what they're using, if you don't have informed consent,
• 13:54
  you're not actually consenting.
• 13:56
  What seems as basic as, we have no guarantee, when people say things like, can we see every
• 14:03
  post that was seen more than 100,000 times in the United States, right, like none of those
• 14:09
  posts violate anyone's privacy, right, if 100,000 people see something, that's not private.
• 14:14
  They turn around and they play things like the privacy card.
• 14:17
  They say, we can't do that because we violate people's privacy.
• 14:20
  And so it's one of these questions of, you know, if you don't get to see what echo chambers
• 14:25
  exist in society, so you don't get to decide how to govern systems like this, you don't
• 14:29
  get to see what echo chambers you're in, though we are a little bit better than we were
• 14:33
  four years ago.
• 14:34
  Right.
• 14:35
  So four years ago, I started talking about the right to reset that you should have the
• 14:39
  right to reset any algorithm that you interact with and go back to whatever is the default
• 14:44
  that they start with because people notice when they start going down rabbit holes often.
• 14:49
  It's not just a thing where people have no idea what's going on.
• 14:54
  You know, for the first time that got interested in Instagram a couple of months ago, you know,
• 14:59
  it's not super obvious.
• 15:00
  You have to know how to navigate the menu, but you do have a right to reset today.
• 15:04
  Imagine if we had that in more platforms or if you got learned, if you started sliding
• 15:08
  down a rabbit hole, those are real changes that we can make and not censor anyone.
• 15:13
  So I mean, the right to reset sounds good, but it's not the right to read to opt out
• 15:18
  of an algorithmic curation of content, right?
• 15:21
  I mean, I think this whole question about whether or not a company like Facebook or
• 15:26
  Meta is a publisher or whether it is just a supposed clearinghouse through which other
• 15:33
  individuals are capable of contacting each other is sort of belied by the fact that obviously
• 15:40
  they're putting their finger on the scales of certain types of content and not on other
• 15:44
  types of content.
• 15:45
  It's not like when Facebook started out and it was just a reverse chronological feed
• 15:51
  of all the things that the people you follow had been posting.
• 15:53
  Now, it intentionally elevates some things and intentionally demotes other things and
• 15:58
  it's all very well to say, well, I want to reset so that I no longer see quite so many
• 16:04
  skinny women with big boobs, but it's not going to take very long after resetting for
• 16:09
  it to notice that I spend more time hovering over images of skinny women with big boobs
• 16:13
  or whatever it might be, chocolate cake, then I do over charts of, you know, national
• 16:19
  debt levels during the Greek financial crisis.
• 16:23
  And so isn't the actual solution to provide a restructuring or an option for restructuring
• 16:29
  social media such that there isn't a constant algorithmic feedback loop that is borrowing
• 16:34
  me into whatever rabbit hole happens to be the most attention capturing?
• 16:40
  So I think one of the, this is like a kind of a root thing in terms of a figure out what
• 16:45
  is the path forward is how do you set a system of rules or incentives that gets the
• 16:49
  outcome that we want, right?
• 16:52
  And I think one of the things that we can all agree on is that we would like to give
• 16:56
  consumers genuine choices.
• 16:58
  And the case of what you described, like, I want to have social networks where I only
• 17:03
  hear from the people that I choose to hear from, right?
• 17:05
  I don't want you to push me into mega groups with hundreds of thousands of members so that
• 17:09
  my feet is always full.
• 17:10
  Like I don't want the algorithm to be constantly so hyper personalized and tuned, like I want
• 17:16
  to get an ambient sense of like what other people are being exposed to, you need to take
• 17:20
  a step back and say like, well, right now, what are the incentives that cause these products
• 17:25
  to become what they are?
• 17:26
  And I think one of the core ones, and this is just a quote in the, probably the other
• 17:30
  biggest whistleblower.
• 17:31
  I should know, Sarah Williams has come out.
• 17:33
  I used to say Arturo Behar was the largest other whistleblower from Facebook, but that's
• 17:38
  not true because Sarah is doing a great job.
• 17:41
  So I would say, so Arturo Behar, who was the big whistleblower about two years ago.
• 17:46
  He came forward and said, I worked, I did a bunch of really in-depth work, polling children,
• 17:51
  right?
• 17:52
  You know, asking children about their experiences on Facebook, asked hundreds of thousands
• 17:56
  of kids in the United States.
• 17:57
  Now they found was that one in eight girls, between the ages of 13 and 15, which really
• 18:04
  means all the way down to those seven-year-olds, just to be super honest.
• 18:09
  One in eight girls trained 13 and 15, so they received an unwanted sexual message in
• 18:13
  the last seven days, in the last week, one in eight.
• 18:18
  And he said, in front of the sudden, you know, if we want that number to go down, the only
• 18:23
  way it's going to go down is if Mark Zuckerberg has to report it next to his profit and loss
• 18:28
  numbers each quarter.
• 18:30
  And I think there is a basket of simple metrics.
• 18:33
  You know, you can't play the privacy card here and say, oh, people's privacy is being
• 18:37
  violated.
• 18:38
  We can't tell you anything about the platform.
• 18:39
  I mean, there's a basket of metrics.
• 18:41
  There are things like, when you poll users, maybe this could be kids, this could be adults,
• 18:45
  there's various ways of infrastructure this.
• 18:47
  And as in the question, I control my usage of the platform.
• 18:51
  I feel in control of how much I use the platform.
• 18:53
  I can stop whenever I want to.
• 18:55
  Or if I say, does this interfere with your ability to go to bed when you want to?
• 18:59
  Do say yes.
• 19:01
  If you start after a report metrics like that, you will start to push products away from
• 19:07
  being so addictive and so hyper-tuned, because the only way you make those numbers go down
• 19:11
  is by giving people more control, giving them a little bit more of a great debt crisis,
• 19:17
  and a little bit less of whatever their most vulnerable foibles.
• 19:24
  I mean, that strikes me as being a little bit like a gigantic company is producing toxins
• 19:32
  that contaminate everybody, and your solution to it instead of addressing the production
• 19:37
  of toxins is to force them to report in their annual reports about how many toxins they're
• 19:43
  releasing.
• 19:44
  It's better than nothing.
• 19:45
  And quantity.
• 19:46
  Right.
• 19:47
  It's better than nothing, but it'd be better to address the toxins themselves in the sense
• 19:51
  of addressing it.
• 19:52
  So the challenge, so it's interesting, like I love your analogy.
• 19:56
  I think it's very germane, right?
• 19:58
  Like there's this question of like in lots of other arenas, we have really basic bright
• 20:03
  lines, like we have ways of distinguishing, is this good or is this bad?
• 20:08
  What's so different about technology?
• 20:09
  What's so different about social media?
• 20:11
  And the thing that we have seen time and time again is if you write a law and you prohibit,
• 20:16
  you describe a thing, and even if you describe it very precisely, the companies go and make
• 20:20
  a slightly different thing and say, this is no longer the same thing.
• 20:23
  It's very, very hard.
• 20:24
  So for example, are you going to ban all personalization?
• 20:28
  People like, like the personalization and Netflix people love that, you know, I'd say
• 20:32
  even on YouTube, like YouTube has a much less like, you know, people reporting, I've been
• 20:37
  damaged by YouTube compared to saying I've been damaged by Instagram or damaged by TikTok.
• 20:42
  You know, there, there's, there's, it's really difficult to come in and say with a blanket
• 20:47
  like, oh, you, you can't do this anymore.
• 20:50
  No, I mean, I don't, yes, so, so, but, but somewhere between the bludgeon of you can't
• 20:56
  do this at all and the, the feather duster of tell us how bad it is, and hopefully be
• 21:02
  shamed.
• 21:03
  So I think there isn't in between.
• 21:04
  Yeah.
• 21:05
  So I think one of the interesting questions is happening right now with section 230.
• 21:08
  So I, there, I've heard rumor, I think this is incredibly dangerous.
• 21:11
  I don't support it at all.
• 21:13
  I've heard a rumor that certain people in, in, in the Senate are saying, hey, we should
• 21:19
  just sunset 230, we should set three years from now.
• 21:23
  Uh-huh.
• 21:24
  So, so for context, so I like to apologize to Australia and apologize to the world.
• 21:30
  It's my duty as a journalist to ask silly questions because lots of people and not a, don't
• 21:34
  follow.
• 21:35
  No, it's not a sling question.
• 21:36
  It's very, it's very important.
• 21:37
  So way back in 1996, we just have these things called bulletin boards and a bulletin
• 21:41
  board is kind of what it sounds like, you know, you know, bulletin boards like that cork
• 21:44
  board thing.
• 21:45
  You put a little pin in it and other people can read stuff that you pin to the bulletin
• 21:48
  board.
• 21:49
  Bolton boards were basically like chronological feeds where there'd be different conversation
• 21:53
  topics and you'd just post a thing in and fall into that flow, that chronology of whatever
• 21:58
  that topic was, not very stimulating, but people loved them.
• 22:03
  People loved them.
• 22:04
  People spent all their time on them back in 1986.
• 22:07
  So this problem arose on two sides.
• 22:09
  So on one side, there was one of these bulletin boards and they didn't do anything and someone
• 22:17
  said something disparaging about a company that was not true and sued the bulletin board
• 22:24
  for hope soon.
• 22:25
  And the courts ruled because they weren't doing anything, they were like a neutral conduit.
• 22:30
  They weren't a publisher.
• 22:31
  They weren't responsible for what people are saying.
• 22:33
  But a different court case came out that was almost identical.
• 22:36
  Only this platform did take a certain amount of proactive action.
• 22:40
  They went in there and they removed meal pornography, they removed a few other things.
• 22:45
  And the courts said, hey, because you are going in there and curating that experience,
• 22:50
  you've taken responsibility, you are a publisher.
• 22:53
  So this very unfortunate situation arose where companies that were trying to create higher
• 22:58
  quality experiences were being fined, were being punished and companies that were reckless
• 23:03
  and doing nothing were viewed as like, oh, then you have no responsibility.
• 23:08
  So Congress stepped in and said, hey, you have the right to curate without giving up
• 23:12
  this safe harbor position.
• 23:15
  And for publishers in general, the service providers in general, if you are providing
• 23:19
  a space for people to speak on the internet, you are not responsible for the content that
• 23:23
  those people produce.
• 23:25
  So fast forward in time and we get into a very, very different world.
• 23:29
  So instead of having everything you see, we picked out by other humans in your community,
• 23:34
  even if that community is a digital community.
• 23:36
  And the order is just, you know, you are, you are the one acquiring information.
• 23:40
  You are hunting down and learning things.
• 23:42
  We move into a world where these things called recommendation feeds appear.
• 23:46
  And it's quite late.
• 23:47
  It's like 20, 20 years, close to 20 years, it's like 2011 is when like the first algorithmic
• 23:53
  feed shows up in Facebook and it takes another like five years for Twitter to fall.
• 23:58
  Did Facebook find any of that?
• 24:00
  By the way, Francis.
• 24:02
  Oh, I think Netflix had an algorithmic feed at that point.
• 24:07
  So I don't think we can purely say that they were the, oh, no, no.
• 24:10
  So the first commercially used recommender system was on Amazon.
• 24:13
  I went and looked up all this.
• 24:16
  So the first one was like, so it was like, you know, you know, you might want this.
• 24:20
  You might also like this toothbrush.
• 24:21
  You might.
• 24:22
  Okay.
• 24:23
  But Facebook was the first social media platform to, yeah, to start, start fucking around
• 24:29
  with what gets, gets to go at the top and what gets to go at the bottom.
• 24:32
  And I think, I think one of the ways to think about this and I, I, I want to be clear,
• 24:36
  I am partially responsible for the world that we live in today because I worked at Pinterest
• 24:40
  and I was, I was the product manager who, who picked out what you saw in Pinterest,
• 24:45
  you know, and, and one of the things that Pinterest saw, and this is the same arc that took
• 24:49
  place on Facebook.
• 24:50
  It basically is a very similar arc to anywhere that chooses to go down the algorithmic
• 24:54
  direction is they started out with a network where you chose who you received content from.
• 24:59
  They found that people didn't consume enough that it was too hard to go and, and build
• 25:04
  out a community to hear from.
• 25:05
  So they started getting new topics, tops of kind of a little bit like groups, you know,
• 25:09
  they're getting chunkier things where you'd say, you like cooking.
• 25:13
  Now you're going to get, we'll always know we can fill your feed with cooking stuff.
• 25:17
  And then they went one step further and said, we now understand the significance, like
• 25:21
  the meaning of this content better.
• 25:24
  We know that you don't just like cooking, you like vegan cooking, you like gluten-free
• 25:28
  cooking.
• 25:29
  We're going to give you specifically that.
• 25:31
  And the challenge is that as we move from humans to computers, you know, we say, hey,
• 25:37
  we want you to consume more and so we are going to figure out using computers how to make
• 25:41
  sure we always are showing you things, you never are going to get bored.
• 25:44
  We also get into a zone where those computers don't actually know what they're showing you.
• 25:48
  And they don't understand significance, they don't understand the ancient idea that often
• 25:53
  our greatest vulnerabilities are the things that we're drawn to.
• 25:57
  And so we end up in this place where, you know, section 230 is now being applied to those
• 26:04
  algorithms.
• 26:06
  So before it was this question of, well, if we provide a space at all for people to post
• 26:11
  it, then we have to give immunity to the people who are providing that space because
• 26:15
  there's no way to ensure that never, like never anything gets through.
• 26:20
  But the flip side now is, when matter is more than any given piece of content is that
• 26:26
  content selection system.
• 26:29
  And right now there are no consequences for optimizing that system to be harmful to you
• 26:33
  other than the risk that you may leave.
• 26:36
  Right.
• 26:37
  And so just to make a few analogies here, because this is obviously a hybrid space, which
• 26:42
  is neither a newfangled kind of newspaper, nor an old fashioned phone service, right?
• 26:48
  You know, because I hear all these analogies being made.
• 26:51
  Well, the internet is just like a phone service, right?
• 26:53
  So it's like a utility essentially.
• 26:54
  It's a series of tubes.
• 26:55
  It's a series of tubes, as the Republican Senator famously said.
• 27:00
  The internet is not something that you just dump something on.
• 27:03
  It's not a big truck.
• 27:05
  A series of tubes.
• 27:07
  The obviously, if there were no algorithm in the room, if I was just receiving the posts
• 27:14
  that my friends and businesses that I follow had been posting in a reverse chronological
• 27:18
  order, then that would be true.
• 27:19
  That would be akin to a phone company being held liable for the conversations that I
• 27:24
  have, you know, on the telephone that would be unfair.
• 27:29
  On the other hand, if it is a lot more like a newspaper where it's choosing what to
• 27:33
  show me and what not to show me, you know, even if it's not writing the articles itself,
• 27:40
  let's suppose it's a news aggregator service or something like that, Apple News or, you
• 27:44
  know, the associated press or something where it's taking in content from other people
• 27:48
  and then arranging them.
• 27:49
  Then obviously, it does have a responsibility to be held accountable for the kinds of things
• 27:53
  that it's showing me and any double serious consequences they might have.
• 27:57
  It's interesting that you point out the existence of the human versus non-human because when
• 28:01
  you mention Netflix, that got me thinking we, as you say, we do want a certain amount
• 28:07
  of curation, like one of the great clever tricks of Netflix was in the early days to
• 28:14
  identify what productions it should green light on the basis of user behavior, you know,
• 28:20
  House of Cards comes about because they notice that people who watch Kevin Spacey movies
• 28:25
  also watch David Fincher movies and also like British political crime shows.
• 28:31
  And so they buy the rights to a British political crime show and they get David Fincher to direct
• 28:37
  it and they get Kevin Spacey to star in it and you end up with a great product that tickles
• 28:41
  the likes of me.
• 28:44
  Where are we, like I am sympathetic to the anti social media argument that says that as
• 28:51
  long as there is an algorithm involved in the loop, you are a publisher because you
• 28:56
  control the algorithm.
• 28:58
  But it sounds like you're not crazy about abolishing section 230, which would seem to
• 29:03
  me be a logical consequence or conclusion.
• 29:06
  I want to agree.
• 29:07
  I want to say that that's my position.
• 29:08
  I think my position is not that we should rip the bandaid off and say we're going to
• 29:11
  sunset it in three years because we are in a position right now where we're very bad
• 29:16
  at passing laws.
• 29:17
  Right, we pass progressively less laws every year, like every new session of Congress,
• 29:24
  we pass less laws from the one for it.
• 29:25
  And it's the point where we don't really do budgets while we end up a year behind
• 29:33
  on passing the budget because we can't even agree on that.
• 29:37
  Then you threaten to drive the government off a cliff and have the United States default
• 29:40
  on the internet and then everyone comes down.
• 29:42
  We're classy like that.
• 29:44
  The first time I went to Australia to talk with representatives, I got to go to
• 29:51
  De Canberra.
• 29:52
  I got to meet all these interesting people.
• 29:54
  It was while we didn't have a speaker of the house and I kept being asked, so why are
• 30:01
  you in Australia?
• 30:02
  And I'd have to be like, well, we haven't had a speaker of the house for 11 days.
• 30:05
  That hasn't happened since 1860.
• 30:09
  We're back to a phase.
• 30:12
  Hopefully it won't last forever.
• 30:13
  We'll see how it goes.
• 30:14
  I feel for you.
• 30:15
  I feel for you.
• 30:16
  Also, can I just say we're about to have a federal election in less than two weeks and
• 30:22
  I've never felt more grateful to live in such a boring country because you have these
• 30:26
  two very boring, gray, faced, middle-aged, bureaucratic men fighting about basically nothing.
• 30:34
  And I'm like, thank goodness for boring politics.
• 30:40
  Anyway.
• 30:41
  Carry on.
• 30:42
  To flatter all the Australians, my husband and I have had conversations about where would
• 30:47
  we live in the world if we didn't live in Puerto Rico.
• 30:49
  And we would probably, I hate to pick a team, we'd probably live in Melbourne.
• 30:53
  Yeah.
• 30:54
  I like Melbourne.
• 30:55
  It's delightful.
• 30:56
  Yeah.
• 30:57
  It's right.
• 30:58
  But to finish the thought, so this question is, okay, so if Section 230 is flawed, if
• 31:02
  like literally we've gone to the point where, for context, 44 U.S. states have sued
• 31:08
  Meta saying, hey, you straight up lied about hurting kids.
• 31:13
  Like you knew you were hurting kids, you were like, let's just drive this bus off the
• 31:16
  cliff.
• 31:17
  No one's going to catch us.
• 31:19
  When they went to the courts, Section 230 is so flawed that the judge said, hey, you
• 31:24
  don't have a case because it's not illegal to intentionally harm children, like knowingly
• 31:29
  intentionally harm children under Section 230, like that's where we've gone to.
• 31:35
  So if Section 230 is so flawed, like why, what should we do with it?
• 31:39
  Because that's the real issue.
• 31:40
  The issue is there's the world without Section 230, the internet will break, like you cannot
• 31:46
  run the internet as it currently is.
• 31:48
  So what's the alternative?
• 31:49
  And the thing that I've been suggesting, and this will potentially make your blood
• 31:55
  boil because it does let the tech companies a little bit off the hook, is I think they
• 32:00
  should have to, like for some specified set of harms, right?
• 32:05
  We have to actually structure a little bit more than the Europeans did, even though I love
• 32:09
  that the Europeans just said, if you know there's a harm, you've got to deal with it.
• 32:12
  You guys tell us about it.
• 32:14
  I think for some set of harms, and we need a process for how we add harms over time,
• 32:19
  so for some set of harms, maybe start with 5 or 10, the companies have to disclose enough
• 32:26
  data, and we can, we have a conversation, even say what some initial data should be,
• 32:31
  enough data that we can tell the magnitude of the problem, and they should have to disclose
• 32:35
  what they're doing to make that problem less bad.
• 32:39
  And if they do those two things, and they give us ways to substantiate that they are
• 32:43
  actually executing that plan, remember we're not specifying the plan, they're specifying
• 32:49
  the plan, they set some kind of sense of you must be making progress, or we will rule
• 32:52
  you in contempt, give them safe harbor for those specific harms, because right now our
• 32:58
  situation is we have right no right to any data, and so questions like, does this interfere
• 33:05
  with your sleep?
• 33:06
  You know, a third of kids in the United States are on, a third of high schoolers are on
• 33:12
  screens till midnight or later, most school nights, which means probably 10 or 20% are
• 33:18
  on till two.
• 33:19
  Right?
• 33:20
  Like we have so little data that we don't have the right to know what fraction of the population
• 33:23
  says, I have trouble going to bed because of this application.
• 33:26
  Why can't we pull them independently?
• 33:28
  Why do we need the, the, the, the consent of the social media companies?
• 33:32
  I think that's a great idea.
• 33:34
  Like I think it's one that that is definitely a thing where there's value in doing more
• 33:38
  large skill polling.
• 33:40
  Um, I strongly encourage people have funding sources for that, or you know, an appropriate
• 33:45
  statistics background, um, but I think one of the key challenges there is in the United
• 33:50
  States, we've actually really struggled with polling in the modern era, as you might have
• 33:54
  noticed from our presidential figures, because, um, the ways of accessing people that has
• 33:58
  gone actually quite hard.
• 33:59
  Right.
• 34:00
  So, um, I do strongly support that, but we, we have to figure out funding sources for
• 34:03
  that.
• 34:04
  Right.
• 34:05
  And I guess if you put it inside the app, if you coerce the social media companies to
• 34:07
  comply, and there was a pop up, you know, inviting people to take part in a 22nd survey,
• 34:13
  uh, every once every few weeks, then that would probably be a more effective way than
• 34:17
  trying to call the model landline during dinner and asking to speak to a 10 year old member
• 34:21
  of the household.
• 34:22
  And that's the only thing too, is like, how do you contact native roles?
• 34:25
  Yeah.
• 34:26
  Right.
• 34:27
  Other than through the app.
• 34:28
  And just to just before we leave section 230, I hear this as a, as a, as an article
• 34:32
  of faith from people I trust who know a lot about tech that it would be disastrous for
• 34:36
  the internet to abolish section 230, you know, the whole edifice would collapse.
• 34:42
  Bear with me for a second, since we're in a moment of, uh, US government is playing
• 34:47
  with devaluing the dollar and erecting, you know, and turning its back on all kind,
• 34:51
  on NATO and the global financial system and kidnapping people and kidnapping people.
• 34:56
  Like if we're in the, it's not, it's not immense words.
• 34:59
  Yeah.
• 35:00
  So if we're in the blue sky zone of like everything is possible and none of our previous
• 35:04
  assumptions can be held as, as givens, why is it a given that the internet, the way
• 35:09
  that social media currently conceives of it has to remain.
• 35:13
  Like if you abolish section 230 and all of a sudden, they can't have the algorithms
• 35:18
  that they have without being accountable and they get delused by lawsuits from private
• 35:22
  citizens and they have to change the algorithm.
• 35:26
  Why would that not just be similar to when Ralph made a forced car companies to introduce
• 35:31
  seatbelts and they were like, well, we can't.
• 35:32
  There's no way you can't put a seat on every single car or the expectation that we have
• 35:37
  of McDonald's that not a single McDonald's hamburger anywhere in the world should make
• 35:41
  somebody sick.
• 35:42
  You can imagine if we live in a world where McDonald's hamburgers made people single
• 35:45
  the time.
• 35:46
  McDonald's would say, you can't pass a law.
• 35:48
  How?
• 35:49
  There are billions and billions of burgers.
• 35:50
  How can we possibly, you know, it's going to change the whole, well, yeah, maybe it
• 35:53
  just has to change.
• 35:54
  Maybe the whole thing just has to be burned down the way that it is and they'll find a
• 35:57
  way to deal with.
• 35:58
  I feel like I think that my job has two, two planks to it.
• 36:03
  So one is to give people hope that things can change because they actually have changed
• 36:08
  much last four years as a result of the kind of pressure that we're putting on the platforms
• 36:12
  now.
• 36:13
  But it's to also, I remind people that there are things we want technology to be able
• 36:18
  to do.
• 36:19
  The technology is likely never going to be able to do.
• 36:21
  At least not a way that we would want that wanted to be executed.
• 36:24
  I'll give you an example.
• 36:25
  So some of the things that right now would go into effect if I, I section G31 away is
• 36:33
  like, for example, right now, if you are seen as promoting terrorism, like you are liable,
• 36:40
  right?
• 36:41
  And that's actually a thing that is true, that is for systems that aren't like social
• 36:44
  media platforms.
• 36:45
  That's true on the internet today.
• 36:47
  What is terrorism?
• 36:48
  Right?
• 36:49
  You know, in the case of the Luigi Mangione shooting, that's where a guy went and assassinated
• 36:55
  the CEO of United Healthcare.
• 36:56
  It's the largest healthcare company in the United States.
• 36:59
  He never put out a manifesto.
• 37:01
  He never, like the manifesto that was released, the cops released it, he didn't release it.
• 37:07
  It's he killed one person, like there's not really a basis to say that he's a terrorist,
• 37:11
  but he's being charged with terrorism.
• 37:13
  You know, there's a lot of different things where this administration particularly would
• 37:16
  come out and say these protesters aren't protesters, they're terrorists because they
• 37:19
  broke these windows, but they scare people.
• 37:22
  And as soon as you start having fuzzy lines, now we need to go to the platforms and say,
• 37:26
  you need to go find all the people who potentially are dangerous.
• 37:30
  Those are now liable for it.
• 37:32
  And when that happens, like one of the core, the reasons why I always end up coming back
• 37:37
  on, you know, we want to have metrics, we want to see improvement in metrics, there's
• 37:40
  a lot of different ways to get from point A to point B when it comes to check.
• 37:44
  Lots of different ways we do this.
• 37:46
  The reason why I don't say you need to take down every act of hate speech is because computers
• 37:51
  don't understand what we're saying, right?
• 37:55
  There's a huge amounts of gray lines.
• 37:58
  And you end up being in a situation where if you want to guarantee you get every last piece
• 38:02
  of hate speech or every last piece of terrorism content, you will take down huge spots of totally
• 38:08
  innocent content.
• 38:09
  Okay.
• 38:10
  It was the computer, not exactly sure.
• 38:11
  Okay.
• 38:12
  I mean, you're all right.
• 38:13
  You've won me over on that point.
• 38:14
  Let me try to see if I can split it apart and split hairs here.
• 38:18
  Suppose you...
• 38:19
  I will welcome it.
• 38:21
  Suppose you retained immunity for the social media companies from state prosecution, right,
• 38:27
  either at a federal or a state level.
• 38:28
  So you can't go after them for terrorism content.
• 38:31
  You can't go after them for hate speech or whatever.
• 38:34
  But you make them for the first time become civilly liable to lawsuits from individuals
• 38:41
  who claim that they've been harmed by being sent down a rabbit hole because their daughter
• 38:46
  was looking for dieting tips and ended up consuming lots of anorexia content, for example.
• 38:53
  So they have more of a responsibility or more of a duty of care towards their uses, but
• 38:57
  you're not opening the door for them to be hounded by the authorities.
• 39:01
  Would that assuage your concerns there?
• 39:03
  I think there's an opportunity to go in and pick categories of harm.
• 39:08
  And this is one of the places where it gets really can not just the United States is if
• 39:12
  we were to sit down and find where the overlap is.
• 39:15
  Like the place that I think is possible to start as kids are like coming in and saying
• 39:19
  like there are categories of known harms.
• 39:22
  Like one of the projects that my nonprofit is working on is we've made a taxonomy of
• 39:26
  harms and kids and we have about 40 discrete harms that have at least some evidence.
• 39:31
  There's a huge range.
• 39:32
  Some of them are very, very meticulously researched.
• 39:34
  Some are more conversations that are just becoming prominent, things like labor exploitation.
• 39:41
  The first laws in the United States around making money off your kids on social media
• 39:44
  are just being passed now.
• 39:47
  You know, I think you can go in there and say, let's try it out.
• 39:50
  Riley, let's try it out for eating disorders.
• 39:52
  Let's try it out for depression.
• 39:54
  Let's try it out for sleep deprivation or impacts on schools.
• 40:00
  But the challenge there again, I think the path actually is going to let you get there
• 40:05
  or like what would be the remedy from the court is things like saying, let's go and
• 40:10
  agree to this in metrics, let's make sure you are actually asking children and adults
• 40:14
  on what their experiences are because to quote our show again, when it comes to things
• 40:19
  like unwanted sexual imagery, Facebook was saying there was basically no unwanted sexual
• 40:25
  imagery being distributed.
• 40:26
  And when you ask kids, lots of them said, this is what my experience is online.
• 40:31
  And so it's one of these questions of we need to change the process of how we create technology
• 40:36
  and we need to change the way that we watch how that technology is being developed
• 40:41
  because it's been basically acting in a vacuum up until now.
• 40:45
  One thing that occurs to me as a potential entry point that's between the bludgeon we
• 40:49
  were talking about earlier of banning algorithms or banning personalization on the one hand
• 40:55
  versus the light feather of mere transparency on the other would be to change the business
• 41:01
  model of social media companies and people who love that.
• 41:04
  Play with the question of the only metric at the moment is the amount of time that you
• 41:09
  spend on the website.
• 41:10
  I mean, these are free platforms, which means they have to deal in your attention and sell
• 41:15
  it to advertisers.
• 41:17
  Is there a way of, so that you don't have to go around and play a game of whack-a-mole
• 41:21
  where you try to fix all of these problems that are emerging from a single cause which
• 41:25
  is that they have to hold your attention as long as possible and therefore make their
• 41:28
  products as addictive and radicalizing as possible so that they're sexy.
• 41:33
  Can you, how would you do, could you do that?
• 41:36
  Could you basically say like, okay, you're no longer going to be driven by time on site
• 41:40
  as the metric.
• 41:41
  Now it's going to be X or you have to have, I don't know, a paid system, I mean, not that.
• 41:46
  Yeah.
• 41:47
  What's possible?
• 41:48
  So one of the things, when I'm feeling more mischievous, the thing that I always try
• 41:55
  to get people to just imagine, like imagine if we did this is, you know, back when I
• 41:59
  was working at Pinterest, so I had a job very similar to what exists at Facebook.
• 42:04
  We were tweaking the algorithms, you know, as time we could get used to the site, how
• 42:08
  many as we could get you to click on, you know, this was maybe 2016, 2017, a more innocent
• 42:15
  time on it.
• 42:17
  And back when we thought Facebook could change the world in positive ways, you know,
• 42:23
  we pushed out a single change, this is just one change, we came in and said, hey, one
• 42:31
  of the challenges of picking content with computers is computers don't understand the
• 42:35
  concept of boredom, right?
• 42:37
  So if I'm picking out content and I'm like, I want to give you the thing you're most likely
• 42:41
  to save next on Pinterest, I'm going to give you a thing as identical to the last thing
• 42:46
  you did.
• 42:47
  So for example, if you pin a daisy, like a pillow with a daisy on it, I'm going to find
• 42:52
  all the white flowers I can for you, like all the pillows with white flowers.
• 42:57
  But the problem when you do that is people get bored.
• 43:00
  And so we went in there and we made a relatively simple model where we put all the content
• 43:05
  in the world into 6,000 buckets, not a ton of buckets, given how complex the world is.
• 43:09
  But we went there and said, hey, we're going to make sure we sprinkle content from different
• 43:14
  buckets in your feed.
• 43:15
  You're not going to get deluge in any given thing.
• 43:19
  And that one change increased global consumption by 6%.
• 43:23
  And for context, we have 300 million users globally.
• 43:27
  Like I sucked up lifetimes.
• 43:29
  What does that metric mean in the context of pins?
• 43:32
  It means how many pins did you look at?
• 43:34
  No.
• 43:35
  People spent 6% more time.
• 43:37
  Like we spent, we took whole lifetimes away from people from one change, right?
• 43:43
  But the interesting thing was while people looked at 6% more pins, Facebook, Facebook,
• 43:48
  Pinterest only made about 2.5%, 3% more revenue.
• 43:52
  So you can say, how did that happen?
• 43:54
  Like if you're looking at more content, like I thought we had this direct relationship,
• 43:57
  like you look at content, I get to show you ads, I get money.
• 44:00
  So why did people not, why did Pinterest not make 6% more money?
• 44:05
  And it's because the people who we impacted the most were people who were already consuming
• 44:12
  a lot of content.
• 44:13
  Like the most profitable second you spend on a social network is the first second you're
• 44:18
  there.
• 44:19
  My first minute you're there.
• 44:21
  The longer you're on there, each incremental minute the platform makes less and less money
• 44:25
  from you.
• 44:26
  Because they show you all of the ads they think you're going to most likely click on right
• 44:29
  up front.
• 44:30
  Like all their best ads are going to go right up front.
• 44:33
  And so there's this interesting question of if they make progressively less and less money
• 44:37
  off you.
• 44:38
  And we know that harm is at least partially correlated with people who get sucked into
• 44:43
  these platforms really intensely.
• 44:45
  But if you went in there and said, hey, there's some syntax, you know, we're going to tax
• 44:50
  you at a certain rate for a amount of content consumed, we're going to base it on whatever
• 44:57
  your platform is.
• 44:58
  And basically make it such that if you have this person who is beyond some cut off, you
• 45:04
  stop making money off them after they've been on there for an hour or two hours, something
• 45:07
  like that.
• 45:09
  And use that money to fund in-person community.
• 45:12
  That's community centers.
• 45:14
  It's places to dance.
• 45:15
  That's playgrounds for kids because right now these platforms are pulling in billions
• 45:20
  and billions, tens of billions, hundreds of billions of dollars, taking a small amount
• 45:24
  of that from disincentivizing the most severe addiction would change the way our communities
• 45:30
  worked.
• 45:31
  That's fascinating.
• 45:32
  You said a sin tax.
• 45:33
  I heard sin tax.
• 45:34
  But it's like a sin tax.
• 45:35
  Yeah, sin tax.
• 45:36
  Yeah, sin tax like we applied to alcohol or tobacco.
• 45:39
  That's interesting.
• 45:40
  We've got that.
• 45:41
  So everyone gets an hour a day on social media and then after that the company has to
• 45:46
  pay a 10% tax on the second hour and a 20% tax on the third hour and a 100% tax on
• 45:52
  everything beyond the third hour day or something.
• 45:56
  I like it.
• 45:57
  Because then platforms would come in there and say, hey, the thing that's most valuable
• 46:00
  to us now is you continue to use the platform six months from now, right?
• 46:04
  You've shifted the focus away from optimizing for the pure short term and made it so that
• 46:09
  if you really want to maximize the value of this user, retaining the user is now the
• 46:13
  most valuable thing.
• 46:14
  And when you look at how people churn off of these platforms, like I had to go and start
• 46:20
  rereading through the documents that I brought out of Facebook recently because I'm going
• 46:24
  to be participating in a lawsuit like that we're trying to bring justice to some kids.
• 46:31
  And I've never had to testify before, like non-acor case, because I'm sitting there rereading
• 46:35
  the documents and one of the things that I was shocked by and I was like, God, this
• 46:39
  just shows you how short-term thinking this organization is.
• 46:43
  They had a chart that showed if you were in the top tiers of consumption, like basically
• 46:50
  everyone else had the similar chance of still using the platform six months from now.
• 46:56
  But if you were in the top tiers of consumption, you got a real ding off that, like people
• 47:00
  could tell their lives are being harmed but they left.
• 47:04
  And right now, there's nothing that they get evaluated on a quarter by quarter basis
• 47:09
  in terms of how much money they make.
• 47:11
  There's no incentive in there to be like, we need to actually pair it back a little bit
• 47:15
  now so that we can have more users in the future.
• 47:18
  But this would create that incentive.
• 47:20
  It's totally logical.
• 47:21
  I mean, it makes sense.
• 47:22
  If a person drinks twice a week, then they're likely to still be drinking in six months.
• 47:27
  Then if they get up in the morning and drink a bottle of vodka, if they get up, they're
• 47:30
  probably not going to be drinking.
• 47:31
  Like something's going to give, you know, it's six months later.
• 47:34
  Same, you know, if you use cocaine once every three weeks, maybe you're still going to
• 47:37
  be using cocaine in six months.
• 47:39
  If you're using cocaine all day every day, you're probably not going to be using it again
• 47:42
  in six months time.
• 47:43
  Like something's going to happen.
• 47:45
  That's really interesting.
• 47:46
  Something's going to go.
• 47:47
  Yeah, exactly.
• 47:48
  That's right.
• 47:49
  One way or another.
• 47:50
  I mean, the other thing that interests me that your organization is doing is this idea
• 47:53
  of a simulated social network.
• 47:55
  What is that?
• 47:56
  Oh, you know, I wanted to do a similar social network one day.
• 48:00
  We currently don't have the resources to that.
• 48:02
  So you're just focused in on standard of care.
• 48:05
  So to give you a little bit of context.
• 48:08
  So when we teach physics, you know, people think, oh, sciences, like sciences are objective.
• 48:13
  You know, like there's one true truth.
• 48:15
  Like when we teach physics to anyone except for like graduate students or like later
• 48:21
  years of being a physics major, we teach with these simplifications that are not real.
• 48:27
  You know, we talk about spherical cows.
• 48:29
  We talk about springs that don't weigh anything that never heat up, no matter how much
• 48:35
  we move them, that moves through a vacuum.
• 48:37
  You know, we do simplifications, but those simplifications are still powerful enough
• 48:41
  that we build intuition.
• 48:43
  So for example, in the case of that spring that has no weight that never heats up, we still
• 48:47
  learn about the math that causes something to oscillate.
• 48:50
  Right?
• 48:51
  It's really powerful.
• 48:54
  When it comes to social networks, we don't need to have social networks that are perfectly
• 49:00
  high fidelity to what we see in real life.
• 49:02
  We need to have ones that are follow the general rules that we know of social software.
• 49:08
  And at this point in my life, I've worked on four social networks.
• 49:11
  I can tell you wholeheartedly, there are patterns that were peaked over and over and over again.
• 49:17
  You know, we can go in there and take a bunch of people, people, you know, make people
• 49:22
  simulate people. They're not super complicated.
• 49:25
  They have simple interests.
• 49:27
  They're basically numbers.
• 49:28
  I wanted to use emojis.
• 49:29
  You like all content on here is just emojis.
• 49:32
  People have emoji interests.
• 49:34
  Make it really simple.
• 49:36
  But it allows you to then go in there and say, let's try out different ways of distributing content.
• 49:41
  You know, we can simulate a little social graph of who's connected to him.
• 49:45
  We can simulate how much you like each of these people.
• 49:48
  We can simulate how likely are you to seek attention?
• 49:50
  How lucky are you to actually post?
• 49:52
  And we can go through and generate some of this with some synthetic data
• 49:56
  and make it such that you can teach a class where you can learn data science
• 50:01
  by going and looking at data that is very, very messy, very, very complicated.
• 50:06
  But it has fidelity for these generalized principles of social networks.
• 50:09
  And the reason why that's important,
• 50:12
  No, you were just about to answer the question, which was what is the utility?
• 50:17
  So the reason why this is so important is, you know, when I sit down with people and I say,
• 50:20
  you know, there's other ways of making a social network safer,
• 50:23
  other than choosing which ideas Ken or Ken not be said.
• 50:27
  Right? There's ways to safety that don't involve censorship.
• 50:31
  People look at me like I'm talking about how gravity is only optional.
• 50:35
  Right? People like, what do you mean?
• 50:37
  Like there's no way we can make it better.
• 50:38
  But I'll give you an example.
• 50:39
  Like Arturo Bayhara talked about the idea of like for bullying, right?
• 50:43
  Bullying is a great example where AI's are not going to understand what is bullying.
• 50:48
  They just don't like you, they can understand a few slurs.
• 50:51
  They can maybe understand a little bit tone.
• 50:53
  But like if you actually want to stop bullying,
• 50:55
  you would take down so much innocent content it would make things really dysfunctional.
• 50:59
  But if you come in there and say, let's, let's use this as a suggestion.
• 51:03
  You know, we don't have to be right for suggestions.
• 51:05
  We can be wrong lots and lots of time.
• 51:07
  But if you come in there and just say, hey,
• 51:09
  people have said that content like this,
• 51:11
  when they receive it, it hurts their feelings.
• 51:14
  About 50% of kids don't send that message.
• 51:18
  Right?
• 51:18
  The platform didn't decide what they could or couldn't say.
• 51:20
  That kid decided, am I going to actually,
• 51:22
  that gave them a moment of pause?
• 51:24
  Or you know, you were talking about this.
• 51:26
  Like should the algorithm be picking on content you receive or should other people?
• 51:30
  How does that play out?
• 51:31
  And we can actually do that if we have a simulation.
• 51:33
  We can teach people so they have intuition themselves.
• 51:36
  And they can go out and build better social networks.
• 51:39
  And is the simulation inhabited by bots for the purposes of the students and researchers being able to
• 51:46
  analyze it or is it, I mean, because I could also see the utility in having
• 51:50
  a simulation that is populated by some cohort of real users.
• 51:55
  They may be in turns or researchers themselves.
• 51:57
  But in order to understand the group dynamics of the way that people spawn to different incentives,
• 52:02
  different algorithms, when you tweak this, I mean,
• 52:05
  given that the social media algorithms are essentially a black box and we don't really
• 52:08
  understand what they're prioritizing or how, couldn't you also sell up your own social media,
• 52:13
  networks, small scale sort of scientific, I guess, AB.
• 52:18
  I think you would end up getting, so I'll take a step back.
• 52:22
  So one of the most important data sets that has ever been made available
• 52:27
  for studying how humans actually relate to each other.
• 52:29
  How information flows within a company is the Enron email database.
• 52:34
  So Enron was an energy company in the United States.
• 52:37
  They did some very naughty things like you'd say,
• 52:40
  how can an energy company kill people?
• 52:42
  People died in California because of how the power got shut out,
• 52:45
  shut out, how they made power surge and price.
• 52:48
  All kinds of things that were very, very illegal.
• 52:50
  But when they went bankrupt and their executives went to jail,
• 52:55
  the company email database got released as like an exhibit in the lawsuit,
• 53:00
  like you know, in the bankruptcy or in the criminal proceedings.
• 53:02
  And so researchers can actually go in there and see genuine people who have genuine
• 53:07
  motivations in genuine relationships, in genuine context,
• 53:10
  and understand how they relate to each other.
• 53:13
  When you go and put a bunch of interns in a room and say, talk to each other,
• 53:17
  you might be able to get 20, right?
• 53:21
  Like it's one of these questions about scale.
• 53:23
  That when we look at the phenomena of social networks,
• 53:25
  we're talking about somebody known as a long tail effect.
• 53:28
  So long tail effect says, you know, we often use the shorthand
• 53:32
  99 1% rule.
• 53:35
  So that means 90% of people say nothing.
• 53:39
  They just consume.
• 53:40
  They just scroll.
• 53:41
  They read 9% will do light actions.
• 53:44
  It's like a like comment and 1% of people produce all the content.
• 53:49
  So even if I had 100 interns,
• 53:52
  I wouldn't really have a social network.
• 53:54
  Well, to be fair, I was thinking more along the scale.
• 53:57
  I was thinking more along the scale of a drug trial where you'd have a thousand people.
• 54:00
  Oh, when you're a thousand.
• 54:01
  But sure, I mean, yes, you're right.
• 54:03
  You're not going to replicate perfectly 100 million people.
• 54:06
  But even in the skill of like 5,000 people, right?
• 54:08
  Like you have a situation where where, you know,
• 54:11
  the 5,000 people you recruit are going to be much more involved than real people are.
• 54:15
  Right?
• 54:16
  Like when I talk to people about like, where do the dangers lie?
• 54:19
  Or where the dangers come from?
• 54:20
  One of the first analyses I ran when I was at Facebook was we went and looked at
• 54:27
  well, how was voice shared amongst people who participated on Facebook?
• 54:32
  So like voice is known as, so example, like if I do a post,
• 54:37
  I and is seen by 100 people, I have 1% the voice of someone who writes a post
• 54:44
  and is seen by 10,000 people.
• 54:47
  And when you go look at it, the United States came out as the most like little
• 54:52
  de-democratic country in the world because about 14% of people made up 80% of the voice.
• 54:58
  Like there was actually a good number of voices there, like 1 in 7 people made up 80% of the voice.
• 55:03
  When we looked at other countries around the world, 1 to 2% of people made up 80% of the voice.
• 55:10
  And my theory on that is that Facebook existed without algorithms for a long time in the United States.
• 55:17
  Right?
• 55:17
  That meant when they started putting an algorithm in place, it knew about a lot of people.
• 55:23
  One of the biggest challenges when writing algorithmic systems and we face as a Pinterest
• 55:27
  was when you have a new voice, how do you know how much audience to give them?
• 55:32
  Right? It's really hard.
• 55:33
  And I think at the time that I was at Facebook, you had in other countries because it rolled
• 55:37
  out so much later, like you rolled out, got popular after there were algorithms.
• 55:42
  Those only a very small number of people were there to bootstrap those systems.
• 55:47
  And they ended up getting much, much larger audiences.
• 55:50
  Fascinating.
• 55:51
  So just to clarify, assume that you've got a simulation, it's populated with bots,
• 55:57
  and you're then kind of analyzing and studying it.
• 56:00
  What's the best case scenario in terms of what you would get out of this thing?
• 56:03
  Suppose we could give it infinite funding.
• 56:06
  I think the way I think of it is like right now, the way that we,
• 56:12
  there really aren't college classes around the United States on designing social networks.
• 56:16
  Like if we were to look at architecture, like how we design a space, like you know,
• 56:20
  a church is echoally, it makes you feel small, right?
• 56:24
  It makes you, it motivates you to be quiet or motivates you to sing and you know,
• 56:28
  some of others because of how the space is designed.
• 56:30
  Versus a cafe is designed such that we can have conversations that it encourages us to speak,
• 56:35
  or like a club is allowed so that we go and get close to each other in a whisper to
• 56:38
  or like yell at each other, right?
• 56:40
  We spend so much time trying to figure out how to design our physical spaces to create
• 56:46
  the relationships and interactions that we want to spend our lives doing.
• 56:49
  You don't have schools of architecture for social media platforms.
• 56:53
  An imaginary world would sit down in the past.
• 56:56
  I just assumed that people in Silicon Valley are spending a lot of time analyzing
• 57:00
  precisely what shade of blue on the Facebook logo makes people spend more time on site.
• 57:06
  I just assumed that the problem was that their incentives are all aligned
• 57:10
  simply to capturing my attention rather than making the space conducive to
• 57:15
  same conversation.
• 57:16
  So one of the challenges is that all these patterns that I'm talking about because you don't,
• 57:22
  let's say, let's talk about SpaceX.
• 57:25
  If you show up at SpaceX to go build rockets, to go to Mars,
• 57:29
  you come in with at least five years of graduate education in some kind of
• 57:34
  aeronautical engineering that teaches you like this is what happens when a rocket fin is this
• 57:39
  shape, when you make the engine this shape.
• 57:41
  This is what happens when these two gases are combined.
• 57:44
  When people show up at Facebook, they show up blind.
• 57:47
  They start from zero.
• 57:49
  And one of the challenges is when I was at Facebook,
• 57:52
  the last of the people who really understood the whole system were leaving.
• 57:57
  Right, like these are the people who because they had been there at every step of the way,
• 58:00
  they knew where the bodies were buried, they knew what kind of work didn't work.
• 58:03
  Like it was super fascinating for me getting to read the documents of these people,
• 58:08
  because you could tell that the number of voices was diminishing.
• 58:11
  And like the ability to actually have conversations with all the context, it was going away.
• 58:17
  And for context, like, you know, places like Google go through these phases too.
• 58:21
  Like any large company, it gets more and more complicated over time,
• 58:25
  and you likely only work on a tiny slice.
• 58:27
  So you start from zero, you work on a tiny slice.
• 58:31
  It makes it really hard to have these larger interaction conversations,
• 58:37
  like what do we actually want to build?
• 58:40
  Because there isn't a systematized way to even learn about it internally.
• 58:44
  And so what ends up happening is you do run the 50 shades of blue experiment.
• 58:49
  You don't run the, what if we re-architect it in a really novel and different way?
• 58:54
  Right, right, because nobody is the rigid Feynman to physics.
• 58:58
  Nobody is like a general practitioner in medicine.
• 59:01
  Nobody is like an all purpose.
• 59:03
  So Feynman is a great example.
• 59:06
  So for context for your listeners, Feynman was a physicist who became famous,
• 59:09
  because he actually could explain to people how physics works.
• 59:12
  He explained to them the beauty of the world that we live in with numbers.
• 59:17
  He got to launch into physics with probably an undergraduate physics degree.
• 59:22
  So four years, probably a PhD, five to six years.
• 59:26
  And then he got to spend 30 years becoming more and more, you know, steeped in the beauty of the universe.
• 59:33
  I love social software.
• 59:36
  Like I've said this repeatedly that if I could go work at Facebook again,
• 59:39
  if I could work at TikTok again, I would.
• 59:41
  Because these systems shape our reality.
• 59:43
  And it's just beautiful.
• 59:45
  Like getting to watch people interact with each other at the scale,
• 59:48
  there's no social science that is set in the world.
• 59:52
  The problem though is the vast majority of people who work at Facebook have been there
• 59:56
  for two years or less.
• 59:58
  Right, like when I left, my senior already in the company was like 75%
• 1:00:02
  75% tell or something wild like that.
• 1:00:04
  And I'd been there for two years.
• 1:00:05
  And that's because in Silicon Valley, if you don't hop from company to company to company,
• 1:00:09
  you actually end up making substantially less money, substantially less money.
• 1:00:13
  And so you have this challenge where people start from zero.
• 1:00:16
  They don't have a lot of context.
• 1:00:18
  They probably don't stick around for that long.
• 1:00:19
  And I want to add one more problem, which is we said before,
• 1:00:22
  there's not one social network.
• 1:00:24
  There's many social networks.
• 1:00:26
  One of the things that was brought up in the recent antitrust trial,
• 1:00:31
  so just for context, again, for listeners,
• 1:00:33
  at the FTC, the Federal Trade Commission,
• 1:00:36
  in the United States has sued meta.
• 1:00:37
  Then the middle of a giant court case,
• 1:00:39
  it's really like soapy and like dramatic,
• 1:00:42
  like they've interviewed Mark from multiple days.
• 1:00:45
  Cheryl Sandberg, like it's chef's kiss.
• 1:00:48
  Just fantastic.
• 1:00:51
  In the government's published a bunch of their exhibits,
• 1:00:53
  you can like look through 100 pages of emails between these seconds.
• 1:00:57
  It's the Blake Live Blake court case for now.
• 1:00:59
  Oh, oh, it's so good.
• 1:01:01
  It's so good.
• 1:01:03
  But you look at how they treated even Instagram after they bought it.
• 1:01:11
  They hoarded the technical talent for designing those algorithms
• 1:01:15
  in the main company, like they intentionally
• 1:01:17
  understaffed it so that if it ever got spun off,
• 1:01:19
  it wouldn't have intellectual capital.
• 1:01:21
  You look at Pinterest.
• 1:01:22
  Pinterest had 300 million users.
• 1:01:24
  And my team had maybe 15 people, maybe 15 people,
• 1:01:30
  across the whole kind of like quantitative area,
• 1:01:33
  picking out all the content that everyone got to see in the world on Pinterest.
• 1:01:37
  And so it's one of these challenges of, you know,
• 1:01:39
  we assume there is this infinite pipeline of talent,
• 1:01:42
  this pool of talent, but literally each one of these companies
• 1:01:45
  is reinventing things from scratch over and over and over again.
• 1:01:49
  And the idea that we could be so casual
• 1:01:53
  about designing the spaces that, you know,
• 1:01:54
  the average kid in the United States spends like three and a half hours
• 1:01:58
  a day online, average, right?
• 1:02:00
  So that means like the 70th percentile is probably like five hours, right?
• 1:02:05
  The idea that we are so casual about the design of these spaces
• 1:02:09
  is something we really should think and ponder on.
• 1:02:12
  How do you think about free speech, Frances?
• 1:02:14
  A lot of this gets framed around that sort of, you know,
• 1:02:19
  I would call it a punchline, I suppose.
• 1:02:21
  You know, Elon Musk bought Twitter ostensibly to reintroduce free speech to that platform.
• 1:02:28
  It's certainly a free place to say certain kinds of things and a less free
• 1:02:32
  and less conducive place to say other kinds of things.
• 1:02:35
  When you and I were on television together in Australia,
• 1:02:38
  I felt like that was a moment when the conversation was,
• 1:02:42
  I guess being portrayed in a fairly crude binary where I was like
• 1:02:46
  cast as a person who's supposedly really pro free speech,
• 1:02:49
  which means that anything goes and then other people are like,
• 1:02:52
  but hate speech, you know,
• 1:02:53
  and the Australian government is like,
• 1:02:55
  we need to impose laws to decide what people can say.
• 1:03:00
  Oh, yeah. How did I go?
• 1:03:02
  Did you guys actually end up passing?
• 1:03:03
  No, that's a questionable loss.
• 1:03:06
  They weren't thinking about that.
• 1:03:07
  That law, which was not as bad as,
• 1:03:09
  ultimately I did a lot of reporting on that that I hadn't done prior to
• 1:03:12
  our appearance together.
• 1:03:13
  And it was not as bad as it seemed to be.
• 1:03:16
  It was not going to allow the Australian government to police what people could say.
• 1:03:19
  But it was going to allow the Australian government to punish American tech companies
• 1:03:23
  that failed to report on how they were moderating what Australians say.
• 1:03:29
  It was perceived as being a sloppy backdoor way.
• 1:03:32
  But it was really about transparency of algorithms.
• 1:03:34
  It was mostly about trying to compel.
• 1:03:36
  It was a dog's breakfast of a law, as we would say.
• 1:03:39
  It was a hard podge.
• 1:03:40
  It was ultimately killed in the Senate,
• 1:03:41
  so it passed the House of Representatives,
• 1:03:43
  a few independent senators,
• 1:03:45
  and some greens, I believe, shot it down.
• 1:03:48
  So that probably puts an end to that level of reform.
• 1:03:53
  Nonetheless, it strikes me that you have a much more nuanced and kind of a third way
• 1:03:58
  of thinking about this than to see it as a battle between a free-for-all and censorship.
• 1:04:08
  So we talked about before this idea of power loss.
• 1:04:12
  So a power law, the 90 just watch 9% toggle things, do a like, do a comment,
• 1:04:18
  1% create the content.
• 1:04:21
  Even within that 1%, it again follows another power law.
• 1:04:25
  So the vast majority of people in that 1% post maybe once a week, once a month.
• 1:04:31
  A much smaller fraction post every day, or maybe multiple times a day.
• 1:04:34
  And the thing that I think is really unfortunate about how we have kind of,
• 1:04:39
  it comes again, like you know what we said before, we don't have a good,
• 1:04:43
  popular understanding of what's even possible or what's available to us.
• 1:04:47
  Because we don't have access to the idea that there are a bunch of different
• 1:04:51
  leveraging and poem, the thing people come to first is why don't we just take the bad thing down.
• 1:04:58
  And we spent a while earlier talking about the idea that like,
• 1:05:00
  AI's don't know what we're talking about.
• 1:05:02
  Like they can't figure out gradations, they don't understand jokes,
• 1:05:05
  they don't understand, uh, implicate like, uh, when you apply something like,
• 1:05:10
  if you actually want to remove the bad things, you're going to take down lots and lots and lots
• 1:05:15
  and lots of good things. Like that is not a realistic path forward. But it is a realistic path forward
• 1:05:20
  is to say, who is pushing ideas? And one of the things that I think has been really
• 1:05:26
  obviously in the United States recently, I don't know if this led over into Australia.
• 1:05:30
  But in the last week or so, there has been these viral videos across a bunch of different social
• 1:05:35
  platforms of Chinese factory people exposing Western companies for like how they made things.
• 1:05:41
  And they all came out of nowhere and they're everywhere all at once. And when you look at these
• 1:05:45
  things, you have to take a step back and say, how did this happen? And the reality is there's a
• 1:05:51
  lot of people out there that are trying to manipulate how the algorithms function.
• 1:05:56
  Like we can take for granted the idea that people say hateful things. We have to take one step
• 1:06:01
  back and say, why is it some hateful things seem to go everywhere and other hateful things don't
• 1:06:05
  go everywhere? Like it's not a random set of things. One of the things that we saw during
• 1:06:11
  the referendum you guys had around Aboriginal rights was there were transparency folks,
• 1:06:16
  there were actively monitoring that situation. And a huge fraction of the loudest voices that
• 1:06:21
  were happening were pushing narratives in Australia, they didn't live in Australia. They lived
• 1:06:25
  in the south of the United States, right? Or there was a giant, probably the largest social media
• 1:06:31
  facilitated genocide to date. It's about half a million people died in Ethiopia a couple of years
• 1:06:35
  ago. Probably never heard of it. Half a million people died. That wasn't even run by a state. It was
• 1:06:41
  run by a bunch of people who left Ethiopia who were living abroad who didn't like the ruling ethnic
• 1:06:45
  group and just started slamming them continually online. So where am I going with all this?
• 1:06:52
  Computers are good at detecting behavior. They're bad at understanding meaning. So what is that?
• 1:06:59
  Again, let's unpack that again. If I see that you are not a real person, you don't have friends,
• 1:07:06
  you don't message back and forth of people, you don't send emojis to your these friends. They
• 1:07:10
  all send stuff back to you. You are not a person, but you seem to be talking a huge amount and
• 1:07:15
  pushing out stuff. We have to ask questions about, is that how we want our platforms to work?
• 1:07:23
  What are the obligations of companies to watch from when people manipulate platforms? Because
• 1:07:30
  if we just did a better job at that one thing of going in and saying, hey, if you're coming in
• 1:07:35
  here to just store the platform intentionally, we're just turning you down. You better be really,
• 1:07:42
  really compelling because we're going to turn the volume down a little bit. It would completely
• 1:07:47
  change the tone. Why not make social media platforms only for human beings? I love that too.
• 1:07:56
  I'm saying there's a broad continuum here of things that we could step in and say,
• 1:08:02
  we see that there are foreign actors. We see that there's interference.
• 1:08:05
  Facebook, TikTok, Instagram, Twitter, you have obligations to tell us exactly what level of
• 1:08:13
  investment you're putting into these things. Because one of the things that I documented when I was
• 1:08:17
  there was I worked with in threat intelligence. There was a team that was doing this. They were
• 1:08:23
  looking at behavior, they were looking at content. They were looking specifically for people weaponizing
• 1:08:27
  the platform for influence operations. I was in the sister team that looked at ironically
• 1:08:33
  counter espionage. This is things like foreign actors intentionally sending malware to you because
• 1:08:39
  you're a part of a dissident ethnic group, the Wiggers and you're living in Bangkok. If I can get
• 1:08:46
  you to click on this malware, I can see all your photos, I can figure out where all you hang out,
• 1:08:51
  I can see your geolocation, we went and looked for the people who were doing that.
• 1:08:58
  The thing I saw across all those teams, the people who were searching for the child predators,
• 1:09:03
  looking for the people who were doing scams, looking for the people who were facilitating
• 1:09:08
  the growth of drug cartels and terrorists, was if they had a tiny amount of more help,
• 1:09:15
  like we're talking like five engineers, ten engineers, we would have taken 80% of the problem off the
• 1:09:21
  platform. We intentionally didn't get that. There's things where you could imagine passing a
• 1:09:27
  loss saying, hey, you have to disclose the size of these teams. You have to let us directly pull
• 1:09:34
  them for whether or not they feel adequately supported. There's things like this where we could go
• 1:09:39
  and set it in place, systems of incentives, where we could say, you can't let people abuse your
• 1:09:45
  systems, because it steals the opportunity from real people to talk to each other.
• 1:09:49
  There's over-to-beuse like that, the installation of malware, hoodwinking people. There's the case
• 1:09:57
  of bots just flooding the zone with hostile nonsense. What about the edge cases that are the ones that
• 1:10:04
  seem to become the most controversial? This came up during the Australian attempt at legislation
• 1:10:09
  where there were provisions about holding social media platforms accountable for misinformation,
• 1:10:15
  disinformation, hate speech and things like that. Let's exclude bots for a moment, just talk about
• 1:10:19
  human beings. One of the things that people like me who are regarded as being on the free speech
• 1:10:25
  side of things in Australia would say, well, what exactly constitutes misinformation? What
• 1:10:29
  constitutes hate speech? It's why the computers can't do it. If I say that we should have been
• 1:10:34
  talking years ago about the integration of Muslim migrants in the United Kingdom,
• 1:10:39
  is that a form of hate speech in xenophobia and Islamophobia or is illegitimate part of the
• 1:10:44
  national conversation? The government would come back, the Safety Commissioner, and say things like
• 1:10:49
  there are clear cut cases of speech that we want to ban on social media, for example,
• 1:10:54
  telling people that the polling place that they were going to go to on election days closed.
• 1:10:59
  That, to me, strikes me as, yes, despite my free speech by una fides, I would be happy to ban
• 1:11:08
  someone from posting that my local polling place is closed when it's not, because that strikes me
• 1:11:13
  as election interference. Do you have any insight into how we deal with those edge cases? Does AI
• 1:11:20
  provide any prospect of assisting us in making those adjudications? How do we manage misinformation
• 1:11:27
  and disinformation when we all live online? I think there's always exceptions that prove the
• 1:11:34
  rule. I'll give you an example. Child abuse material. Child abuse material, so this is people taking
• 1:11:40
  inappropriate photos of kids by them online. That's one of the great examples of where
• 1:11:47
  content moderation actually works pretty well. Because we have been fighting porn in general
• 1:11:52
  for the last, what is it, 30 years? We've been trying to take porn off of public spaces
• 1:11:58
  since we had an internet. Because we have been detecting pornography for so long, and we can
• 1:12:02
  find kids, we're pretty good at finding child abuse material and just not allowing that to be
• 1:12:08
  distributed, and we're okay when the computer is raw. That's one of these exceptions that prove
• 1:12:14
  the rule. I think there is a world where you could go through and enumerate publicly and say very,
• 1:12:20
  very cleanly, election interference, as a thing that interferes with our ability to have a
• 1:12:24
  democracy. Here are the specific examples of things that are not okay, or you have to give us a
• 1:12:30
  feed of content that you think is election-related on the weeks leading up to the election,
• 1:12:35
  and in a public way that can be adjudicated, using these criteria, we can say what kinds of
• 1:12:41
  means are things that need to be taken down. That's the thing that you can do with relative
• 1:12:45
  precision, but you can't do a blanket category, like I say, you can't have hateful content because
• 1:12:51
  what does that mean? Or you can't come in there and say you can't have misinformation. The reality
• 1:12:55
  is those influence campaigns that foreign interference I talked about before, they don't even care
• 1:13:00
  of stuff's false. They care if it's divisive. You know, you can have true facts that manipulate people,
• 1:13:06
  and if you go out there and intentionally find the wedges to divide society, you can have serious,
• 1:13:10
  serious consequences. So it's more of a question around how are our systems designed,
• 1:13:15
  what are the incentives to get better systems, and how are we able to hold platforms accountable
• 1:13:20
  to move in that direction? Francis, from your lips to God's ears, it's great to talk to you.
• 1:13:25
  I'll ask thanks so much for being on the show. It's hasn't any. Happy to be here.