Loading summary
Transcript107 lines
- [00:01]
A
Thank you for calling the Bombas Comfort line. Bombas make socks, slippers, tees and underwear made with the highest quality materials. Press 1 for comfort, 2 for style, 3 for donation. You chose style. Bombas is styles for whatever you enjoy. You can run in Bombas, lounge in Bombas, dress them up, dress them down, but always give back in Bombas because with every item purchased, another is donated Bombas comfort. Worth calling for. Go to bombas.com audio and use code audio for 20% off your first purchase. That's B O-M-B-A-S.com and use code audio.
- [00:33]
B
This episode of the Times Tech Podcast is sponsored by ServiceNow.
- [00:36]
C
Danny One thing we keep hearing from business leaders right now is AI sounds great, but how do you actually make it work inside a company?
- [00:45]
B
Exactly. Because most organizations aren't neat, shiny systems. They're layers of software, legacy tech and teams all doing things slightly differently.
- [00:54]
C
ServiceNow sits across all that, acting as a control tower for making work move seamlessly through the organization, connecting people, systems,
- [01:02]
B
data, and increasingly AI agents so that things don't happen in silos.
- [01:06]
C
Learn how ServiceNow puts AI to work for people@servicenow.com Good afternoon, my name is
- [01:13]
D
Mike and I will be your conference operator today.
- [01:16]
B
At this time I would like to welcome everyone to the Facebook Results Earnings conference call. On May 3, 2017, Mark Zuckerberg held a conference call as CEO of the world's biggest social network. This is meat and potatoes. Every three months, Facebook unveils its financial progress and Zuckerberg talks about it with Wall street and the press. But on this call, he also slipped in a big announcement. Facebook was starting to get a lot of heat about the posts that were making it onto the platform. Suicide videos, terrorist recruiting, hate speech. So he called in some backup.
- [01:55]
A
Facebook's Mark Zuckerberg has announced that they will be hiring 3,000 new employees to help combat violence that people are streaming onto Facebook Live.
- [02:05]
B
It wasn't enough six months later on November 1, 2017 first today, there are 10,000 people working at Facebook on safety and security. Across our security, product and community operations teams. By the end of 2018, there will be more than 20,000. January 30, 2019 the most important work
- [02:26]
D
here is to keep executing our roadmap
- [02:28]
E
to build systems that can proactively identify harmful content so we can act on it Sooner.
- [02:33]
B
We ended 2018 with more than 30,000
- [02:36]
E
people working on safety and security.
- [02:41]
B
Keep in mind, at the start of 2016, Facebook had 12,000 workers total. This was a hiring spree of every epic proportions. And Facebook is not alone. As the web has expanded, so too has a whole new shadow class of worker content moderators. They are the human filters, charged with seeing and stopping the worst things you could imagine from ever getting to someplace you or I might see. But who is doing this work? Under what conditions? And why do we hear so little about them? Stay with me and we'll go inside this world of the millions of shadow workers who are manning the digital barricades, keeping the web safe for the rest of us, often to their own detriment.
- [03:32]
D
Fifteen years ago, there was no one doing anything like this. And now on various social media platforms, whether it's Facebook or Instagram, it's essential. And what they're exposed to is truly horrific Imagery, repeated imagery, over and over. That's your job. Nine to five.
- [03:58]
B
I'm Danny Fordson, west coast correspondent for the Sunday Times, and this is Tales of Silicon Valley, an eight part documentary series on the tech industry from the inside. Episode 8 Ignore, delete, escalate.
- [04:17]
A
Hello.
- [04:18]
B
Hello. Hello.
- [04:19]
F
Hi.
- [04:20]
B
Marlena Griffin is 37 and lives in Corpus Christi, Texas. She and her husband have a 13 year old son who has special needs. The care he requires means that they need flexibility. A 9 to 5 would be very difficult. So they both work as content moderators, not for Facebook, but as freelancers. And they find most of their work on Mechanical Turk or Mturk, the world's largest marketplace for content moderation and other quick episodic jobs. It's owned by Amazon, of course.
- [04:55]
F
My husband was looking for ways to make money online and he found something on Reddit that linked him to Amazon Mturk. And that's where he started and he told me about it. And still most of my work is through Amazon Mturk. But I've also branched out. Other sites have popped up doing the same thing.
- [05:18]
B
Most days start out the same way.
- [05:21]
F
I open my dashboard, I open their list of available hits. That stands for Human Intelligence Task.
- [05:28]
B
Human Intelligence tasks Hits. They can be almost anything. Analyzing the sentiment behind a tweet, taking part in academic research, even playing games.
- [05:40]
F
The. The other day I did a hit Solve a murder mystery with a few other people. It's possible they were other workers. It's possible that that was just AI. I was told they were other workers. And you know, I, I didn't stop long enough to really try and figure out, you know, what are they trying to get from this data.
- [06:01]
B
But as often as not, the job involves sifting through the worst of humanity.
- [06:05]
F
Child pornography, gore and Death. Yeah. I have to. I have to walk away from my computer.
- [06:12]
B
Sometimes it is grueling work, and often it pays literally pennies. Jobs are doled out in micro contracts and are under high pressure. As soon as you click yes, the clock is ticking.
- [06:26]
F
And these hits have timers, and so it. It's harder to do when I can't give my full attention to the hit. Sometimes it's like one click and then that hit is over. And that will be. Maybe it'll pay a cent, 5 cents, 10 cents. And then sometimes it's five images per hit, sometimes more. And those will maybe pay like 25 cents for five or 10 images. A lot of workers on MTURK won't work for less than I think they say 10 cents a minute, which is six an hour.
- [07:04]
B
For what it's worth, the federal minimum wage in America is $7.25. Griffin lives with family. The money she and her husband make is enough to make ends meet, but wouldn't be if they had to cover rent as well. The flexibility, though, is key enough, she says, to counterbalance the downsides.
- [07:25]
F
There was a hit. It was funded by the Department of Defense. And I'm not sure exactly what qualified me for that. I don't know if it was my mental health or my past. I've experienced trauma. But they showed images of actual gore and crime scenes and death. A lot of pictures from what I think were war, just aftermath. In between these images, sometimes there were flashing images and they wanted to see responses. Some of these images were of those. And then also with them, sometimes they would flash images of, like, puppy dogs and flowers and cats. Was very strange.
- [08:19]
B
Millions of people around the world, be they in Corpus Christi or Bangalore or Manila, are doing this every day, every hour, every minute. An army of anonymous people working in their living rooms, kitchen tables or featureless office parks. Scrolling and clicking.
- [08:36]
F
I'm a guinea pig, a mental guinea pig. And that term has been used by workers like me ever since this work has existed. That's exactly what we do. That's what we are. We're. We're guinea pigs.
- [08:53]
B
And a quick word about MTurk, which is the beating heart of this world. Jeff Bezos came up with the idea back in 2005. Amazon was booming, but it was having trouble with quality control. People would leave typos in their book reviews. Workers would input erroneous product information. Amazon software simply wasn't up to the task. So Bezos outsourced the job of keeping the website tidy. He set up a marketplace for gig workers before that was even a term we would recognize. His timing was fortuitous. Facebook was founded the year before, in 2004, YouTube launched in 2005, and Twitter 2006. It was the start of those three companies that really marked the dawn of the modern Internet. The web of user generated content. All of a sudden anyone could share a cat video or a taco recipe or an execution. The arrival of the smartphone supercharged this dynamic. Today, everyone, literally everyone with a smartphone is a publisher, which is great. It is also a problem because the billions of pictures and videos and posts that we upload every day need a filter.
- [10:14]
C
New Zealand's Prime Minister Jacinda Ardern said this is one of her country's darkest days.
- [10:20]
A
There is no place in New Zealand
- [10:22]
C
for such extreme acts of unprecedented violence.
- [10:26]
B
I've watched that 17 minute video that that gunman, he streamed live on Facebook, and it is one of the most horrendous things that I have ever seen in my entire career. It's something that nobody wants to see.
- [10:41]
E
And in March, police say at least 40 people watched the gang rape of
- [10:45]
D
a teenage girl as it happened on Facebook.
- [10:47]
B
The joke used to be that Mturk was a way for college kids to make beer money, but it has morphed into something else entirely. It's a bazaar for the web's dirty work, where anonymous digital janitors log in and clean up the Internet 10 cents at a time. But whereas MTurk is a marketplace for mercenaries, big companies often need their own standing army to clean up their content. And in America, no one has amassed a bigger army than Facebook.
- [11:19]
D
This is a class action. Correct.
- [11:22]
B
Steve Williams is not to be trifled with. He's a trial lawyer who specializes in suing big companies. One case he's working on is against Mylan and Pfizer, the pharmaceutical giants. He claims that they are fixing the price of EpiPens, a life saving treatment for extreme allergic reactions. If he wins, damages could run and into the billions. But perhaps even more eye catching than his EpiPens fight is the case he is leading against Facebook. Last year, a woman named Selena Scola got in touch. Scola is from the San Francisco Bay area, and in 2017, she was hired by a company called Pro Unlimited. Remember all of those announcements that Facebook made about hiring tens of thousands more people to filter content? That number included engineers, data scientists, policy experts, and about 15,000 content moderators, most of them contractors. Pro Unlimited was one of the companies that Zuckerberg drafted in to staff his army. Scola's first day was June 19, 2017, a Wednesday.
- [12:37]
D
And I'm not sure anyone could really have a sense until they start doing this job of what they're going to be exposed to. And what they're exposed to is truly horrific imagery, repeated imagery over and over. That's your job 9 to 5 or 9 to 6 every day to see beheadings, rapes, child sexual abuse, animal cruelty, all manner of things.
- [13:03]
B
Over the next nine months, she showed up to work at Pro, which, despite being an independent company, had its office on Facebook's campus. Over that time, according to her lawsuit, she was exposed to a catalog of horrors. The experience left her with severe Post Traumatic stress disorder, which she claims can now be triggered by the subtlest of cues, like touching a computer mouse or entering a cold building. Two other contractors who worked over a similar time period have joined the suit. They filed their claim in September 2018. No one would argue that the above doesn't sound dystopian. But is Facebook to blame legally? Williams argues that it is, and he frames the case as a question of workplace safety. He says Facebook is asking people to do a job without providing the support the job needs.
- [13:59]
D
These people, like Ms. Scola and the others who are in the position of Ms. Scola, are the filters to look at this, identify it, and keep it from appearing. But you know, it's sort of this horrific dystopian vision of people sitting in rooms with monitors, constantly flashing just truly awful imagery at them so that they can be the ones who are taking the step to get it off of Facebook, but not themselves being protected from the harm that they're suffering by seeing these things over and over again.
- [14:31]
B
Amid an avalanche of complaints, Facebook has been racing to revamp its approach. The company claims that its moderators can now avail themselves of services, including round the clock on site counseling and tools that allow them to blur upsetting content on their screens. In May 2019, it raised moderator wages. But Facebook is not the only company in this bind. Any social media platform you could think of, YouTube, Twitter, Instagram, Snapchat has human beings doing the same thing. Scola's case could have huge implications for how those people are treated.
- [15:10]
D
Since we filed this complaint, we've been contacted by content moderators throughout the world. In Germany, in the Netherlands, in Ireland, in Japan, in India, in the Philippines, in Malaysia. And their experiences are no different. So I have done cases involving medical monitoring, and that's really what's at issue in this case, which is this concept that California recognized first in a toxic tort case. This was a case in which people were exposed to harmful chemicals in their drinking water and didn't necessarily show illnesses yet. And the question became, well, you're not hurt. But what the court in California decided was if someone violated the duty that they owe you as a matter of law, as a matter of negligence, and placed you at risk of that harm happening, there should be an entitlement to some procedure that would permit you to be monitored and have access to medical care to protect yourself at the earliest point from one, hopefully the harm happening, and two, if it does happen, getting you treatment as quickly as possible. We're taking that model from a chemical exposure and we're applying it here in the instance of a psychic exposure.
- [16:27]
B
For American football fans, this argument may sound familiar. For years, the NFL, America's biggest richest sports league, claimed that it was not responsible for former players who were suffering debilitating brain injuries, which researchers eventually proved could in fact be traced back to the hits they sustained during their careers.
- [16:48]
D
I think that is very analogous. People in the NFL are getting these injuries. Some it manifests in horrible ways and we've seen the suicides and some seem to go on and never really manifest those harms. That doesn't mean they weren't exposed to an at risk for those harms.
- [17:05]
B
In that NFL case, the league agreed in 2017 to hand over a billion dollars to help cover medical bills for former players and their families. Williams is arguing that Facebook should set up a similar mechanism for content moderators, a system to care for people when the alleged effects of this difficult work catches up to them. Whether it's A week or 10 years later, Facebook is fighting tooth and nail. Last year they filed a motion to dismiss the case, and the two sides are currently fighting over Facebook's demand for detailed medical records, which could prove the alleged trauma may have been caused by other events in the plaintiff's lives before they became content moderators. And the law firm that Facebook has chosen to lead this charge, Covington and Burling, the giant Washington, D.C. outfit that for years advised the NFL in the concussion case. In June 2019, an investigation by the Verge website revealed Dickensian working conditions at the site of a Facebook contractor in Tampa, Florida. Florida people reported being put under immense pressure to rattle through hundreds of pieces of content per day, and drug abuse was rampant. Workers were given two 15 minute breaks, a half hour for lunch, and nine minutes of wellness time per day. Nine minutes. What is happening here? I went to Facebook's sprawling campus in Menlo park to find out. The first person I met Was Jeff King, the engineer in charge of building the systems to catch and route problematic content. He's got a hard job, not least because of the sheer numbers. Facebook, remember, has 2.7 billion users.
- [19:00]
E
We've got thousands of reviewers looking over essentially billions of pieces of content. We filter that down into things we can automate and then things that require a human to look at to make a decision. And all of that technology essentially works together to help improve really the experience overall is what we're looking for.
- [19:18]
B
Most of the horrible stuff posted by terrorists and racists and murderers, no one ever sees it, is caught by machine learning algorithms, which are trained to automatically cull images or text which cross the line. It's the stuff that falls in those gray areas which get sent to humans for the final say.
- [19:38]
E
It's actually incredibly cool how much progress we've made where it's like 96% of the nudity is detected and deleted automatically. 99% of child nudity and terrorism propaganda is detected and deleted automatically. Graphic violence has gone from the mid-50s to like 85%.
- [19:57]
B
The numbers may sound impressive, but dig deeper and a different picture emerges. And how many violations do you have in a day?
- [20:05]
E
Millions of violations a day. Let me see if I can quote this off the top of my head. I'm not sure. I think the prevalence of adult nudity on Facebook is 10 basis points of all views on Facebook a day. That's a reasonably small amount, Right? That's manageable. Things like hate speech are higher prevalence. I think closer to 30 basis points.
- [20:26]
B
Let me translate that. 30 basis points is 0.3%. So for every billion pieces of content uploaded to Facebook, 300,000 of them are hate speech that find their way past the filters. This is why Zuckerberg went on his hiring spree.
- [20:44]
G
This is what kind of a reviewer would see. So this is a piece a comment
- [20:48]
B
was reported in between meetings. Facebook press officer Carolyn Glanville showed me the filtering tool that Jeff king had built. Yeah.
- [20:56]
G
And we have blocked out, like, different things.
- [20:59]
A
Yeah. So this is like an action that they can take.
- [21:02]
B
They can say, so what are their options there?
- [21:04]
G
It's ignore, delete, escalate are the main ones.
- [21:09]
B
What is hard to square is that if people are the last line of defense, if safety is such a priority, why is Facebook leaving it to a bunch of contractors? The company is already facing the threat of being held liable for the content which appears on its pages. Why not keep this vital operation in house?
- [21:29]
A
Yeah, I go back to actually some of the scale we talked about in the beginning. So we have 50 languages to support. We have 20 different sites across the world we're using to help scale this program.
- [21:41]
B
That's Abigail Soy, one of the executives who helps manage the 30,000 people charged with keeping Facebook a clean, family friendly place.
- [21:50]
A
In many cases, we really do need that mixed workforce to help us do this. Well, you know, we're hiring languages that are quite difficult to source if we put all this work in one place, or if we only look to hire FTEs for a given role. And so it's really through that sort of unified workforce that we can actually meet the needs and scale.
- [22:09]
B
I'll remind you here that Scola and her fellow plaintiffs worked right here in America. In fact, they were in a building on Facebook's campus.
- [22:19]
A
So, as we think back to the last couple of years, we've realized there may be places where we needed to put more structure in place to help do audits at the vendor level to understand what's happening. And so over the next year, we're really going to be putting in place a process to help to make sure that what we think is happening and what we want to be happening actually is happening.
- [22:38]
B
So Facebook brought on 15,000 people to trawl through the worst things that humans could conjure up. It laid out clear guidelines on how they should be treated and supported. It even helped draft the industry standards. But only now, two years on, is it putting in place a formal audit system to ensure that its army of moderators are indeed being treated humanely. I would argue that there are two things going on here. One philosophical, the other economic. The latter, in fact, is pretty basic.
- [23:10]
G
Yeah, I think it's a cost question. Yeah.
- [23:13]
B
Mary Gray is a principal researcher at Microsoft and co author of Ghost Work, a book on the rise of content moderation and other mundane jobs that keep the Internet ticking over.
- [23:25]
G
Arguably, it's how much of tech has dealt with what is effectively customer service in the past. So there's a history here and an infrastructure of depending on outsourced labor to handle customer calls and customer service responses. And so that companies effectively followed the same pattern of procurement, of looking for subcontractors who had the language skills and the cultural capacity to be able to respond to what are more or less customer complaints.
- [24:03]
B
But this goes deeper than dollars and cents, and it's a theme that we've touched on earlier in this series. It has to do with the worldview of Silicon Valley. At its heart, it is techno utopian. There's a belief that any problem can be Solved with code for content moderation workers seem to be seen and treated as a stopgap which companies must tolerate until machine learning and artificial intelligence are good enough to do it all themselves.
- [24:31]
G
And you know, the challenge in front of us is that technology companies that have built not just their profits, but their reputation and their sense of self and their value in the world on the solutions that technologies provide. The world have to come to grips with the limits of what technology has to offer. And that's no small concession for a technology company. Their sheer grit and optimism about the potential for computation to solve that problem is unshakeable.
- [25:10]
B
And there's a twist to this tale. Every day that humans are tagging photos, blocking videos, judging posts. They're helping the machines get smarter. Remember Mturk? Way back in 2007, Fei Fei Li, a Stanford computer science professor, turned to Amazon's gig Work marketplace for a specific task. She was trying to train AI to recognize simple images, say a dog or a couch. But the only way to train the algorithms was was with lots of pictures where everything was labeled over time. Once the program had been told enough times what a dog looked like, it could then pick one out on its own. She started out with grad students doing the grunt work, but then did the calculation it would be 19 years to build up a big enough store of labeled photos. Mturk was the solution. She outsourced it to an estimated 50,000 workers who completed the job in just two and a half years. Years. The result was ImageNet, which became the gold standard of data sets and was used to develop increasingly powerful algorithms that today are fueling the AI revolution. And you can probably see where I'm going with this. That revolution is supplanting humans in fields from radiology to accounting to image recognition itself. Mturk's army of content workers and in other words, helped build what Silicon Valley hopes will be their replacement.
- [26:52]
E
Jeff King Again, where we want humans in the loop is essentially to make the decisions where it's less clear. And also, of course, to continue to audit and train the machines over time.
- [27:05]
B
What do you mean, train the machines?
- [27:06]
E
I mean the machines don't learn by themselves, right? What you do is you take thousands or millions of pieces of content and you have humans label them right through a machine learning labeling process. A simple example, you get thousands of pictures of streets and you have humans label all the ones with stop signs. At some point, the machines understand what a stop sign looks like and it can pick up future pictures of an object that has a stop Sign in it. So you need the humans in the loop.
- [27:37]
B
For what it's worth, every person I spoke to at Facebook was very clear. They're not racing to replace people with machines. Like King said, they need humans in the loop. But if Mary Gray is right, and I think she is, that Silicon Valley's faith in the ability of technology to solve any problem is truly unshakable, then how could they not view humans as anything other than placeholders?
- [28:05]
G
Worst case, we keep pretending like this work is going to go away any day, that we're going to automate it out of existence, and therefore we, we don't have to care about the people doing this work.
- [28:16]
B
And that's the thing. Technology may be coming on leaps and bounds, but the Internet itself, the amount of stuff that is migrating to the web is simply outpacing it. The algorithms have their limits. So content moderation, the shadow work from which we all benefit, is a growth industry, even if Silicon Valley wishes that it wasn't. Marlena Griffin, again, there's ugly stuff on
- [28:43]
F
the Internet and a lot of it, real people filter through. Because these workers are basically independent contractors. We're not really protected by any union or laws or anything like that. And so a lot of people just, they just work for what they can find.
- [29:08]
B
And is this. Do you think you'll continue to do this?
- [29:11]
F
For now, yes, while I can. If I move and I'm not in this living situation anymore and I have, say rent, then I would definitely consider even a fast food job.
- [29:28]
B
Mary Gray spent years interviewing people like Selena Scola and Marlena Griffin. Believe it or not, she's optimistic.
- [29:37]
G
Society has to ask for more of tech companies than we have in the past, and we're not that deep into it. It's only been since 2005. That's not much time in the history of technology. So it's early days. It's early days.
- [29:53]
B
It is early days. The tech industry has moved fast and broken things, including some of the people, the ghosts in the machine. And whether it's from pressure in the courts or news stories or public outcry, the industry will have to catch up to raise the standards of how it treats the people who make this whole revolution go. As you might have gathered, Facebook declined to comment directly about its legal case, but it did provide a statement. A spokesperson said, quote, what happens in the world makes its way onto Facebook, and that unfortunately often includes things that don't belong on our platform. Many of the decisions we make, including cultural nuance and intent, still require a human eye and judgment, which is why people will continue to be part of the equation. We are committed to working with our partners to continue to improve our operating model and provide support for those that Review Content. The technology revolution that we are all living through makes me think of another the Industrial Revolution. We look back on it today with a mix of awe and bewilderment. It revolutionized societies, increased living standards, and changed the course of human history. But it was also powered by child labor factories polluted with abandon. Eventually, of course, these excesses were stamped out. Laws were passed. In many ways, we're in a similar position today. The web has turned society upside down. It has made our lives immeasurably better. For some, it has made it worse. But like Mary Gray said, it's early in the history of technology and how we all learn to live with it. Tales of Silicon Valley was written and narrated by me, Danny Fortson, with production by Jim Carrey at Rethink Audio. Matt hall is the executive producer for Wireless Studios. It was a Wireless Studios production for Times Newspapers. If you've enjoyed this series, I'll ask you just two favors. One tell a friend and two give it a rating on Apple Podcasts. That helps other people find the show. I hope that after these eight episodes, you understand Silicon Valley better in all of its soaring ambition, immense power, and very human shortcomings. And if you want more, head on over to my other podcast, Danny in the Valley, where you can hear interviews with everyone from Bill Gates and Marc Andreessen to the anonymous startup founder working on what they hope will be the next big thing. After a few weeks off, we'll be revving that back up with a whole new season of interviews. That's Danny in the Valley. Wherever you get your podcasts.
- [33:02]
C
This episode of the Times Tech Podcast is sponsored by ServiceNow.
- [33:06]
B
There's a lot of excitement around AI right now, but the problem is what happens after the demo when you have
- [33:12]
C
to plug that technology into a real company.
- [33:15]
B
Different clouds, different data, different systems that were never designed to talk to each other.
- [33:19]
C
ServiceNow's platform is designed to help people by connecting these pieces, enabling organizations to coordinate work across departments, tools, and increasingly, AI agents.
- [33:30]
B
In fact, the company says more than 80 billion workflows run on its platform
- [33:34]
C
every year, which gives you a sense of the scale of operations it's designed to handle.
- [33:40]
B
Learn how ServiceNow puts AI to work for people@servicenow.com
- [33:47]
D
Leadership used to mean having all the answers, but today's best leaders embody a more human approach. I'm Jack Myers and I'm Tim Spengler.
- [33:54]
E
Tim and I have spent our careers inside media marketing and culture and we
- [33:58]
D
partnered with the ACAST Creator Network to start Lead Human to answer one simple question. What does it really look like to lead in this AI dominated world?
- [34:08]
B
The biggest tip for being a creator? It's a job. What I learned from Michael Jackson Here's
- [34:13]
G
a man who understands precision.
- [34:16]
B
It's about answering the questions that are hard, not about answering a bunch of teed up questions that are fake.
- [34:21]
D
What we're looking for are real stories and practical advice that you can use with your teams right away.
- [34:26]
B
Subscribe to Lead Human with Jack Myers
- [34:29]
E
and Tim Spengler wherever you get your podcasts.