Oral evidence: Fake News – 8 February 2018 (George Washington University, Washington DC), HC 363

Oral evidence, 8 February 2018.

Witnesses

Juniper Downs, Global Head of Public Policy, YouTube; and Richard Gingras, Vice President of News, Google.

Chair

Welcome, everyone, to this special hearing of the House of Commons Digital, Culture, Media and Sport Committee. Today’s evidence sessions here at George Washington University form part of our inquiry into fake news and disinformation. We are thrilled to be here and absolutely delighted with the support and welcome we have received from George Washington University. We are grateful for their efforts in helping us put on these hearings. This is the first time a Select Committee has taken live evidence in this way outside the United Kingdom, so I am grateful for the work of everyone who has made the logistics behind that possible. I thank the witnesses from the different companies we will see today for their evidence, too. I have one small piece of housekeeping for people who are unfamiliar with the work of House of Commons Select Committees but perhaps more familiar with the work of congressional committees. It is understood that anyone taking part in a proceeding of Parliament answers questions honestly and truthfully to the Committee, and it is an offence to mislead Parliament. I am not suggesting for a moment that our witnesses would do that, but we do not require witnesses to swear an oath in front of the Committee because, in our rules, it is implied that they will be telling the truth. We do not need to see an oath being sworn to affirm that. We have a number of sessions to get through today, and I am conscious that we need to do that in good time and in good order. I ask members of the Committee to be clear in directing their questions to individual witnesses and to be clear what they are asking for. Let me also say that, because of the time constraints, I would be grateful if witnesses answered the questions put to them and avoided, as far as possible, being drawn into general statements of policy. If a question is directed to you personally, you should be the person who answers it. I hope that if we can do that, we will keep to some sort of time. I would like to start with you, Mr Gingras. Do you, and does Google, regard campaigns of disinformation and fake news as harmful to your customers?

Richard Gingras

First of all, thank you very much, Mr Collins, for the opportunity to be here to address the Committee and present evidence. There is no question that the issue of misinformation and untoward influence on our societies is indeed an issue. It is certainly an issue for Google. I should point out that, as you know, our mission is to organise the world’s information and make it accessible and make it useful. Billions of users take advantage of our services every day and rely on us to do the right thing and to provide the right information. So this is not only important to us as a company, and important to our users and to society, it is crucial to what we do. I often think of ourselves as being in the trust business. We will continue to retain the loyal usage of people around the world to the extent that we retain their trust with our work every day. So I agree it is an important and significant issue. We will continue our efforts to make sure that, throughout what we do on Google Search or Google News, we are surfacing the best information from the best possible sources and doing our best to make sure that untrustworthy information does not surface.

Chair

Thank you. I think we would certainly agree it is important and I am glad that you think it is important too. I asked you whether you thought it was harmful—whether disinformation can be harmful to society and to your customers, who may be innocent recipients of it.

Richard Gingras

Without question, it can be harmful. I think of misinformation in a broad range of areas, not just in news and politics. People come to Google every day looking for, for instance, medical information. It is not hard in the medical area if you are concerned about health. You could take, “Do peach pits cure cancer?”—you can go to Google and indeed you might find information that suggests that. Across our surfaces, as we look at content, it is how we make sure that they have the right information that is helpful, and not harmful, to them.

Chair

If disinformation is harmful to society and individuals, what responsibility do you think companies like yours and the other tech companies that we will hear evidence from today have? In what sense do you think you have a responsibility to protect your customers from being unwittingly exposed to harmful disinformation?

Richard Gingras

We feel an extraordinary sense of responsibility. As I was noting, the loyalty of our users is based on their continued trust in us, to the extent that if they don’t trust us, they will stop using our products and our business will collapse. So from that perspective alone, it is crucial to us—but it goes beyond that. We believe strongly in having an effective democracy. We believe strongly in both supporting free expression and supporting a sustainable, high-quality, journalism ecosystem to make sure that quality information is out there and to make sure that our citizens, not just our customers, have the knowledge and information they need to be good citizens.

Chair

This is to Juniper Downs. Research has been done looking at the role that the “next up” feature on YouTube plays in supplying fake information to users through recommendation. What steps are you taking to address that problem?

Juniper Downs

Thank you, and thank you to the Committee. Good morning, everyone. I am Juniper Downs, global director of public policy and government relations for YouTube. Our recommendation engine was designed to give users the kinds of content they want to see and it actually works quite well in the main use cases of YouTube—educational content, comedy, music—providing users with more of the kinds of content they want to see. It is quite useful in those genres. News content makes up a fairly small percentage of YouTube’s overall watch-time—less than 2%. We recognise that there is work to do on our recommendation engine in terms of making sure that we are surfacing the right news content to our users. We have been investing a lot of resources in making sure we surface authoritative news sources and demote low-quality content, but as some of the press coverage has shown over the past week, we still have work to do. There is progress to be made. We recognise that and take responsibility for it.

Chair

YouTube has supplied the Senate Intelligence Committee with information related to 18 YouTube channels that were backed by the Internet Research Agency in St Petersburg, which account for 1,100 videos and 43 hours of content. Could you confirm whether the identification of those channels was based on analysis conducted by the Intelligence Committee itself, based on information it received, or was that YouTube’s own research looking for channels like that?

Juniper Downs

The security and integrity of our products is core to our business model. We co-operated with the congressional investigation into whether there was any interference in the US election, and the channels we discovered on YouTube that were connected to the internet research agency were due to a thorough investigation that we conducted using our own resources. We have publicly reported that information to Congress through the Intelligence Committee, where our general counsel testified back in the fall.

Chair

Was the identification of those accounts based on your own research, and not on intelligence or information you were given?

Juniper Downs

Our research and our investigation included leads and intelligence provided to us by others, including other companies who were conducting similar investigations.

Chair

But what did you do yourselves that was based not on leads, but on your ability to analyse how people are using the platform?

Juniper Downs

Correct. We looked at the leads we were provided and went far beyond that in looking at any advertisements that had a connection to Russia. We also looked at organic content to see if there were channels on YouTube that were connected to the internet research agency that were not purchasing advertising but were simply uploading content.

Chair

You may be aware that the Committee has written to Twitter and Facebook asking for their analysis of whether Russian-backed agencies were involved in interfering with elections in the UK, particularly the Brexit referendum and our recent general election. Facebook is conducting that research for us, and I hope we will get an update from Twitter later, but I don’t really know what they are doing about it. Would YouTube be prepared to conduct the same analysis for the UK that you have conducted in the United States and look at potential Russian interference in and around the Brexit referendum and other elections?

Juniper Downs

Absolutely. We are happy to co-operate with the UK Government’s investigation into whether there was any interference in elections in the UK. We have conducted a thorough investigation on the Brexit referendum and found no evidence of interference. Again, we looked at all advertisements with any connection to Russia, and we found no evidence of our services being used to interfere in the Brexit referendum. We are happy to co-operate with any further efforts.

Chair

Obviously, that is the investigation by the Electoral Commission, which is an agency of the Government. I am asking about this Committee’s investigation. We are not necessarily just looking for paid-for advertising linked to the election, but the operation of channels or the uploading of films that can be linked back to Russian agencies and had a political purpose or message during the referendum. Would you be able to do that for us?

Juniper Downs

We are happy to co-operate with that investigation.

Chair

So that’s a yes?

Juniper Downs

Yes.

Chair

Good, thank you.

Jo Stevens

I would like to ask about the search function on Google. You may be aware of a British journalist, Carole Cadwalladr. We met some members of your company earlier this week, and she was talking about the auto-fill or auto-finish search. She conducted a simple exercise: when she typed “Jew” into the search function, it would auto-finish the rest of the search for her, and all of it came up with far right, extremist anti-Semitic searches. I am interested to know why it took a UK journalist doing such a simple exercise to identify this issue. Why had Google not identified it previously?

Richard Gingras

I thank you for that question. These occurrences happen from time to time. Many we catch ourselves; others we do not. The auto-complete feature on Google search is an important service to our users: you come to search, you begin to enter a query and we offer suggestions based on what other users have searched for. It is a dynamic feature. Everything we work on in terms of the corpus of expression of the web or the corpus of queries that people ask is a live and vibrant corpus that changes every day. There are indeed also malicious actors out there who will seek to game that as well. As we have constructed those, we continue to build defences against that, and also to build mechanisms such that users inside or outside the company have the agency to identify these things, mention them and bring them to our attention so that we can quickly correct them. That is what we look to do. Clearly, the kinds of terms that you might see in those are offensive—offensive to all of us, egregiously offensive—and we will look to take care of them. Again, it goes back to the trust of our users.

Jo Stevens

Your algorithm development and the data they gather is a cyclical thing, isn’t it?

Richard Gingras

Yes.

Jo Stevens

So that then determines what phrases are picked up. What safeguards are you putting in place to prevent that cyclical enhancement of hate content, if people are searching using those terms? Can you give us some examples?

Richard Gingras

It is about the continuous advance of our systems, in terms of what kinds of terms and strings of words to look for and making sure they don’t occur. It is also about maintaining constant efforts on our part to evaluate the results we surface. We have a large team of what we call raters—10,000 people around the globe who are constantly at work assessing our results against different queries. Are we surfacing the right kind of results? Are we surfacing results that have appropriate authority? They use those hundreds of thousands of bits of data from those evaluations to continue to train our systems. This is an ongoing effort. As much as I would like to believe that our algorithms can be perfect, I don’t expect they ever will be, simply because of the dynamics of the ecosystem upon which we work. We will continue to strive to make sure situations like that occur as infrequently as possible.

Jo Stevens

This is a final question on developers and algorithms. Do you have an ethics policy or framework that your developers work within, or do developers’ inherent biases—we all have inherent biases—influence the algorithm they build?

Richard Gingras

Absolutely, that is crucial. There are ethics policies that guide our work. For the raters specifically, we have a 160-page set of rater guidelines—it is public information and is available for anyone here to look at—that guides their assessment. We work on top of policy, we constantly evolve that policy and we constantly train those who have to apply those policies to make sure we are doing the right thing.

Jo Stevens

Sorry, can I clarify that? Your raters, as you call them—are they developers?

Richard Gingras

They are not necessarily developers.

Jo Stevens

Right. So do the developers who build the algorithms have an ethics policy?

Richard Gingras

Yes, we have ethics policies. For instance, we have an internal policy called the “honest results policy”, which prevents our engineering teams and people like me from trying to influence the algorithms in untoward ways or allow third parties to influence our algorithms in inappropriate ways.

Ian C. Lucas

I was very struck by your statement that you are in the trust business. Trust is based on knowledge. Do you believe your customers know what information you retain about them and how you use it?

Richard Gingras

I would hope so. We have made great efforts to provide transparency and control to our users about the information we collect as they use our services. We collect that information to make the services better for them, but they can go to a control panel—hundreds of millions of people have done this—to look at what information we are collecting and, for that matter, to change the settings. They can tell us not to collect certain kinds of information. That is always there for them. It is hugely important that we maintain dialogue with our users about the information we collect from them so we can provide better services to them. We never share that information with third parties—we never have and we never will. It is crucially important that we protect it. Again, it is part of the trust relationship that it is important we maintain with our users.

Ian C. Lucas

So you will give me all the information you hold about me if I ask for it?

Richard Gingras

Indeed. You can come to Google and look at the information we have about you in our different services.

Ian C. Lucas

And will you tell me how you use that information? For example—I am a politician—would you tell me how I could use that information for political purposes?

Richard Gingras

Well, we don’t use any information for political purposes. We try to be as transparent as we can about how we use that information. I will give you an example. Google search is not personalised, except that we will tune the results based on where you are geographically. If you are looking for a restaurant in London, we will surface restaurants in London. Otherwise, we think it is important not to personalise search results. We expect that what you want to see and find is any information that is out there in the corpus of expression without us trying to guide you in one way or another based on what we think you might be interested in.

Ian C. Lucas

Do you market your capabilities to politicians?

Richard Gingras

We market various services—for instance, our ads—to various folks who want to advertise using our products. To some degree, that does use data to help target information.

Ian C. Lucas

Do you have a specific team that markets advertising to politicians?

Richard Gingras

I am not that familiar with our sales organisation selling ads to know how we approach these things. We have advertising sales teams in countries around the world. I imagine some focus on different categories, but it is not an area of my expertise in my role as vice president of news.

Ian C. Lucas

Is it within your area of expertise, Ms Downs?

Juniper Downs

I want to reiterate some of what Richard said about how we use the information we collect from users. The main principles of privacy at Google are transparency and control. For YouTube, for example, we do collect the watch history of signed-in users, but the user can at any time pause or delete that watch history. The way we use that information is to improve the service for users. So, earlier we were talking about recommendations. If we know that someone is a lover of comedy or a particular kind of music, we can use that watch history to optimise the service. We never sell the information to advertisers. We do provide aggregate data to advertisers, to help them optimise their campaigns. In terms of the work our sales team does with various constituents who are interested in using Google’s ad products, that is the kind of information that would be provided—never information about individual users.

Ian C. Lucas

Thank you.

Chair

On that point, if you can provide such precise aggregate data, why do you find it so difficult to identify bad content on the platform? Why do you still have problems of fake news being filtered in next up?

Juniper Downs

Identifying and managing content on YouTube is the No. 1 priority for us this year. It is mission-critical for the business. It is critical to our users, to our creators, to our advertisers and to us as a company, so we invest tremendous resources in terms of both technology and the people working on these issues. Our executive team is absolutely engaged. We meet for hours every week to figure out how we can improve our systems to make sure that the policies on YouTube are followed and that we quickly identify content that violates them and removing it. This is a top priority for the business.

Chair

What percentage of your advertising revenue do you reinvest in the site in this way?

Juniper Downs

I do not have an exact percentage, but we have spent tens of millions of dollars fighting spam and abuse across our products. We have committed to having 10,000 people across Google by the end of this year working on these issues.

Chair

Tens of millions of dollars?

Juniper Downs

Tens of millions.

Chair

What is the annual advertising revenue for YouTube?

Juniper Downs

I am sorry; I don’t know the answer to that question.

Chair

Just roughly—you must have a rough idea.

Juniper Downs

I am sorry; I don’t know the answer to that question, but I want to reiterate that we will invest the necessary resources to address these issues not only in terms of people we employ but the technology we develop. We are continually staffing up and we have seen good progress in our management of these issues. For example, over the past eight months, I think the work we have done on violent extremism speaks for itself: not only have we removed over 100,000 videos but the speed with which we have been able to identify this content and remove it from the site has gotten faster and faster. We are now at the point where 70% of that content is removed within eight hours of upload and 50% within two hours of upload. That is the kind of progress that we demand of ourselves and that we will continue to strive for across all of these issues as we move forward.

Chair

We will see if we can work out during the course of the morning what that percentage is—my suspicion is that it might be quite small. Many people still say, though, that you can sell the service to micro-target people—that is how you make your money—so you must be capable of using the same tools to more effectively identify harmful content. Of the content you do take down, though, what proportion of that is user referral of bad content and how much is bad content you discover for yourselves?

Juniper Downs

The technology we have developed to identify content that may violate our policies is doing more and more of the heavy lifting for us. In the area of violent extremism, 98% of the content we have removed has been identified by our algorithms. That number varies from policy to policy, depending on how effective and precise the technology is on particular issues. When it comes to misleading and deceptive content, our spam detection systems are quite effective at identifying the behavioural patterns associated with mass-produced misinformation that is being distributed through our services. In 2017, we removed over 130,000 videos for violating our misleading spam policy. Those were virtually all identified by our technology.

Chair

If you had evidence of someone being heavily exposed to or distributing violent extremist content or other forms of extreme content that might be dangerous to children, what sort of responsibility do you think you might have to share that knowledge with law enforcement agencies and people who might take an active interest in that person?

Juniper Downs

We co-operate and meet regularly with law enforcement to share information. When it comes to the disclosure of user data, obviously we do that pursuant to valid legal process. There are mechanisms in place where law enforcement can request information from us, pursuant to their investigations. There are also emergency disclosure provisions, where if we identify content on our services that poses an immediate risk to life, we disclose that information proactively. When it comes to the safety of children, obviously we report to the National Centre for Missing and Exploited Children any solicitation of minors or content that is exploitative in that way. It co-operates with law enforcement internationally to make sure that incidents are followed up on.

Chair

I appreciate that you co-operate with authorities upon request. If someone had a family member who was a victim of a violent act carried out by someone who, from their interests on YouTube—perhaps the channels they have set up—quite clearly was potentially quite violent and dangerous, some people might ask whether you have a responsibility to notify the authorities of people who give you cause for concern, and not just wait to co-operate with an investigation after a crime has been committed?

Juniper Downs

Certainly, if we identify someone on our services who is posing that kind of threat to individuals. We see instances where someone may be threatening to perpetrate a violent attack, we would proactively disclose that because we would consider that part of the emergency disclosure provisions.

Rebecca Pow

I want to look at the ethics of YouTube, your “up next” button and your scoring system for videos. Would it be right to say that, usually, the content with the highest score is something that has had the most views? Are hits important?

Juniper Downs

The recommendation engine is quite complex and it varies depending on the video that is being watched. One of the factors is content that is popular on YouTube. There is also content that is associated, so if I have a real niche interest in a particular kind of knitting, there may not be a lot of highly popular videos that are about that type of knitting, but we will continue to recommend videos that are similar and that provide more instruction on that type of knitting.

Rebecca Pow

Popular usually means lots of hits, doesn’t it? I want to mention some research that was done recently by The Guardian that suggested that the largest source of news for false news channels came from these videos on YouTube that tend to be recommended by the “up next” button. More research by Oxford academics demonstrated that a lot of that had a far right bias. I put it to you that if that is the case, you are indoctrinating society.

Juniper Downs

I am not here to comment on the particular methodology of The Guardian research, but when it comes to recommendations, we recognise that we have more work to do. Recommendations are largely a reflection of user interest. When it comes to news content, our goal is to promote more authoritative sources and to demote lower-quality content from less established sources. It is not only popularity that fuels the recommendation engine. It also tries to draw from well-established, known sources of news and put that near the top. Do we succeed at every time? We do not. I will give you a recent example. Earlier this week when the stock market fell, our algorithms recognised “stock market plunge” as being a newsy query because established news sources were using the word “plunge” in their coverage, but they did not recognise “stock market crash” because, for some reason, that term wasn’t being used. That was flagged to us and we immediately corrected it and made sure that we were surfacing authoritative sources across both of those queries, but even those subtleties in language can cause our systems to misfire on occasion and we work quickly to try to correct that.

Rebecca Pow

The Guardian information fed 8,000 YouTube recommendations into their software system, which tracks political disinformation campaigns. In every case, the largest source of traffic for these fake news channels came from YouTube’s “up next” algorithm. Are you clearly saying that you do not agree with that?

Juniper Downs

I am saying it is an area of progress for us. That research was done in the lead-up to the election. We have made a lot of progress since in making sure that the more authoritative sources surface near the top of our recommendation engine. It is an area that we will continue to work on and improve. We recognise that some of those results are not where we want to be as a product, so it is an area of investment for us.

Rebecca Pow

There is one other area I want to touch on. It seems like, in choosing to rank videos and to have a system of what is good, what is better, what is higher and what is lower, you are in fact acting as an editor—that is what an editor of a newspaper would do—yet you are not calling yourselves editors; you are calling yourselves hosts. I wonder whether you think the description of what you are and the whole name of your platform ought to be changed so that you take on more of the responsibilities of a bona fide newspaper and you have to apply broadcast and newspaper regulations to yourself. At the moment, you are unregulated.

Juniper Downs

When it comes to the scale of our services, we have well over 400 hours of content uploaded by creators to YouTube every minute. There has to be some organising principle. When someone comes to YouTube and searches for “funny videos”, there has to be an algorithm that delivers the most relevant and useful results to that individual user. That is how our algorithms function. The question of whether we are a publisher really gets to the question: do we take responsibility for the content on our site? The answer to that question is absolutely yes. It is a top priority for us to make sure that we are providing a useful service to our users and that we are exercising responsibility.

Rebecca Pow

But you are self-policing, aren’t you? Is that right?

Juniper Downs

We have a set of community guidelines that set the rules of the road for YouTube. Those have been in place since we acquired YouTube as a service. They are dynamic. They evolve in response to changing trends in the world and changing trends in what kinds of content and misuse of our services we see. We have developed those guidelines because we are trying to maintain a certain kind of community for our users. They do go above and beyond what the law requires, but we also follow the law in the jurisdictions where we are launched. When we are alerted to content that may violate a particular law, we block that content for the relevant jurisdiction. On the model of self-regulation I have just described, the EU Internet Forum, where we have been co-operating with the European Commission for several years on the removal of illegal hate speech from our services, is a good model for how Governments, NGOs and the tech industry can collaborate to ensure that we act responsibly and quickly to deal with these content issues. The Commission has acknowledged the progress of that process.

Rebecca Pow

You are described as “the most overlooked story of 2016”, and people are describing this whole social media issue as a disease on society that is affecting behaviour. It is surprising that you do not seem to register any of that.

Juniper Downs

We take the criticism there very seriously. Again, it is a top priority for the company to make sure that we continue to invest in detecting content that may violate our policies and watch for bad actors who try to manipulate or exploit our systems. The openness of YouTube has brought tremendous benefits to society when it comes to educational content, culture, music and the arts. Those benefits are tremendous. Of course, that openness also presents challenges, and those challenges are ones that we are committed to managing and dealing with responsibly.

Chair

The Committee Clerk has been busy trying to work out the answer to the question about how much money YouTube makes. Independent market analysis—I appreciate the figure is not disclosed by the company—estimates that YouTube’s global revenues are about $10 billion. Does that sound about right?

Juniper Downs

I really cannot confirm any of those numbers. It is not information that I have access to.

Chair

Well, let us say it is, for easy maths. If you are investing $10 million a year in making YouTube safer and your revenues are $10 billion, you are reinvesting 0.1% in making the site safer. If that is true, it sounds to me like a small sticking plaster over a gaping wound, and it is not likely to be satisfactory in addressing some of the concerns that Rebecca Pow has raised. I do not know whether you want to comment on those figures. I appreciate the company may not want to disclose revenue figures, but if it could give us a percentage figure for the amount that is reinvested in the site as a proportion of revenues from YouTube, we would be very grateful for that. As I say, based on publicly available figures, 0.1% seems to be where that comes out.

Juniper Downs

I think your question gets to whether we are investing adequate resources in addressing these issues. There really is no constraint on the resources that we will put towards getting this right. We have invested tremendous time—time from our engineering teams, from our executive teams and from the trust and safety operation, which has nearly doubled in size over the past year. We take these issues very seriously. If we decide that we are unable to make our desired progress because we do not have adequate investment, we will invest more in addressing it.

Chair

All I would say is that, for a multi-billion-dollar business, if tens of millions of dollars is the investment to address these widely held and increasingly well-documented concerns, it sounds like a pretty unambitious programme of investment. I think, understandably, given the huge size of the platform, people would question whether that sort of investment is ever likely to be successful in addressing some of these concerns.

Christian Matheson

Ms Downs, I want to follow on from Ms Pow’s questions about the Paul Lewis article in The Guardian. It quotes Guillaume Chaslot, one of your former engineers, who says that “Watch time was the priority” for your algorithms in prioritising what came up next, not necessarily truthfulness, decency or honesty. How would you comment on that allegation?

Juniper Downs

Watch time is obviously an important metric for us, because it demonstrates that we are providing a service that users love, where they want to spend time on the product or they are enjoying the experience of YouTube and finding it valuable. For certain kinds of content, such as watch time for music, I know I listen to hours of music on YouTube because it keeps delivering the music I want to hear. We know that for news and for certain other verticals, the veracity of the content we provide to our users is also incredibly important. That is why some of the changes we have made over the past years have been about making sure we are surfacing more authoritative content to users, where they are getting the information they are seeking, and demoting lower-quality content. We have created dedicated new surfaces. We have a new breaking news shelf, so when there is a breaking news event we surface high-quality content near the top, and then the recommendations are also limited to those authoritative sources. We also invest in making sure that the users of YouTube have the skills they need to assess the content they are watching. In the UK, we created a campaign called Internet Citizens, where we are committed to training 20,000 young people in identifying fake news and learning how to check sources and bias and so on. It is about an effort that supports the whole ecosystem. We also work with publishers. We launched a programme called Player for Publishers, which now has 50 European news providers in it. We provide them with the technical support to have a video player embedded in their own website, which reduces complexity and cost for them, and we also work with them to optimise the content they produce for YouTube. Their watch time has doubled, and they really appreciate the fact that their watch time has increased through our investment in them. Watch time can be an important metric.

Christian Matheson

So you do accept that there is a problem of bogus news, misleading news and disinformation, and you also accept that you have both a responsibility and the means to address that?

Juniper Downs

We recognise that there is a problem of misinformation and we are dedicated to making sure that we are promoting authoritative content when it comes to news and demoting misinformation and low-quality content.

Christian Matheson

Do you have a sense of moral responsibility to take those actions?

Juniper Downs

We have a sense of social responsibility and it is a business priority for us. The trust that users have in our services is reliant on our providing a high-quality experience when they come to the site. Presumably, when people come looking for news rather than entertainment, what they expect to find is factual reporting from a variety of sources. We do provide information from diverse news sources from around the world. We think that is valuable. So yes, our investment in this area is because we recognise that it is an important policy matter, but also because it is an important business priority.

Christian Matheson

Paul Lewis confirmed the Chaslot conclusions in relation to the United States presidential election, which suggested that YouTube was six times more likely to recommend videos that aided the Trump campaign than the Clinton campaign. Do you accept those figures?

Juniper Downs

Again, I am not here to comment on the particular methodology, but I will say—

Christian Matheson

No, go on, comment on it.

Juniper Downs

We did publicly say that we did not agree with the methodology that was used.

Christian Matheson

Why not?

Juniper Downs

We were not provided the 8,000 videos to examine them, but this is an employee who left the company—

Christian Matheson

You’ve got a bigger selection, because you have got the whole set, so maybe you can run their methodology past your bigger set and get what you might consider to be a more accurate solution. How about doing that?

Juniper Downs

Let me explain how I think the recommendation engine functioned in the lead-up to the election. It is a reflection of user interest, so to the extent there was more content recommended about one candidate versus the other, that was because there was a lot more user interest being expressed about that candidate. So we saw more searches for that candidate. I think we saw similar trends across broadcast coverage of the US election, where one candidate got more coverage than the other. So that is a reflection of what users wanted to see, what they were interested in and what they were looking for when they came to the site.

Christian Matheson

But Ms Stevens has already identified the possibility that that becomes a self-fulfilling prophecy, because your recommendations and “up next” simply reinforce what has gone before, and it tends to spiral in one direction, as opposed to providing balance.

Juniper Downs

This is an important question for us to figure out how to ensure that we are providing information that users actually want to watch and will watch—that is also diverse. We do not build any political bias into our algorithms. We design products for everyone and we invest heavily in ensuring that there is no bias built into the algorithms we design. There is a human element here, too, though. People want to watch what they want to watch, so if someone is not expressing an interest in a particular person or type of content, it is hard to insert something opposite to that and expect them to watch it. We see abandonment of the service when we do that, because humans have their own will. So we try to create a diversity of content but, at the same time, keep it topical enough that the user feels we are fulfilling the interest they have expressed to us when they come to the service.

Christian Matheson

Finally, if a video is getting hundreds of thousands or even millions of hits and, all of a sudden, when it is at the top of its performance, it gets taken down, would that not be a little strange, in reference to your business model.

Juniper Downs

I am not sure I quite understand the question.

Christian Matheson

If a video is getting hundreds of thousands or millions of hits—which is good for you and for the person who posted the video—and then they take it down when it is at the top of the game, would that not strike you as a little bit odd?

Juniper Downs

The community guidelines and content policies we have for YouTube apply to all of our creators equally, no matter how popular they are. We identify content when it is flagged by users or detected by our own internal systems. We evaluate those videos and, if they violate our policies, we remove them immediately. It does not matter who the creator is or how many views the video has.

Christian Matheson

What if the creator takes it down?

Juniper Downs

The creators on YouTube always have control over their own content and are free to delete it at any time.

Chair

Thank you.

Julian Knight

You have talked about how much progress you have made, and you believe that your moves in this area are quite effective, yet we have just discovered that potentially you are spending 0.1% of your turnover. More recently, in fact this morning, the front page of The Wall Street Journal has a forensic investigation—old-style journalism, if you like—by Jack Nicas, who says that your recommendations “often present divisive, misleading or false content, despite changes the site has recently made to highlight more-neutral fare”. Potentially up to 70% of viewing time is taken up with these recommendations. In the light of that, why has your self-regulation so demonstrably failed, and how many chances do you need?

Juniper Downs

Again, our recommendation engine was designed with the main use case of YouTube in mind—the things that people to come to YouTube and love the site for: comedy, music, cooking vlogs and beauty vlogs. It works quite well for those use cases. We recognise that for news the effort we have put in to promote more authoritative sources and demote lower-quality content is a work in progress. We will continue to invest in getting it better.

Julian Knight

But you stand by the fact that it is quite effective? That is what you said at the start.

Juniper Downs

It is quite effective for the majority use case of YouTube. But news is less than 2% of our watch time, so it is actually not the majority use case of the platform, but is an area—

Julian Knight

So you have got the 98%, but you think for that 2% your algorithms do not work at the moment.

Juniper Downs

I think they are working better than they were six, eight months ago and so on. This is an area where we are continually investing to make sure that we are providing the right news experience for users when they come to YouTube. We are committed to doing better. When we see some of the results in that Wall Street Journal article, frankly we are not proud of them. That is not the experience that we want to provide to our users—

Julian Knight

So you agree with the methodology of this article and take that on board? Potentially, you dispute the one from the past in The Guardian. We have a phrase in Britain—“It’s a fair cop.” Is this a fair cop?

Juniper Downs

I think the results pictured in that article were taken from YouTube, so to the extent that we look at those and think we could do better, we certainly look at those and think we can do better.

Julian Knight

One area where you seem to do remarkably better is over things like sporting rights. For example, if I wish to post a final touchdown from the Super Bowl—I am saying that as I am in America, but I am probably thinking about a premier league soccer goal—that would be taken down within minutes. It would be almost instantaneous, and you would be right on to it. Why does it take so much longer, if at all, when it comes to misinformation by foreign powers that are specifically looking to undermine the west?

Juniper Downs

When it comes to copyrighted material, we have invested a lot over the years to make sure that we protect people’s intellectual property rights. Part of why the detection and removal is so much faster is that rights holders provide us with a digital file of their copyrighted material, so we have that as the starting place. We already know what content the systems are looking for, so we can identify matches with that content relatively quickly using our technology. If we were provided with a digital file of every misinformation video, we would be able to identify them just as quickly. Unfortunately, our systems have to look for patterns, and they are quite sophisticated at doing that, but bad actors are also constantly evolving and trying to evade detection. It is a little bit of a cat and mouse game, and we are obviously always trying to stay one step ahead of those who attempt to misuse our services.

Julian Knight

So it is not just about the money or the fear of being sued that you act so quickly when it comes to copyright? However, when it comes to misinformation from, as I say, foreign actors, the evidence is abundant and out there that it is effectively not in the same ballpark, to stretch the sporting analogy.

Juniper Downs

It is a top priority for us. As I said, the security—

Julian Knight

0.1%.

Juniper Downs

I am not sure that the financials behind this are the right metric, in that the technology we deploy, which is a hard thing to monetise, actually does a lot of the work at scale to identify this content with speed. If you try to quantify in people hours how many hours of human endeavour are being saved by the technology, we would get to a much vaster number. In order to address these issues at scale, we really need to invest in the technology that can find the content and enable us to act on it quickly.

Julian Knight

Okay. I will turn to you for a question in a moment, Mr Gingras—I am sorry to leave you out. Just ahead of their elections, Germany effectively regulates in terms of removing hate speech from social media. I think they were concerned about the reaction to immigration. Social media responded, and many commentators suggested that there was a demonstrable reduction in the level and effectiveness of their elections—it also happened in the French elections. Surely that is strong evidence that the way in which western democracies protect themselves is to regulate you?

Juniper Downs

We work with elections and campaigns around the world to help them with digital tools to protect their websites and campaigns from any interference—from phishing attempts and so on. We are very committed to investing in the integrity of elections and to providing security support to political candidates and campaigns. That is a service that we can help to provide. We collaborate with others on that. In terms of the German law that was recently passed, there has been robust public debate about the risks and benefits of that approach. I think, since it was implemented in January, there has been a lot of public criticism of the way it has played out. The debate is ongoing on whether that is the right approach.

Julian Knight

Mr Gingras, your company very kindly hosted us in New York in order to talk about the Trust Project, which was very interesting. While we were in New York, we also received some quite stark evidence of the crushing impact on local press, and increasingly on national press, of the sucking out of advertising revenue by companies such as yours from more traditional media. You talk about how you want to train journalists so they are able to use your platforms more effectively, but what happens when all we have are a few global players, such as The New York Times, Washington Post and, on our side of the Atlantic, the British Broadcasting Corporation? It is pointless having a journalistic programme if there are so few journalists left. What else will you do in order to ensure that proper journalism survives?

Richard Gingras

That’s a very important question, and it is one we are deeply concerned about as well. I should point out that the success of our efforts with Google Search and our ads business is really dependent on there being a rich ecosystem of knowledge out there, including at the local level. As we often say, our success is dependent on the success of publishers. I will point out that the publishing business, in every dimension, has without question been changed and affected by the introduction of the internet. I will also point out, however, the various things that we are doing to assist in this. First, we share 70% plus of our display ad revenue with publishers around the world, to the tune of some $12 billion—that is one element—as well as providing them with ad systems and tools that allow them to derive revenue. Some 2 million publishers around the world benefit from that. We also drive traffic to many new sites, to the tune of well beyond 10 billion visits per month, which has been valued by third parties at between 5p and 7p per visit. That is another $5 billion to $7 billion a year in revenue to the industry. So that is one component, but it is just one. We need to recognise that the ecosystem has changed. One way to think about this is through our own actions. I started out in the newspaper business—I started out in the press rooms of The Providence Journal when I was a teenager.

Julian Knight

Does The Providence Journal still exist?

Richard Gingras

Yes, it does still exist.

Julian Knight

There are not that many local newspapers in the US that do. It is good that you are giving this background, but the point is what happens when the journalistic landscape is so decimated and people do not know what to trust? Do you have a sense of responsibility to ensure that we have that trusted news?

Richard Gingras

We do feel a strong sense of responsibility, for the reasons I pointed out. It is important to society and it is important to the nature of our business, so we have mounted many efforts. My role at Google is, on the one side, to oversee our efforts to surface news on Google Search and Google News, but another part of it is the various efforts we have mounted to enable the publishing systems and legacy publishers to make the transition from the old marketplace of information before the internet to today’s one. Behaviours have changed. Think about it. When I was young and looking for a used car, I went to the classifieds in my local newspaper. If you are doing that in Birmingham today you will go to Gumtree, not your local newspaper. If I was looking for a house or a rental apartment as a kid, I would go to a newspaper. Today, in Liverpool, you will go to Zillow. That is even when it comes to national news. I went to local newspapers for my national news—much of it might have been from wire services, but I went to them. Today, people’s behaviours have changed. They go to The Telegraph or the London Times. The marketplace has changed. We feel that part of our responsibility is how we help and enable news organisations to figure out what kinds of products they can serve to their local communities. We do that in various ways. One is through deep collaboration with the publishing industry. It is important that we understand their challenges and they understand what we can do. We have mounted efforts such as the subscription project. Can we bring better tools to help to drive subscription revenue? Can we bring better data and knowledge to help them to identify targets of opportunity in their markets? We helped with tools for digital storytelling and training for journalists through the Google News Lab. We have trained thousands of journalists in the UK, and will continue to do more. We have provided innovation funding in Europe through the digital news initiative—some $83 million to date, and £6 million, I should say, in the UK, helping folks like Trinity Mirror develop news services like InYourArea. There is a lot of opportunity to continue to develop, and we are seeing good signs of success around the world, but there is no question that there is more work to be done and more innovation that needs to take place to reach that sustainable ecosystem of local news publishers as well.

Julian Knight

Thank you.

Brendan O'Hara

I have a couple of follow-up questions—one on Mr Matheson’s and the other on Mr Knight’s. Following on from Mr Matheson, your business relies on people being on your site and viewing for as long as possible. Hence, you have these sophisticated recommendation engines, which are designed to do that. How do those recommendation engines differentiate between children and adults? Is it the goal to get the user, regardless of age, to stay on the site and consuming content for as long as possible?

Juniper Downs

When it comes to Google, we only allow users who are 13 and older. That is the limitation on the site. Your question gets to the issue of public concern around tech addiction, and particularly young people’s use of smartphones and social media. When I think about the goal of Google Services, it is really to provide products that enhance people’s lives, not detract from them. We think the question of how we can better fulfil that goal is a really important question for society, so we are investing in research to better understand these issues and to better understand what kinds of product design we can implement to make sure that we are providing a product that is an enhancement to people’s lives and enhances people’s productivity. If you look at YouTube, we have 1 billion views of educational content every day. That is a tremendous asset to society. If you think about something like Khan Academy, where an uncle started making videos—

Chair

Sorry to interrupt, but I need to direct you back to the question and away from making general statements about YouTube.

Brendan O'Hara

Thank you, Chair. The question is how you differentiate between adult consumption and child consumption.

Juniper Downs

We don’t look at adult consumption versus child consumption. The recommendations are based on the video that is being watched and content associated with that video, or with the watch history of the individual signed-in user. Again, only people aged 13 and older can sign in to YouTube, so the watch history there would be teenagers and above.

Brendan O'Hara

Okay, so there is no mechanism, other than an arbitrary “You have to be 13 to view”, that allows you to say who is watching what? Anyone of any age—any tech-savvy five or six-year-old—could probably get around that 13-year-old barrier. Any tech-savvy child could consume as much content as they like and you wouldn’t know.

Juniper Downs

Protecting teen users on YouTube is a top priority for us. One thing we do is age-gate videos that are more mature. You have to be signed in and aged over 18 to view those. That is the one place where more mature videos aren’t eligible to be viewed or recommended.

Brendan O'Hara

I am not talking about more mature videos. I am talking about the unrestricted consumption of YouTube videos by children. Is there a way that you have managed to address that? I suspect from your answer that you haven’t.

Juniper Downs

The investment that we have made is in developing a dedicated YouTube experience for families and children, which is a dedicated app called YouTube Kids. When we designed that product, the goal was to create a new service that gives parents control and information. It was informed by collaboration with child safety experts. We designed a product that is a much more limited corpus of YouTube videos—the family-friendly content—and we give parents controls such as a timer for how long their child can view the app, the ability to turn search off, so that the child is limited to a much smaller set of videos, and other controls. That was because we want to provide a good experience for families, but we recognised that it really needed to be in its own space.

Brendan O'Hara

Are you happy that you are doing everything you possibly can—this goes back to the trust issue that you talked about, Mr Gingras—to protect children from this unlimited consumption that we hear so much about? That is why people have so many issues about child interaction with the internet at the moment.

Richard Gingras

I would point out that we are never fully satisfied with our work. As I have mentioned, this vibrant, live ecosystem changes every day. For instance, with Google Search, we get billions of queries every day, and 15% of those are queries we have never seen before. It is an ongoing effort. As I often mention, our algorithms are not perfect and I doubt they ever will be, which means we have to be ever vigilant. I noted earlier, when we were talking about how many dollars we have invested in this, that when we mention, for instance, $10 million of additional investment in security efforts and so on, that is just additional recent investment. It is building on a foundation of effort. We have thousands of engineers at YouTube, thousands of engineers at Google Search, security systems and security people that we are building upon, and will continue to build upon, such that we can understand the new challenges and address them as quickly as we can. I just want to be clear that there are no laurels that we rest on here.

Brendan O'Hara

My final question relates to what Mr Knight was saying. I am talking about the fairy-tale idea of a fun space where everyone could share and enjoy. It really has been superseded by events. It is no longer possible to have YouTube operate in the way it did way back at the beginning, because of people who are exploiting it and social issues that have arisen. Are you satisfied that the current legislative framework is sufficiently robust to meet these new challenges, or would you welcome further legislation; or, as the debate is now happening, how would you like to participate in that debate, to shape a legislative framework that works for everyone?

Juniper Downs

We are intrinsically motivated to address these issues. Again, it is the top priority for YouTube as a company, and we welcome any conversation about legislative proposals. A big part of my organisation’s role is interacting with policy makers and other stakeholders to hear ideas and to participate in conversation about them, so we are always happy to do that. I want to reiterate that we do not need extra motivation to get this right. It is mission critical for the business for us to address these issues responsibly, and that is what we are committed to doing.

Richard Gingras

But more collaboration, more engagement—these are problems that we are facing as a society, that we are all facing. There are no silver bullets for any of these questions; they are big questions, whether we are talking about the challenges of misinformation or of the evolution of the publishing ecosystem. If there is anything that we have learned over the last several years—this didn’t start in the last 18 months—it is that our best approach to these is to engage with the community, whether they be publishers, people in the public policy space or whoever, such that we can all, collectively, gain a better understanding of how to approach the problems and what they are and how we partner together on finding better solutions.

Chair

Thank you, Mr Gingras. We take your point, but we have heard the expression “top priority” a lot. If we judge the company based on what it does rather than what it says, the top priority is maximising advertising revenue from the platform, and a very small proportion of that is reinvested into dealing with some of the more harmful content. That is one of the reasons why we are here and why social concern about this is growing. Giles, do you want to come in?

Giles Watling

To use pretty much every analogy under the sun, it seems to me that you have opened a Pandora’s box, probably unintentionally, and you have got a tiger by the tail. And you are now, to carry on the sporting analogy, somewhat behind the curve. It seems extraordinary. You were talking about misinformation. Let’s face it: misinformation, disinformation, fake news can be sort of sexy. People like ghosts, conspiracy theories and UFO stories, and they follow it. This creates, I would imagine, a huge amount of traffic, both propagating those theories and so on, and from those who want to quash it. That would be good for you, because you can then harvest the data from the people who are doing that, and that would have a commercial and potentially political value, would it not?

Juniper Downs

Let me talk a little bit about the spectrum of fake news, because the term is used to reference a lot of different categories of content, and then I can speak to how we deal with it and address your question. On one end of the spectrum, you have fake news farms, content that is produced in bulk, which there is a financial incentive to try to distribute as virally as possible. This pattern is caught by our spam detection systems. It violates our policies; we remove it. Then there is what I think you may be referring to, which is clickbait—content that might be more sensationalistic. We see videos with titles in all caps, or salacious thumbnails. We see this clickbait content. What we do there is demote it. Our algorithms are trained to identify clickbait and demote it. It is curious that you ask the question about watch time and user interest. What we expected, when we started demoting clickbait content, was that that would create a decline in watch time, and that was the initial result. We were willing to absorb that, because we did not think that it was high-quality content, so we were willing to absorb the loss of watch time, but—this may make you feel better about humanity—over time, the watch time actually picked back up and increased. So when we were surfacing the higher-quality content, people actually started consuming that as well. If you think further down the spectrum of what people call fake news, you also have, somewhere in the middle, various opinions that people express on YouTube about current events, political matters and so on. We do not consider that content to be news. It is subject to our content policies. If it crosses the line to hate speech or harassment, obviously we are going to take it down, but that is not news content, in our view.

Giles Watling

But you did say a little while ago that people watch what they want to watch. You are sort of on the back of that and you are not able to keep up with it. I put it to you that it is actually not in the interest of large corporations like yours to put a stop to this, because you are harvesting data.

Juniper Downs

This is really about the kind of site and service we want to provide to our users and the long-term play. It is not necessarily about short-term results. When it comes to news content, our commitment is to provide our users with authoritative information from established news sources. We want to demote that lower-quality content, not necessarily because we think that we will get more watch time. Your assumption that people will watch more clickbaity content because it is entertaining, or for whatever reason, may be accurate. That is not enough for us to prioritise it. Our goal is to demote it and to promote the more authoritative sources because we think it is a better long-term play for the business to be seen as a trusted provider of news and information to our users.

Giles Watling

Thank you.

Chair

A final couple of questions from me before we close this session. I notice that YouTube’s policy now is to put a label or a tag on content from broadcasters that are publicly funded in their countries, I think in response to requirements in the United States that that be done. Is that correct?

Juniper Downs

We recently started rolling out a new transparency feature for certain news sources. We made a public commitment to provide users with information about the sources of news they consume. We have started fulfilling that commitment by introducing a new label that informs users if a news outlet receives Government funding, and a separate label to inform users if a news outlet is a public broadcaster. It has just started rolling out, but we plan to extend that and think that providing transparency to users is an important part of what we can provide as a service.

Chair

There has been some criticism of this, because it means that RT, a Kremlin-backed propaganda station, is in the same bucket as the BBC, a publicly funded, editorially independent public broadcaster. Do you think mixing RT and the BBC together is making navigation easier for consumers or muddying the waters?

Juniper Downs

As we developed this new feature, we did reach out to various news outlets, including the BBC, and we got that feedback directly from them, which is why we created two separate labels: one that states that an entity receives Government funding in full or in part—that is what applies to RT—and a separate one that labels things as public broadcasters, which is what applies to the BBC. There are actually two distinct labels. Each label links to an online source where users can read more about the news source and get additional information about who funds them and what their structure is.

Chair

There is a degree of confusion here. I think some people reading that might say, “What’s all the fuss about RT? They’re a public broadcaster. They’re state-backed. So is the BBC. It’s all the same thing.” Actually, they are radically different things. It is not just about the ownership structure or the model of the organisation but about the way it is run and the degree of editorial independence it has, which exists at the BBC and does not exist at RT. Certainly from the evidence we have received from people while we have been in America this week, it is very strongly suggested that this is something that needs to be thought through again, because it may unwittingly sow more confusion rather than adding the transparency that you have asked for.

Juniper Downs

We agree that more information and context is better, which is why the label links to a site where people can get additional information, like what you describe, to distinguish some of these news providers from each other.

Chair

That’s all well and good, but given the way people consume content—we heard earlier about “Up Next” providing a constant stream of content for people to see—I am not sure that that is a service that people in reality will use that much of. What they will go on is the labels. If labels seem to be categorising content in a similar way, they may trust them in a similar way, and that may not be helpful to them in terms of making sure they are consuming the highest quality news. I think we have completed our questions for this session. Richard Gingras and Juniper Downs, thank you very much for your evidence. Examination of witnesses

Witnesses

Monika Bickert, Head of Global Policy Management, Facebook; and Simon Milner, Policy Director UK, Middle East and Africa, Facebook.

Chair

We will go straight into the second panel. Monika Bickert and Simon Milner, thank you very much for joining us here this morning for our evidence session. I will start by asking Monika Bickert for some background information about the Senate Intelligence Committee’s investigation into Russian activity during the election. Quite a lot of that work has been quite helpful and insightful for us and, as you will know, it has obviously led us to contact Facebook and ask for a similar analysis of whether there was any Russian activity in the UK. I want to ask about the information that Facebook has provided. Is it correct to say that all the evidence so far—the analysis of the number of people who were exposed to content created and distributed by the Internet Research Agency in St Petersburg—comes from the analysis that Facebook did, identifying payments made in roubles to promote advertising around that content?

Monika Bickert

Thank you for the question. The information that we gave to the Committee in testimony—it was my colleague Colin Stretch—is really the best source of information for the details about that. In that testimony, Colin described how Facebook conducted that investigation. Ads being paid for in roubles was one of the signals, but we considered other things as well. Mr Stretch’s testimony is really the best record of that.

Chair

From the information that we have received from the Senate Intelligence Committee, it is very clear that everything has been extrapolated only from those accounts where rouble payments were made. Analysis has been done to try to identify accounts linked to that behaviour, but Facebook simply analysed what some people might call the lowest-hanging fruit—linking it from the accounts where rouble payments had been made—and did no wider analysis of whether other agencies were involved or other similar activity took place that was not linked to rouble payments for promotional advertising.

Monika Bickert

I think I can just refer you to Mr Stretch’s statement. We also put out some posts, including in April 2017 and several over the summer—I believe one in August and two in September—in which we talked about how we looked for that sort of content.

Chair

I am pleased that Facebook has now agreed to conduct an analysis of whether Russian agencies were involved in distributing content on the platform that was linked to the Brexit referendum in the UK and other elections too. Just to note, however, as we are in the hearing now, that we expect that that will be based not just on whether rouble payments were made for promotional advertising but on an analysis of the origin of the content and whether it is likely to have come from a Russian agency due to the nature of that content and the way it was distributed. Rouble payments for advertising are one tool that could be used, but there are many, and we hope that the analysis will be wider than just that.

Simon Milner

Perhaps I could add to that. I can tell you that, as I explained previously in my letter to you, a second investigation is under way. It is not yet completed, but we expect to be able to report back to the Committee by the end of February on the results of that investigation. We will be prepared to share with you—possibly in private, because we do not want to tip off the bad actors about how we do those investigations—exactly how that work was undertaken.

Chair

Thank you. We look forward to seeing that at the end of the month. Monika Bickert, to what extent do you feel that the company has a responsibility to ensure that your customers are protected by knowing the source of the information that they see—where it is coming from and who is producing it?

Monika Bickert

We feel very responsible for letting our community know, first that they are in a safe community—some of the questions that were put to the previous panel spoke to questions of safety, and we can speak a bit about that. That is a huge priority. It is also a big priority for us to help people to connect with authentic information. From talking to our community, we know that that is something they care about, and it is something we care about. We are investing a lot, and not just in the policies for keeping our communities safe, which tend to be simple in the sense that it is fairly black and white—it either crosses the line into something that is unsafe and remove it, or it doesn’t and we leave it up. In the area of fake news, which is very different, we have developed a four-pronged approach where we are trying to make sure that when people come to connect with news on Facebook, it is reliable news and they have the ability to make informed decisions.

Chair

Do you feel that not just with news services but with other information that people might share—community pages, pages that may have political content—it should be really clear to people where those pages are being administered from? If a community page is giving me information about Kent in England, where I live, it should be clear that that is being run by someone who lives in England, and not by someone who lives in another country.

Monika Bickert

There is a spectrum of different types of what people might call news or information. One of the things that has been notable about social media is that it has given a voice to many people who want to share what is happening to them, especially in regions of the world where traditional news media outlets do not necessarily reach, or do not reach frequently. There is a spectrum of information. Our job is to try to make sure that people can connect with reliable information and make their own decisions about what they want to see.

Chair

Isn’t one of the ways in which you empower people to make those decisions for them to understand where the content is being created and whether the person creating that content is actually who they are pretending to be?

Monika Bickert

There are a couple of things that we try to increase transparency of source information. One thing that we do on Facebook, which is distinct from many other services, is have a policy that requires that you use your real name. If you think about the worst types of false news—the financially motivated spam that has links and that takes people to ad farms—that sort of content is very typically propagated by fake accounts. That transparency requirement in our policies is very important for removing those accounts. Another thing we do is look at using context as a way to inform people about their news source. This is something that we are testing right now. When people see information from a news source, if there is any signal to us that that news source might be unreliable, they can click on a little icon recalling the article’s context. We released this in November 2017. From that icon they are taken to information that is taken from across the internet about this source and its reliability.

Chair

On the point about labelling and transparency, people do set up fake accounts. You have identified fake accounts that have been taken down and there are very many of them.

Monika Bickert

We remove them every day.

Chair

So it is possible to do. But what would be the harm in making really clear the origin, at least, of material as you see it and consume it on the platform? Seeing where it is being created from may be a very big signal to you about whether it is a source of information that you should trust.

Monika Bickert

Yes, and because of our real name policy, people can see who is publishing that information.

Chair

As long as they are using their real name.

Monika Bickert

If they are using their real name. If they are not, we take that account down.

Chair

But people may not know and it could be there for a while. People do not necessarily know the country from where the page is being administered.

Monika Bickert

You are very right that we do not catch every fake account at its inception. We find and remove many fake accounts every day. This is also an area of tremendous technical investment for us and we have got a lot better at it. In the run up to the French, German and UK elections, we used our technical tools to remove thousands of fake accounts—not that those were necessarily related to spreading disinformation or information about the election. We are using those technical tools to reduce the chance that they might be used to spread disinformation.

Chair

Finally, I want to ask you the same question that I asked YouTube earlier on. What sort of proportion of your revenues from Facebook do you reinvest in identifying this kind of bad content?

Monika Bickert

I can’t give you a revenue percentage, but what I can say is that we just put out an earnings call where our CEO said that more than 14,000 people at Facebook are working on safety and security issues. That includes the engineers who are working on the technical systems to identify fake accounts, terror propaganda or other violating materials. It also includes the work of our content reviewers who are looking at the sort of content that has been reported to us and removing it if it crosses the line.

Chair

YouTube said that they were spending tens of millions of dollars on this sort of work. How much is Facebook’s investment in money terms?

Monika Bickert

I wouldn’t have a number to give you. This is something that is such a priority around the company that more than 14,000 people are working on it. It is not just a question of sponsoring a certain programme. This is these people’s jobs.

Chair

Do you know what the ad revenue is for Facebook for a year?

Monika Bickert

I believe our revenue for the last quarter was around $13 billion.

Paul Farrelly

All my questions are for you, Monika Bickert, because I do not easily get the opportunity to interview you in London. Therefore, hopefully the UK taxpayer will get the best value from my travel. You have just mentioned thousands of fake accounts in connection with the US election. All I have seen so far are the 470 connected with advertising, so could you elaborate on that statement?

Monika Bickert

My statement was that, in the run-up to the French, German and UK elections, we removed thousands of accounts using these enhanced fake accounts technical tools. We have been investing in this area for a long time. I want to be clear that the real name policy is not a new policy and using technical tools to find fake accounts is not new. I have been at the company for years and we have been doing this for years, but we have had a significant advancement in the past year. That has allowed us to remove those thousands of accounts in the run-up to those three European elections.

Paul Farrelly

So the thousands were not in relation to the US election?

Monika Bickert

That is right.

Paul Farrelly

So why only 470 with the US election? Who is better at French—you or the Russians?—than English?

Monika Bickert

I can refer you to the comments that were put to the Committee by my colleague, Colin Stretch. That is really the best place to find the information about that.

Paul Farrelly

You have not briefed yourself before this session?

Monika Bickert

That is part of an ongoing investigation. We are certainly co-operating with the relevant authorities there. We have given information and that is the best source of information for that.

Paul Farrelly

Do you do these sweeps in relation to specific events such as elections, or do you do them all the time?

Monika Bickert

We are doing them all the time.

Paul Farrelly

How many thousands or hundreds of thousands of fake accounts have you suspended that have no connection with specific events such as elections?

Monika Bickert

Without knowing precisely, that is probably the most common scenario. We remove many false accounts every day. Many of those are created, for instance, for the purpose of sending out spam links or engaging in other bad behaviour. We catch some of them at the time of creation and we stop them from creating the account. Others we can remove quickly, after identifying them using certain signals that our technical tools recognise.

Paul Farrelly

If you cannot give us the numbers now, could you provide us with a briefing as a follow-up?

Monika Bickert

We are certainly happy to follow up, but just to set expectations, when you remove accounts very quickly from creation, you do not necessarily know what the purpose of those accounts might have been. Any accounts that we find have been created en masse, or created where there are signals that they are not being accurate in their name or engaging in a false way, we remove regardless of why they came to Facebook.

Paul Farrelly

We will ask you for some statistics afterwards. In relation to the 470 and the $100,000, or a roubles equivalent, why did you accept that money in the first place?

Monika Bickert

With regard to the US inquiry, that is part of an ongoing investigation. We are continuing to co-operate with US authorities and that investigation will continue. For now I can simply refer you to Mr Stretch’s comments to the Senate Judiciary, and to the Senate Intelligence and House Intelligence Committees.

Paul Farrelly

The question is unanswered. Why did you accept the money in the first place? Do you not make any efforts to know your user or your advertiser?

Monika Bickert

We do. With regard to our systems generally, we can speak about ads and also user-generated content. When it comes to ads, every ad that comes to Facebook is reviewed either by automated or manual review before it goes live. An important component of advertising on social media is that it happens quickly, so we try to use these systems to find things such as bad content—an ad that might have a certain word in it—that would suggest that we should take the time to review it before it goes live. There are a combination of signals that might lead us to take that sort of manual review. After the ads go live, the review does not stop. We look at signals like—

Paul Farrelly

Hold on. You have taken money in the first place. If you do not know your advertiser or your user, how can you be sure that you are not in breach of international sanctions or money laundering regulations? What responsibility do you take?

Monika Bickert

We have a team that works very hard to make sure that, when it comes to taking money, we are complying with all laws such as those around sanctioned individuals and countries. As I mentioned briefly before, we also have a policy that requires accounts to be authentic, and the advertisers must have an account before they can purchase an ad on Facebook. Our advertising is a self-service model. What that means is that if you use Facebook and you have an account, you can run an advertisement. If you do so, that ad will be reviewed in some fashion before it goes live, and then we will look at additional signals after the ad goes live, including how people are interacting with that ad. Are they x-ing it out, are they reporting to us or are there other signals that might suggest it warrants further review?

Paul Farrelly

My son, who is now 19, opened his Facebook account at the age of nine, and has since been unable to change his birth date. Your checks on who signs up stumble at the first block. Regarding the integrity of Facebook, it is important for users and advertisers to be reasonably confident that they are dealing with real people, not fake people or bots. What can you do to improve your game in making sure that people are dealing with real people and not fakes?

Monika Bickert

That is a very, very important part of what we do. When people come to Facebook, they expect that they are interacting with real people. That is fundamental; it is a cornerstone of our service and it is something we invest very deeply in. Our minimum age for people to come to Facebook is 13, and we have automated systems to try to detect when a person is putting in a false birth date. As Simon said, some things we can follow up in private because we do not want people to game those systems. We have systems that restrict people from changing their birth dates, for a similar reason. It does not always function perfectly, but we catch many people that way who might try to come online earlier in violation of our policy. Detecting the fake accounts is something we are constantly investing in and constantly improving. Where we have safety-related policies such as not allowing people to share extremist content or not allowing people to bully others, some of that can be done by technical tools—like Google and YouTube, we are investing in that and seeing real gains from it—but a lot of it is also contextual and requires human review. That is also true with fake accounts. There are technical tools that allow us to identify that an account may be fake, and sometimes it is very simple to say it is, and we remove it. Other times it is not as simple, and we will send it to one of our reviewers, who will look at it and say, “Actually, it appears that this is real,” or, “We’re going to ask this person for identification,” or, “It does appear that this is fake,” and they will remove it.

Paul Farrelly

Final question: if I come to Facebook and complain that my identity has been stolen, how long in terms of your policy would it take for that page, post, site—whatever you call it—to be removed by Facebook so that I can be satisfied that you are acting?

Monika Bickert

The vast majority of all complaints that we get from people about violating content or violating accounts and that sort of thing are reviewed within 24 hours. If there is something we think is safety-related, it goes to the front of the queue. If your account were hacked, for instance, and you had that sort of concern, that is something we would respond to very promptly. If you are seeing an impostor account, that is also something we would attempt to respond to very quickly. Impostor accounts, obviously, could really wreak havoc. I will say that that is not always a simple inquiry. I have looked at our reviewers who do this work, and I have watched what they do. Often what it entails is that you have one account and another account, and the second person has said that the other is an impostor. You have to look not only at the dates the accounts were created and the content within the accounts to try to figure out which one is right, but you might have to ask for the upload of forms of identification to confirm. We try hard to resolve all of those within 24 hours, but we do not always hit that mark.

Paul Farrelly

Thank you. I think we all think you could do much, much better.

Julie Elliott

Before I get to what I am going to ask, I want to come back to what you said about taking a large number of fake accounts down. Have you done an analysis of which countries the fake accounts come from in the main?

Monika Bickert

You mean when I said in advance of the French election and the German election?

Julie Elliott

Yes.

Monika Bickert

There are various signals that our team uses, and we can follow up privately and walk through some of those signals.

Julie Elliott

Right. Certainly in the UK, there are significant implications around electoral law for people who are trying to influence our elections from outside, because all materials in elections are very regulated. So where the fake accounts are from is very important, so I would appreciate that, if you could find out.

Monika Bickert

Absolutely.

Julie Elliott

In the UK we had a very controversial referendum some 18 months ago. Do you believe that disinformation campaigns using your platform played a role in that referendum?

Simon Milner

I am more expert on that issue. We understand why people are concerned about it. Concerns have been raised in Parliament, and it is one of the reasons why, when the Electoral Commission approached us around October last year and said, “We would like you to assess whether or not there was indeed misinformation coming from another country—particularly Russia—associated with the referendum,” we were very keen to co-operate on that. We reported some findings from an initial investigation to the Chair of the Committee in December. I think it is fair to say he felt we had not done enough work and, having reflected on that, that is why we are now undertaking more work. So the answer is: we won’t be able to tell you until that work is completed, but we are committed to telling the Committee the outcome of that at the end of February. The one other thing I would say is, unlike the US election, we have still not been furnished with any intelligence reports from the UK authorities to suggest that there was direct Russian interference using Facebook in the Brexit referendum. That is quite different from the US, where there is an intelligence report demonstrating this.

Julie Elliott

If we put Russia to one side, obviously the referendum was about our membership of the European Union—or not—going forward. So countries within the EU other than the UK probably had more of a vested interest. Are you looking at any countries, or just at the Russian influence?

Simon Milner

We have been specifically asked to look at Russia. I have not been aware of any parliamentary debate, any newspaper stories or anything else suggesting that countries apart from Russia might have been doing this. If you have got something you want to furnish to us—

Julie Elliott

It is not whether we are suggesting anything. I am asking, as a company, are you looking at that?

Simon Milner

No, we are not.

Julie Elliott

You are not. There was, very clearly, misleading information on Facebook during that referendum, because, anecdotally, every time you knocked on a door and somebody would tell you something—it would be absolute nonsense—you would say, “Where has that information come from?” and they would say, “Facebook”. That was every time. What as a company can you do to stop that proliferation of false information getting shared and re-shared? I think people were buying into it because those sharing it were people they knew and trusted, but goodness only knows what the source was.

Monika Bickert

Thank you for the question. I want to be clear that we do not accept that sort of proliferation of false content on Facebook. At the same time, I want to make sure that we are distinguishing between the sorts of content like extremist content and bullying content, where there is a bright line of leave it up/take it down, versus fake news. When we hear the term “fake news”, as you point out, many people will share stories of things they have seen online, and those range everything from the financially motivated spammers linking off-site to ad farms, which is the most common type of false news we see on the platform, down the line to the sensationalist headlines where the underlying story may be based in truth and perhaps given a spin or using certain words to get people to click on a headline. We cannot have one policy that addresses all of that; we have to have a more nuanced approach. Since this has become a topic of interest and concern, especially over the course of the past year and a half or so, we have developed a four-pronged approach, which I am happy to walk through. If we think about why it does not make sense to have just one approach, you can take this to the point where you say, “What if we had a policy on Facebook where we required people to only post what they knew to be true and accurate?” Of course, that would lead to a place where not only was it unenforceable, because we would not know whether something an individual posts about his or her observations and daily life was true or false; it would also really inhibit the type of speech that we all engage in on a daily basis, including predicting, speculating or sharing opinions on things. We have to tread carefully. So the first thing we do is remove false accounts and known bad actors from that far end of the spectrum that I talked about. If we do it well—we are not perfect at it—that takes care of a lot of that content. The next thing we try to do is disrupt the financial incentives for these sorts of actors to come to Facebook. I mentioned earlier that they might post links that appeared to be exciting articles, but were clickbait articles that would take you off-site to an ad farm. We are getting better at detecting those ad farms and at detecting things like publishers inserting clickbait into their headlines or something that looks like a video that plays, when in fact the play button is a ruse to get people to click on it and be taken to an ad farm. Our systems are detecting that and removing it. The next thing we are trying to do is prioritise the visibility of content that is trustworthy, and specifically that is deemed trustworthy by our community. That is something we are testing right now, but we are very interested in it. We are trying to reduce the visibility of content where we have a reason to suspect that it is unreliable. People can report to us when they believe news to be fake. We are using a system of fact-checkers, which is growing. If we have an indication that news is fake, we reduce that visibility up to 80% in people’s news feed. The final thing we are doing, which I really want to underscore because I think it is so important in the long term for all of us, is trying to improve the ability of the broader community—not just people using Facebook, but journalists, policy makers, educators and parents—to fight false news by recognising it, distinguishing among news sources and being able to make those responsible choices. We are doing that in the UK. We have been working with young people—we have digital ambassadors in schools, talking among other things about how to recognise false news. In the run-up to the election, we ran Full Fact’s top 10 tips for spotting fake news. That was not just something we did on Facebook; we actually went out in traditional media and published these top 10 tips to help people make responsible choices. It is also something that we have been doing over the past year with the National Literacy Trust in the UK. We are trying to research and understand the best ways to help people to recognise false news and disrupt it.

Julie Elliott

As a company, do you think you have any responsibility for the misinformation that is on your platform? I could take you to half a dozen sites that have absolute nonsense on them about the area I live in, but they get re-shared and everything. That is influencing people’s voting behaviour and therefore influencing elections. I would say that that is having a very negative effect on the democratic process, which is between people of political parties who stand on their platforms. Do you think you have any responsibility to try to sort this mess out?

Monika Bickert

It is very important to us. I mentioned the broader initiatives, but I want to emphasise that we are engaging with the community and trying to learn more about what they are seeing and the way that this news manifests itself. We are exploring and testing different options—I have mentioned a few of them. In addition to the article context icon that you can click on to find information about the publisher, we are also testing a way of giving people information about related articles. If you are on Facebook and you see a news article about a particular topic, underneath that you will see a spectrum of related articles on the same topic, so that you know whether the article you are seeing fits with mainstream sources or not. We are looking at ways of incorporating brand logos, so that if you find content on Facebook—if it has surfaced to you—you know exactly who it is coming from and whether or not you recognise or trust that news source. This is not an area where we are done; it is definitely an area where we are investing in learning and understanding and in testing different options.

Julie Elliott

Thank you.

Chair

I would just like to return to a couple of points that Simon Milner raised. For the record, my complaint about the analysis that Facebook did is that all Facebook had done when looking at the Brexit referendum in the UK was to go back to the accounts that had already been identified as part of the US Senate investigation, and look to see whether those self-same accounts had been active in the UK during the Brexit referendum—nothing else.

Simon Milner

Yes. I did not mean to dismiss what you said. You were very clear about your assessment, and your assessment and the views of other parliamentarians were a significant factor in why we are now doing another investigation.

Chair

And the view of the British Government, too. You also mentioned that the work that had been done in America on identifying the accounts that have been identified was based on intelligence that Facebook was given.

Simon Milner

I said that there was intelligence about Russian interference in attempts to interfere in the US presidential election, and that was a factor in the work that was done. I did not say that their accounts were developed. But I also—

Chair

Sorry, what do you mean by intelligence?

Simon Milner

An intelligence report was produced by the US authorities about attempted Russian interference in the US election. A similar report has not been produced by UK intelligence about attempts at Russian interference in the Brexit vote.

Chair

When I met yesterday with Senator Warner and Senator Burr, they were slightly surprised by that presentation of events. As far as they were concerned, there was no presentation of intelligence by the American Government or by them to Facebook, seeking to identify accounts that had been problematic. The accounts that were identified were simply a response by Facebook to pressure they were receiving from the Senate, which led them to do the bare minimum, which was simply to look for rouble payments that the site had received. This was not based on a dossier of intelligence that had been received externally. That is why, when we first discussed this, I felt that Facebook, in response to a request from Parliament or the Government, should conduct its own research and not be reliant on other people giving you intelligence. Ultimately, this is your system and your platform, and you know it better than anyone else.

Simon Milner

Absolutely. I was not suggesting that therefore, this had a bearing on our ability to look properly at our systems. That is exactly what we are doing. I was just explaining the wider context of the UK situation compared to the US situation.

Chair

Yes, but you were insinuating that there was a lack of intelligence in the UK that had existed in America, and that the absence of intelligence support from the UK meant that that work had not already been done. But in America, it was not intelligence reports from the Government that led to that work being done; it was pressure from Congress, which led to what they would see, I think, as the company doing the bare minimum. I am pleased that at least now Facebook is prepared to initiate that research in the UK, but I felt that the exchange that we had earlier did not necessarily give the clearest view of what had actually happened.

Simon Hart

I have more of the same. You recently announced that there would be 1,000 extra people to vet political advertisements. What problem had you identified that led you to make that decision?

Monika Bickert

Thank you for the question. When we looked at the advertisements that we identified in the wake of the US election, one of the questions we asked ourselves was “Are we doing enough to identify when ads might be coming from bad actors?” There are a couple of things that we asked ourselves. One is, “Are our policies in the right place about political ads?” We have actually decided to tighten those policies overall, for anybody who might run a political ad, to make sure that we are not allowing ads that inadvertently, for instance, contain hate speech. The second thing that we asked ourselves was “Are we looking—“

Simon Hart

This was in response to some evidence, presumably.

Monika Bickert

This was in response to us overtaking a broad look, in the wake of the US inquiry, at our advertising systems and saying, “How do we prepare going forward for what might happen during elections?” So it was not necessarily because of specific ads that we saw in the US-related investigation. The second thing that we asked ourselves was, “Is our review process holistic enough?” We do not want one reviewer looking at the content of the ad and another reviewer looking at how the ad is being targeted; we want to make sure we have one source of information for understanding who is behind an ad, how that ad is being run, what is in the face of it and who it is targeting. We have made some structural changes to our review, and part of that requires investing more in our reviewers. That is why we are adding those additional people.

Simon Hart

You mentioned in response to an earlier question that there was a lack of intelligence, or no intelligence had been passed on to you, about the possibility that UK elections and referendums might have been impacted by political advertising from sources that had mischief in mind or were from other countries. Presumably, there was therefore no intelligence that that was not happening. You just do not know.

Monika Bickert

I am sorry; would you mind reframing the question?

Simon Hart

Mr Milner’s point was rather a shrug of the shoulders; he said that no evidence was passed to them that anything went wrong with the EU referendum. There was presumably no evidence either way, so do you actually know whether there was a problem or not?

Simon Milner

That is exactly what we are looking into. Just to be clear, when the Electoral Commission approached us and said they would like us to look into this, we specifically asked whether they had any examples—had any member of the public, any parliamentarian or anybody from any campaign highlighted to them a page or an ad that was seen during that election that they felt was fishy? Perhaps it did not feel like it was coming from one of the official campaigns or came from somewhere that they just did not recognise? The Electoral Commission was very clear that it had had no complaints. Again, I am not suggesting that there is nothing, but until we have completed this investigation, we will not know. What we have not had is information that has enabled us to target a particular page or a particular phenomenon from another source. To Mr Collins’s point, that does not mean we are not looking very thoroughly; we absolutely are.

Simon Hart

One of the pitches that your company makes to us, as parliamentarians in the UK, is the effectiveness and cost-effectiveness of targeted advertising, so it clearly works, because you are encouraging us to do it most of the time. However, it is quite subliminal. It enables me to target people over the age of 65 with an interest in fishing in my constituency, for example. That is quite a subliminal method of political advertising. Are you happy, in pursuing that, that you are within electoral guidelines, particularly in relation to spending limits, which are obviously quite rigid in the UK?

Simon Milner

I had better take this. We have a very good relationship with the Electoral Commission, by the way. This is not the first time we have met them. We have worked with them on encouraging people to register to vote, reminding people it is election day and so on. We have a good relationship with them, and they are of course the ones who are principally responsible for determining how much money people have spent and where that money came from. That is not something we can necessarily see. However, we absolutely agree with you that there is an issue with the transparency of political advertising. Can you see what your opponent in your constituency is saying to voters, and can you respond to that, if the advertising takes place on Facebook? That is one of the reasons why we are now rolling out a system of transparency around political advertising, such that, in due course—at the next general election in the UK, for instance—you will be able to see every ad that is being run by both the main campaign pages and by all candidates. If you want to see what ads they are running on Facebook, you can see them. We are going to introduce a radical new level of transparency that has never been seen before in elections. We did not have that last time, and we recognised that it would be very valuable for people to address exactly the issue you are concerned about.

Simon Hart

You will have seen this week that the PM said that online platforms “are clearly no longer just passive hosts of the opinions of others” and that she believes it is right that we look at the liability of social media companies for illegal content on their sites. That reflects comments made in other parts of the world as well. Do you read that as being that the age of unregulated social media is actually coming to an end?

Simon Milner

I think one reads a number of things into that. We certainly do not think of ourselves as unregulated social media. Indeed, that is why Monika and her team are responsible for an incredibly extensive set of regulations that apply to our services.

Simon Hart

They are self-imposed. They are not independent regulations, are they?

Simon Milner

They are self-imposed, but I can give you an example of how they fit within a broader structure.

Simon Hart

That is not my point. I do not want to know that. We are talking about regulation that is accountable, democratic and transparent.

Simon Milner

It is certainly transparent, because all our rules are public. It is accountable in that I am here often, and in front of other parliamentary Committees, as are my colleagues, to answer your questions on that. It is accountable in that we let people know, when they complain about content, what decision we have made and why, and if someone’s content is removed, we let them know why we have done that. There are actually multiple layers of accountability. Yes, there is not a single body from any country, or globally, that has said, “Okay, these are the rules that are applied to Facebook” from outside. It is very hard to see how that would work, but many countries have applied laws and rules that we take account of, and we ensure that people do not break those rules on Facebook.

Simon Hart

That strikes me as a pre-emptive way of opposing anything that might resemble state regulation. Do you fear outside regulation? If legislation did come forward that sought to apply a light-touch regulation to your business, in the light of some pretty overwhelming evidence, would you co-operate fully with that, or would you be minded to resist?

Monika Bickert

We would certainly want to be part of the dialogue, because from time to time we do see legislation that results in some unintended consequences that are not good for anybody. We would also note that our incentives are generally aligned with those of Governments. I was in the US Government as a federal criminal prosecutor for more than a decade before I came to Facebook. When I came to Facebook, it was right in step with a lot of the things I worked on as a prosecutor. All of the criminal behaviour and other things that we try to find and remove from our service were very much aligned with the incentives of policy makers. Like Simon said, we do have a process for complying with Government regulations. If a Government tell us that something is illegal on our service, the first thing we do is see if it violates our community standards. If it is something like terror propaganda, it would violate our standards, and so we would remove that globally. If it does not—let us say the law is about something that does not quite violate Facebook standards—we would look at the legal process and we might restrict the content in that country out of respect for the local law. So we do have a system in effect. I would note that sometimes regulations can take us to a place—you have probably seen some of the commentary about the NetzDG law in Germany—where there will be broader societal concerns about content that we are removing and whether that line is in the right place. So we would want to be a part of the dialogue.

Rebecca Pow

Ms Bickert, isn’t the clever part of Facebook that you get people to sign away their data and their rights to their data, and you find out everything you can about them?

Monika Bickert

No. Not only do we not sell people’s data, which we never have and we never—

Rebecca Pow

I am not saying that you sell it. You get them to give you—they do not even necessarily know that they are giving you data. You are harvesting data about people every single time they use Facebook.

Monika Bickert

No, I would not characterise it that way. We are very clear in our data use policy how we do use data. We also allow people to see any information that we have. If you go to Facebook, you can go to “download your information”—it is literally called that—and you will see everything that Facebook has. So we are very transparent about how this works.

Rebecca Pow

But you gather all this data, whether they are playing a game, watching a video or whatever, working out what they are doing and gathering a profile, so then you can target them back with information that they might want. Isn’t that the way you work?

Monika Bickert

The way targeted advertising works—we do allow that; this is how our advertising model works—is that advertisers will say they want to target this particular audience, for instance people who have liked their page, and we will provide an audience. We do not provide people’s data, but we will provide an audience of people who fit that targeting criteria. That is the way the model works.

Rebecca Pow

You are gathering a massive amount of data about the whole of society. Isn’t this a massive surveillance operation?

Monika Bickert

No, this is a system where people can come and communicate with one another and advertisers can target people based on the interests that they have expressed and the way that they have engaged on the site.

Rebecca Pow

There is an article in The Times today in the UK and the headline is, “How to win 2 billion friends and destroy civilisation…the biggest surveillance enterprise in…history”. It is by a very well-known journalist. He suggests that, whether Facebook likes it or not, you have got yourselves into this position. You hold so much data about people that you are now very powerful, and you are completely unregulated.

Monika Bickert

We are definitely regulated in many different ways, and we are careful to—

Rebecca Pow

By yourselves?

Monika Bickert

No, by laws. Simon can speak to the UK.

Simon Milner

In respect of data, we are fundamentally regulated by European data protection law. Every single person on Facebook in the UK is covered by that data protection law, and we are absolutely accountable to it. It is completely wrong to suggest that, in respect of how we handle people’s data, we are unregulated. The opposite is true.

Rebecca Pow

The Times article focuses on Channel 4, which is the biggest news outlet that has been using Facebook to post videos. It gets massive hits—2 billion views. You have placed adverts around those videos, sensibly, because you know they are going to get loads and loads of hits. I believe Channel 4 asked whether you might share the data of who was looking at those adverts so that they could get some information about their own audience. Will you share that data with them?

Monika Bickert

That is actually not how our ads work. Unlike some other services, we do not place ads next to specific pieces of content. That is just not how it works. You target an audience—

Rebecca Pow

That is not what they told me, but anyway.

Monika Bickert

It does not work that way. The way it works is that advertisers create a target audience, and people who meet that audience will see an advertisement, for instance in their news feed. The ad is not tied to a specific piece of content.

Chair

Just to be clear on that, there is ad space around the content. That is the point Rebecca Pow was making. People are buying an audience rather than saying, “I want to buy Channel 4 content,” but there is advertising adjacent to the Channel 4 content.

Monika Bickert

That is correct.

Simon Milner

Yes.

Chair

That is the point that Rebecca Pow was making.

Rebecca Pow

Yes, that is exactly the point I was making.

Monika Bickert

We do not run ads on pages—that is not something we do—but if somebody is in the audience for an advertiser, from the criteria the advertiser submits, they might see an ad in their news feed.

Rebecca Pow

I believe that you actually asked Channel 4 whether they could put some of your adverts in the middle of their videos, but they refused.

Monika Bickert

We may be speaking about something different here, which is an audience network, where there are ads off Facebook.

Simon Milner

It could well be. Perhaps we could follow up afterwards with you, because we are not aware of the specifics of the case.

Rebecca Pow

That would be useful. Sometimes, these videos are taken by other users, posted on and altered. I wonder where you stand on copyright issues in relation to things like that.

Monika Bickert

We have policies against anyone infringing others’ intellectual property rights, and we have systems in place to detect that. Certainly, upon notice, we have a notice and takedown procedure.

Rebecca Pow

There has been a rather unpleasant incident of child pornography getting into the Channel 4 network through the Facebook Messenger service. That has actually been spread far and wide, and thousands of people have seen it. How did that happen? How did that get through your systems?

Monika Bickert

The way we deal with any sort of child sexual abuse imagery is that, first, we have systems in place to automatically identify such content. Those systems are not perfect. Perhaps I can explain the way they work. They primarily work on matching known images. For instance, we work with the National Centre for Missing and Exploited Children and the International Centre for Missing and Exploited Children, and if we become aware of an image of child pornography, we reduce that to a hash—basically a digital fingerprint—and we can stop it being uploaded to Facebook in the future. We do not necessarily have the technical means to stop new images, or images of child exploitation that we have not seen before, being uploaded. As soon as we become aware of such an image, whether it is from someone reporting it to us or from law enforcement sending it to us, we will take steps to stop its dissemination and we will report it immediately to the proper authorities so they can take action.

Rebecca Pow

It actually took you 36 hours to take this down. How did that happen?

Simon Milner

I can talk to that, because I spoke with the editor of Channel 4 News about this incident. We ought to use this opportunity to say to anyone who is watching the live stream: if you ever come across a piece of content like this, do not share it. Report it to the IWF—that is what they are there for—and they will ensure it gets taken off the internet. Unfortunately, a lot of people shared this to condemn it, and were encouraged to do so. Every person who did that broke the law in the UK and in many other countries by doing so. Once we were made aware of it by “Channel 4 News” and once they told us about which in-boxes it was in, we were able to move very quickly to remove it from the in-boxes—to hash it, as Monika explained—and to prevent thousands more shares of that horrible material on Facebook.

Rebecca Pow

Doesn’t this indicate just how powerful you have become and what a Pandora’s box—to use the term my colleague used earlier—you have opened up? And doesn’t it demonstrate that perhaps it is time for regulation and rules that set out the roles and responsibilities, and all of these things, because all of this is affecting our society, not least our children?

Simon Milner

I would encourage you to ask the IWF about this, because what they will tell you is that they see tiny, tiny amounts of this material being shared on platforms like Facebook. There are much more significant concerns about other aspects of the internet. Despite our big size—2 billion people using Facebook, 40 million people using it in the UK—these kinds of incidents are incredibly unusual. To suggest that this is a Pandora’s box, or something we are not in control of, is completely wrong.

Rebecca Pow

My final question is, how long do you hold people’s data for—the data you collect about them? Is there a limit?

Simon Milner

Yes—as long as they want us to. Mr Farrelly mentioned that his son has been on Facebook for 10 years. We have held his data for as long as he wants us to. If he wants to go back and look at data from when he first joined, he can do. But as soon as people decide they do not want to be on Facebook anymore, then we will remove that content. It is entirely up to individuals about how long they want us to hold it for.

Rebecca Pow

So he was a child when he signed up. How long do you hold data on children and do you remove it once they become an adult?

Simon Milner

It depends whether they want us to. It is entirely up to them. Remember, we are holding data for them that is very precious. A lot of that data may be photos and videos and family moments—important moments in their lives—and we are custodians of that for them. It is up to them to decide when and if that data is removed from Facebook. We will not make the decision for them.

Rebecca Pow

You do sound rather god-like, saying you are “custodians” of that data.

Chair

Let us move on. But just to be clear, when you say that it is for as long as they want, by that you mean for as long as they have a Facebook account.

Simon Milner

Yes. Exactly.

Chair

So as long as they have a Facebook account, you hold that data.

Simon Milner

Yes. We look after it for them.

Monika Bickert

Although, for instance, if you removed a specific post from your account, then we would delete that in accordance with the relevant data deletion laws.

Chair

Okay. Thank you.

Ian C. Lucas

Mr Milner, earlier you said something very interesting, which I greatly welcome, concerning the election law, because in 2015 and 2017, and in the 2016 referendum, Facebook advertising was extremely important. In those elections, do you agree that it was not possible to establish, as a candidate, where another candidate’s purchase of Facebook’s advertising was bought from?

Simon Milner

I do not understand what you mean. Can you explain further? Can you give me an example in your constituency?

Ian C. Lucas

In my constituency, my opponent can purchase advertising from Facebook in the campaign. It is unlawful for someone to pay for that advertising from outside the UK. Also, all of the information has to be recorded within my particular district. To date, and as we stand today, it is impossible for me as a candidate to check where that advertising was bought from. Do you agree with that?

Simon Milner

Is that also true for the pamphlets that he paid for? Actually, was your opponent a man or a woman?

Ian C. Lucas

No—they have imprints on them.

Simon Milner

Right. But you still do not know how they were paid for, so I am not sure that it is any different from other forms of advertising—