ETHI Committee Meeting

     Good day, everybody. We're at the Standing Committee on Access to Information, Privacy and Ethics, meeting 148, pursuant to Standing Order 108(3)(h)(vi) and (vii), a study of election advertising on YouTube.

    Today we have with us, from Google Canada, Colin McKay, head of public policy and government relations. We also have Jason Kee, public policy and government relations counsel.

    Just before we get started, I want to announce to the room that the release went out at 3:30, so it's going out as we speak, with regard to the matter that we dealt with on Tuesday, so watch for that.

    We'll start off with Mr. McKay.

     Thank you, Mr. Chair, and thank you very much for the invitation to speak to you today.

    I'd like to start off with an observation. First and foremost, we would like to clarify that we feel there is an inaccuracy in the language of the motion initiating this study. Specifically, the motion invited us to explain our “decision not to run ads during the upcoming election” and our “refusal to comply with Bill C-76”. To be clear, our decision to not accept regulated political advertising is not a refusal to comply with Bill C-76 and the Canada Elections Act, but rather was specifically taken in order to comply.

    Free and fair elections are fundamental to democracy, and we at Google take our work to protect elections and promote civic engagement very seriously. On cybersecurity, we have developed several products that are available to political campaigns, elections agencies and news organizations free of charge. These include, as I've mentioned to you before, Project Shield, which uses Google's infrastructure to protect organizations from denial of service attacks and our advanced protection program, which safeguards accounts of those at risk of targeted attacks by implementing two-factor authentication, limiting data sharing across apps and providing strong vetting of account recovery requests. These are over and above the robust protections we've already built into our products.

    We have also undertaken significant efforts to combat the intentional spread of disinformation across search, news, YouTube and our advertising systems. This work is based on three foundational pillars: making quality count, fighting bad actors and giving people context.

    I'll turn to my colleague.

     We are making quality count by identifying and ranking high-quality content in search, news and YouTube in order to provide users the most authoritative information for their news-seeking queries. This includes providing more significant weight to authority as opposed to relevance or popularity for queries that are news related, especially during times of crisis or breaking news.

    On YouTube, this also includes reducing recommendations for borderline content that is close to violating our content policies, content that can misinform users in harmful ways or low-quality content that may result in a poor user experience.

    We are fighting bad actors by cutting off their flow of money and traffic. We are constantly updating our content and advertising policies to prohibit misleading behaviours such as misrepresentation in our ads products or impersonation on YouTube and to prohibit ads on inflammatory, hateful or violent content or that which covers controversial issues or sensitive events.

    We enforce these policies vigorously, using the latest advances in machine learning to identify policy-violative content and ads, and we have a team of over 10,000 people working on these issues.

    While diversity of information is inherently built into the design of search news and YouTube, each search query delivers multiple options from various sources, increasing exposure to diverse perspectives. We are also working to provide users further context around the information they see. These include knowledge panels in search that provide high-level facts about a person or issue; content labels in search and news to identify when it contains fact-checking or is an opinion piece; and on YouTube, dedicated news shelves to ensure users are exposed to news from authoritative sources during news events and information panels identifying if a given channel is state or publicly funded, and providing authoritative information on well-established topics that are often subject to misinformation.

    ln relation to elections, we are partnering with Elections Canada and Canadian news organizations to provide information on how to vote and essential information about candidates. We will also support the live streaming of candidate debates on YouTube and we are creating a YouTube channel dedicated to election coverage from authoritative news sources.

    Our work to address misinformation is not limited to our products. A healthy news ecosystem is critical for democracy, and we dedicate significant resources to supporting quality journalism and related efforts.

    The Google news initiative has developed a comprehensive suite of products, partnerships and programs to support the news industry and committed $300 million to funding programs. We are also supporting news literacy in Canada, including a half-million-dollar grant to the Canadian Journalism Foundation and CIVIX to develop NewsWise, a news literacy program reaching over one million Canadian students, and a further $1-million grant announced last week to the CJF to support news literacy for voting-age Canadians.

    We're funding these programs because we believe it's critical that Canadians of all ages understand how to evaluate information online.

     In line with this, we fully support improving transparency in political advertising. Last year we voluntarily introduced enhanced verification requirements for U.S. political advertisers, in-ad disclosures for election ads, and a new transparency report and political ad library for the U.S. mid-terms. We deployed similar tools for the Indian and EU parliamentary elections. While we had intended to introduce similar measures in Canada, unfortunately the new online platforms provisions introduced in Bill C-76 do not reflect how our online advertising systems or transparency reports currently function. It was simply not feasible for us to implement the extensive changes that would have been necessary to accommodate the new requirements in the very short time we had before the new provisions took effect.

    First, the definition of “online platform” includes any “Internet site or Internet application” that sells advertising space “directly or indirectly”, and imposes the new registry obligation on any platform that meets certain minimum traffic thresholds. This captures not only social media or large online advertising platforms, but also most national and regional news publishers, virtually all multicultural publications, and most popular ad-supported websites and apps, making its application extraordinarily broad.

    Second, the provisions specifically require that each site or app maintain their own registry. Unlike some companies, Google provides a wide array of advertising products and services. Advertisers can purchase campaigns through Google that will run on both Google sites and/or third party publisher sites. These systems are automated. Often there is no direct relationship between the advertiser and the publisher. While the page is loading, the site will send a signal that a user meeting certain demographic criteria is available to be advertised to. The advertisers will then bid for the opportunity to display an ad to that user. The winning advertiser's ad server displays the winning ad in the user's browser. This all happens within fractions of a second. The publisher does not immediately know what ad was displayed and does not have immediate access to the ad that was shown. To accommodate the new provisions, we would have had to build entirely new systems to inform publishers that a regulated political ad had displayed and then deliver a copy of that ad and the requisite information to each publisher for inclusion in their own registry. This was simply not achievable in the very short time before the provisions took effect.

    Third, the provisions require the registry to be updated the same day as the regulated political ad is displayed. This effectively means that the registry must be updated in real time, as a regulated political ad that was displayed at 11:59 p.m. would need to be included in the registry before midnight. Due to the complexities of our online advertising systems, we simply could not commit to such a turnaround time.

    A final complication is that “election advertising” includes advertising “taking a position on an issue with which a registered party or candidate is associated”. These are generally referred to as “issue ads”. Issue ads are highly contextual and notoriously difficult to identify reliably, especially as the definition is vague and will change and evolve during the course of a campaign. Given these challenges, we generally prohibit this class of advertising in countries where it's regulated, such as our recent prohibition in France.

    We wish to stress that our decision to not accept regulated political advertising in Canada was not a decision we took lightly. We sincerely believe in the responsible use of online advertising to reach the electorate, especially for those candidates who may not have a sophisticated party apparatus behind them, and for legitimate third parties to engage in advocacy on a range of issues. It is also worth noting that any time we opt to no longer accept a category of advertising, it necessarily has negative revenue impacts. However, after several months of internal deliberations and explorations of potential solutions to try to otherwise accommodate the new requirements, it became clear that this would simply not be feasible in the few months we had available. Consequently, it was decided to not accept regulated political ads, and focus our efforts on promoting civic engagement and other initiatives.

    In the coming weeks, our decision to not accept regulated political advertising in Canada will be formally reflected in our ads policies. We will continue the process of notifying all affected parties of the change. Similar to other ads categories that we don't accept, the policy will be enforced by a combination of automated systems and dedicated ads enforcement teams, who will undergo rigorous training on the new policy. We will also continue our work with Elections Canada and the commissioner of Canada elections on interpretation and enforcement matters and the relevant industry organizations that are working on measures to assist online platforms and publishers with the new obligations.

    We appreciate the opportunity to discuss our elections activities in Canada and our decision to prohibit regulated political advertising.

    Thank you.

    Thank you to both of you.

    First up in our seven-minute round is Mr. Erskine-Smith.

    Thanks very much.

    I understand that Facebook and Google together are 75% of digital ad revenue. The decision of Google to not accept political ads is thus pretty significant for the upcoming election. Do you agree with that?

     We think it's significant for us to take a decision like this. However, that number is generalized. It may not reflect the market for political advertising.

    It's a significant decision for the Canadian election.

    Now, I want to contrast and compare two really large companies that operate in this space.

    Have you read the recent report from the OPC, on the Facebook-Cambridge Analytica breach?

    Yes.

    Okay. Take that as an example. If the Office of the Privacy Commissioner investigated Google, and made recommendations consistent with what took place with Facebook, would Google be complying with the recommendations the OPC made?

    I would say that we have historically worked with the privacy commissioners to arrive at agreed statements of finding, and then implemented them.

    Right. I think there's a higher standard that Google has set for itself. The Privacy Commissioner says that the privacy framework at Facebook was empty, and is empty. You don't consider Google's privacy framework to be empty, do you?

    We are two very different companies, with two very different approaches to data protection and privacy.

    Again, you hold yourself to a higher standard. Remind me: How much money did Google make last year?

    I don't know, offhand.

    I have $8.94 billion in income, just in Q4 of 2018. We have a company that is raking in billions of dollars, and holds itself to a higher standard than Facebook, on a number of different issues. Yet, Facebook is able to implement the rules under Bill C-76.

    Why can't Google?

    As Jason touched upon just now—and he touched upon the Senate, when they were considering the amendments to Bill C-76—our systems, and the range of advertising tools we provide to advertisers, are much broader than Facebook's. I can't speak to Facebook's decision in this regard.

     I can say that we spent an intensive amount of time this year trying to evaluate how we would implement changes that would meet the obligations of Bill C-76. Because of the breadth of the advertising tools we provide—and Jason identified the number of different ways that touches upon our publishing partners and advertisers—we reluctantly came to the decision that we would have to not accept political advertising this year.

    When you say this year, you mean that you're actively working on this, to ensure that you will accept political ads in future Canadian elections?

    This is an important point. We do feel committed to encouraging strong and informed political discourse. It was a very difficult conversation for us.

    Is that a yes?

    Yes.

    You will accept political ads in the next federal election after 2019.

    What I will say is that we are trying to evolve our products to a point where we reach compliance with the Canadian regulations. At the moment, we can't do that. It's a tremendously difficult task.

    What about Washington state? Are you going to run political ads at the local level in Washington state?

    We didn't during the last cycle.

    Are you committed to doing so in the next cycle?

     I can't speak specifically to the obligations of Washington state. I don't know if they evolved.

    Here's my frustration. You have a company that makes billions of dollars, and looks at a small jurisdiction in Washington state and a small jurisdiction in Canada, and says, “Your democracy doesn't matter enough to us. We're not going to participate.” If a big player decided to change the rules, I guarantee that you would follow those rules.

    We are too small for you. You are too big. You are too important, and we are just not important enough for Google to take us seriously.

    I'd contest that observation, because as we mentioned, there have been other examples where we've had to make this decision. We don't do it willingly. We look for every route we can to have that tool available to voters.

    What's the largest jurisdiction where you have made a decision like this?

    As Jason just mentioned, the advertising has to be blocked in France. That's the reality. The reality for us here is not a commitment to democracy in Canada. The reality here is the technical challenge we confronted, with the amendments to the Elections Act. The internal evaluation resulted in the decision that we can't implement the technical challenges in time for the election cycle.

    I accept that timing is a constraint. What I struggle with is when you don't give me a direct answer when I ask whether you are committed to doing so for the next federal election, or for the next election at the local level in Washington state.

    That is an obvious frustration. Can you say, “Yes, we're committed to doing so. We'll fulfill that,” as a clear answer to my question?

    The reason I paused in replying to you is that in a parliamentary system, we have fixed election dates. But conceivably, there could be an election date within the next six, nine or 18 months.

    Okay, but let's say you're looking at a four-year cycle. Do you think that's reasonable?

    What I'm saying to you is that we work to improve all of our elections tools, and to meet the expectations of our users, and especially of regulations. Our intent is always to increase both the quality and the breadth of those tools.

    As we look at the obligations under the Elections Act, our intent is to try to reach those standards. We're faced with a time frame right now where we couldn't do that for this election.

     I understand. By 2023, I expect that you'll be able to do that.

    Yes.

    It's a shame that you were unable to do so, but Facebook was able to figure it out.

    The last thing I want to ask about, just because you've raised it, is recommended videos on YouTube. You've recently made a decision. After many years of not considering this to be a problem, in January of this year you decided that borderline content would not be recommended. Is that right?

    That's correct.

    I would dispute the characterization that this isn't something that we considered to be an issue. We've been examining this for a while and doing various experimentation with the recommendation system in order to improve the quality of the content the users would see.

    Then you disagree with the—

    The borderline content policy was introduced earlier this year. That's correct.

    I was referring to and taking my direction from a number of past Google employees who were quoted on Bloomberg suggesting that you two actively dissuaded the staff from being proactive on this front specifically. I take it that you don't think that's true.

    Lastly, I understand the idea of safe harbour, where someone posts a video, posts content, and you can't be liable for everything that somebody posts. However, do you agree that as soon as you recommend videos, as soon as your algorithm is putting in particular content, boosting particular content and encouraging people to see particular content, you should be liable for that content and responsible for that content?

    I would agree with you insofar as, when there's a difference between the results that are being served or the result of a passive query versus a proactive recommendation, there's a heightened level of responsibility.

    With respect to notions of liability, the challenge is that there's a binary that exists in the current conversation between whether you are a publisher or a platform. As a platform that is also, in the case of YouTube, taking in 500 hours of video every single minute and has over a billion hours of content being watched every day, being able to—

    I'm only talking about content that's recommended by you, specifically.

    Yes. In that case, we are endeavouring, through the process of the recommendation system, to basically provide content that is relevant to what we think the user wants to watch, in a corpus that is much larger than conventional publishers are actually dealing with.

    It's a long answer that isn't really saying anything. The answer is obviously yes.

    Thanks very much.

    Next up we have Ms. Kusie, for seven minutes.

    Thank you very much, Chair.

    When Bill C-76 was being drafted, did your organization have the opportunity to meet with the minister?

    No.

    Did you have the opportunity to meet with the ministerial staff?

    While it was being drafted, no, we did not.

    After the original draft was completed, with the proposed clauses, were you then approached by the minister?

    We became aware of the proposed clauses that were being introduced at clause-by-clause at the procedure and House affairs committee, actually when it was reported publicly, at which point we approached the minister's office, first to gather further information, because there wasn't much detail with respect to what those clauses contained, and then to engage robustly with the minister's office to identify some of the concerns we had and to develop proposed amendments that would resolve those concerns.

    It's fair to say, then, that you were not consulted by the minister or by ministerial staff until Bill C-76 came to the clause-by-clause procedure, until it became public.

    That's correct.

    That's pretty significant.

    When you went to the clause-by-clause process of Bill C-76, I'll going to assume that was in the House procedures committee, PROC. Is that correct?

    That's correct.

    Did you explain at the time that you would not be able to comply with this legislation as it was laid out?

    By the time Bill C-76 was amended to include the new online platforms provisions, the witness list unfortunately had already closed. As a consequence, there wasn't an opportunity to discuss that with members of the committee.

    We discussed it with individual members of the committee to express some of the concerns at the time, and then certainly raised it when it was at the Senate justice committee.

    Okay.

    You're saying, then, that you did not have the opportunity, in the House procedures committee, to explain to the committee why this legislation would not work for your organization, why you would not be able to comply. I know it was said in the preamble that you don't prefer this term, but you did express that it would be difficult for you to comply, recognizing that, as Mr. McKay said, in deciding not to advertise in elections, you are not subjecting yourself to the non-compliance.

    Did you express that at that time?

     Do you mean after the bill was introduced? Sorry, I'm....

    Yes, no problem. Were you able to express at that time that you would not be able to comply with the legislation as it was laid out?

    Yes. Once we were aware of the provisions, and also once we were able to obtain a copy of the proposed amendments, we engaged with the minister's office. We essentially reviewed what we have just reviewed with all of you, which is to say that there are certain aspects of those provisions that would be challenging due to our particular advertising systems. One of the possible outcomes would be that we would not be able to achieve the changes we need in the time allowed, and that we would therefore not be able to accommodate political advertising for the federal election.

    The minister and her staff were aware that you would not be able to comply as Bill C-76 was laid out. What was her response to you at that time, when you indicated that your organization would not be able to comply with the legislation as it was presented?

    They were hopeful that we would be able to accommodate the new changes.

    How? What did they expect, if I may ask?

    They didn't articulate anything more than that, except for, basically, a desire for us to be able to implement the new changes. Also at that time, procedurally, they expressed that because it had already proceeded to clause by clause and was going to the Senate, essentially we should raise our concerns with the Senate in the hope to obtain the amendments we were seeking.

    Did you have any conversations with the minister or her staff regarding the tools you rolled out in the U.S., which you went into some detail on here?

    Yes, we did.

    What was her response in regard to the tools that were rolled out in the U.S. comparative to the legislation that was presented in Bill C-76? What was her response to those tools, and did she mention if there was any gap between those two items?

    Well certainly, they were aware of the tools that we had introduced in the U.S. mid-terms, and essentially they were hoping that with the introduction of the changes to Bill C-76, we would introduce similar tools in Canada. It was really a matter of going through the details about why the specific provisions in Bill C-76 would make this challenging, which is where we had a back and forth. Essentially, they were looking for us to introduce tools similar to what we had there.

    In looking to introduce tools similar to what you have within the U.S., why are you not able to introduce these tools within the Canadian system, and why do they not comply with Bill C-76?

    In short, as I covered in my opening remarks, we have advertising systems that serve advertising to third party publisher sites, but we don't have a means of delivering the ad creative and the requisite information to them in real time, which would be required under Bill C-76. There are also the complications with respect to the real-time registry itself, which is why, even for our owned and operated sites like Google Search or YouTube, we also wouldn't be able to comply—at least we didn't feel comfortable we could commit to that.

    Finally, there is the additional complication with respect to the inclusion of issue advertising. To be clear, the registries that we have available in other jurisdictions actually do not include issue advertising because, as I said, it is very difficult to cogently identify. Therefore, we were concerned that we would not be able to identify issue advertising for inclusion.

    To summarize then, you were not included in the drafting stage of Bill C-76. You were not consulted by the minister or her staff as this government went forward with Bill C-76 in an effort to determine electoral reform for Canada.

    That's largely correct. I should also amend the record. We did have one discussion with the minister's office shortly after Bill C-76 was introduced, to discuss the contours of the legislation as it existed at first reading, before any of the online platform provisions were added and before any registry requirements were added. It was a robust discussion, but certainly at the time, we did not contemplate the introduction of the new provisions, so we didn't cover that. Besides that, there wasn't any engagement.

    Sure.

    What advice would you give to the next Parliament in regard to this legislation?

    As we have basically expressed before the Senate committee and elsewhere, and also to the minister's office, we are fully supportive of and aligned with the idea of increasing transparency in political advertising, and we had intended to actually bring the registry to Canada.

    There were a certain number of extremely targeted amendments. Basically, I think it was 15 words we actually changed that would have altered Bill C-76 sufficiently, such that we actually could have accommodated the requirements.

    My recommendation, if possible, would be for a future Parliament to look at those recommendations, basically look at the challenges the new provisions added—and it's worth noting, as covered by the CBC, many platforms have also similarly announced that they wouldn't be accepting political advertising because of the challenges the specific revisions are introducing—and to revisit them at that time, to see if there are tweaks that can be made to help alleviate the concerns.

     Thank you.

    Next, we'll go to Mr. Dusseault for seven minutes.

    Thank you, Mr. Chair.

    I am pleased to join this committee today. I have the privilege of returning to this committee, which I had the opportunity to chair for a few years before you, Mr. Chair. I am happy to see faces I have seen in the past.

    Mr. McKay and Mr. Kee, you know the influence you have. I am sure you are aware of the influence you have on elections and the information that circulates during elections, which influences voters and, ultimately, their decision. It is in this context that it is important to have a discussion with you about the announcements you have made recently regarding election advertising.

    My first question relates to the preamble that other colleagues have made. Other fairly large companies in the market have said that they are able to comply with the Canada Elections Act, as amended. Given the fact that you have made many investments in other countries to make such registries, for example in the United States and Europe, and also the fact that you have invested millions of dollars in certain markets, such as China, to be able to adapt your search engine to their laws—in China, we agree, the laws are very strict, which you know quite well—I have difficulty understanding why Google isn't able to adapt to Canadian legislation like that in preparation for the next federal election.

    In your opinion, what makes you unable to adapt to Canadian legislation?

    Is it because the rules don't reflect international practices or the way online advertising works? Is it because you can't afford to do it or you just don't have the will to do it?

    I apologize, but I'm going to answer in English.

     I want to underline that our decision was led by a technical evaluation about whether we could comply. It was not a specific question about the regulatory framework within Canada or our commitment to the electoral process in Canada and helping keep it both informed and transparent.

    We are on a path in other countries to implement tools like those described in the legislation, to improve those tools and to work on the back-end technical infrastructure to make those tools more informative and more useful for users as well as for all participants in an election.

    We arrived at a very difficult conversation because we were faced with a constrained time frame with amendments to legislation that are very important within the Canadian context and to us as Canadians, both as electors and participants in the electoral process. And we had to make a decision about whether or not we could comply with the legislation in the time frame allowed, and we couldn't. So our compliance was left to not accepting political advertising. It is in no way a reflection of either our corporate or our personal attitude towards Canada's authority and jurisdiction in regulating this space, and it certainly wasn't meant to be a signal about our opinion about the amendments to the Elections Act.

    It was very difficult for us, and the decision was driven by the technical challenges and the time frame we were faced with.

    I have a supplementary question about the registry you set up during the American mid-term elections and the one you plan to set up in Europe for the European Parliament.

    Are these registries, or are these rules, an initiative by Google, or are they implemented as a result of an obligation imposed by law?

    Can you clarify why you are moving forward in the United States and Europe? Are you required to do so, or do you do it on your own?

     It's actually of our own volition. Essentially, it was entirely a voluntary initiative in light of concerns that were raised as a consequence of the U.S. federal election. There were obviously a lot of robust discussions with respect to how to ensure there was enhanced transparency in the course of political advertising. As a consequence, Google, Facebook and a number of the other companies all worked very hard, to be honest, to basically start building out transparency, who gets registries and reports, that would provide our users with much more context in terms of the political advertising they were seeing. This is not just in access to copies of the actual ads themselves, but also contextual information with respect to why it was they may have been targeted, what audience this audit was looking for, how much money that particular advertiser had spent, those kinds of details.

    Once this was built for the U.S. mid-terms, there was—as Colin alluded to—a process of learning. We have global teams that build this out and are basically moving from election to election and actually learning from each individual election and improving the processes. Essentially we had a template in place that we were capable of deploying in India, that we were capable of deploying for the EU, that we expect to be deploying in other places. Also, we had individual processes over and above the registry itself. What is the process we use to verify the political advertiser? In the case of the United States, we would verify not only the identity of the advertiser by asking them to provide ID, but we would then also verify that they were authorized to run political advertising with the Federal Election Commission in the U.S.

    We have had to adapt that process as we move and implement the registry in other countries because not every election's regulator is capable of providing the kind of validation we had in the United States. So, we're adapting that and learning from those processes as well.

    This has been something we have done entirely ourselves versus being compelled by law.

     In these cases, you are happy to do so because they are your own standards and rules. However, when you have to comply with rules set by others, you are much more reluctant to do so. The decision you have made will ensure that you comply with the act, because, on balance, there will be no election advertising.

    Given the time I have left, I would like to address a complementary topic. This is the announcement you mentioned in your introduction, which concerns a channel devoted to the election campaign, intended to compensate somewhat for your decision not to authorize advertising.

    I wonder about the transparency of this platform and the algorithms used to distribute the content. What content will be broadcast? You say it's content from authoritative sources, but what does that really mean? Will the public have access to this type of information on the platform in question? Will it have access to the policies used to disseminate content and ensure that all candidates or parties have equitable coverage?

    Since it's a decision that obviously belongs to Google, who decides on this coverage? You know the influence you have and can have. Who will determine these policy issues that you will publish on this platform during the campaign?

    There are actually two separate discussion points that you raised there.

    One is that we keep on saying the word “authoritative”. What do we mean by that? That's a perfectly fair question. It's basically something that informs Google Search, Google News and YouTube as well. We actually have very robust guidelines, about 170 pages of search rater guidelines. We use external search rater pools that evaluate the results that we get to ensure that we're providing results to users in response to queries that are actually relevant to the queries they're looking for and actually from authoritative sources.

    By authoritative, we actually rank it based on authoritativeness, expertise and a third classification.... I can send you the information. I'm blanking on it.

    It's based on information we get from the search rater guides. As I said, there are pools of them, so as a result it's all done in aggregate, which helps us to surface authoritative information. These are the same signals we use, as I said, to identify this class of information and it tends to weight towards established organizations and so forth that actually do original work. To be clear, we're not evaluating the content of the work. We're just evaluating whether this is a site that's actually expert in the subject that they claim to be.

    With respect to the YouTube channel, it's a very different thing. It's a very specific, individualized channel that will be available on YouTube. We did this in the 2015 election. In that instance, we got a third party organization, Storyful, to curate that for us. It isn't done algorithmically; it is actually curated. We will do a similar thing where we will have a third party that will then curate that by pulling in information from various sources.

    We will establish with that third party what the guidelines for their inclusion will be, which we will be publishing so people are aware of how information is being included in there. It will be predominantly information on YouTube that is posted by established broadcasters and news organizations that we simply populate into the channel as a singular location for people to see.

    All this information would also be available from, for example, the CBC website, National Post and so forth. It would also be available there individually as well.

     Thank you.

    Next up, for seven minutes, is Mr. Graham.

    Thank you. It's nice to see both of you again.

    As you probably know, I'm an early adopter of Google. I've been using Google for a solid 20 years. I think it's a good service, but it's grown massively, and become a very powerful tool. With great power comes great responsibility, so I think it's very important that we have this discussion.

    The part of C-76 we're talking about is 208.1, and you want to change 15 words. What were those 15 words?

    I can happily give you a copy of the proposed amendments we provided, which might be a little more cogent.

    That would be helpful.

    Do you oppose the changes we have on C-76, or do you support the bill, from a philosophical point of view?

    Absolutely.

    Do you support the use of ad registries in general, not just for politics, but ad registries across the board?

    It's something we're looking at. As in the political context, there's obviously a certain urgency to the issues with expected transparency, and so forth. It's something we are examining, in terms of how we can provide increased transparency to our users. In fact, just this week, we announced some additional measures with respect to that.

    Colin, did you want to touch on that?

    Yes. At our developer's conference, we announced a browser extension for Chrome that allows you to see more detail about what ads you're seeing, where they came from—what networks—and how they arrived on your page. We're building it as an open-source tool, so that other ad networks that are also serving ads to sites and pages you're using can feed that information as well.

    The direction we're heading, because we recognize that this is important, is providing as much information as possible for users around why they're seeing ads.

    That leads to my next question. You talk about the technical limitations of implementing C-76, and I'm trying to get my mind around that, as a technical person. I have a fairly good sense of how your systems work, so I'm trying to see where the problem is, in the next five months, with adding the subroutines needed to grok political advertising so that you can actually use it.

    If you can help me understand the technical side of things, I will understand it. I'd like to hear what those are.

    Certainly. The principle challenge we had was this notion that the registry had to be updated in real time, simply due to the fact that we have a very wide array of advertising systems. People who run campaigns with Google will do search ads, so they show up on your search results. They can be YouTube ads—video—or TrueView. You can skip it, and also display advertising that shows up on third party publisher sites.

    In the case of the third party publisher sites, those ads are often being served by third party ad servers. We don't necessarily have immediate access to the creative ourselves. For us to update even our own registry in real time, let alone the registry of a third party that had to maintain their own registry, would be very challenging.

    What if you just have it on your own servers, and not the third party ones?

    Then, again, we could do it faster. Currently, real time would still be a challenge, based on our systems, just by virtue of the fact that we'd have to know the ad server, and then be able to update it immediately after that.

    I forget the name of the page, but there's a page on Google where you can see all the data Google has on you. It's an easy page to find, and you can see everything you've ever done that Google knows about, in real time. If you can do that for a person, why can't you do it for an ad?

    Primarily because the advertising systems are a bit different, in terms of how the creative is being uploaded to us, how it's stored and how it's managed internally. That's why I said it would be the kind of thing that we would work towards, but we didn't feel comfortable this time around that we could commit to that kind of turnaround time.

    What kind of investment of time and resources would it take to make it happen, if you decided to do so? You say it's possible for the 2023 elections, so it is possible. What would it take to do it?

    To be honest with you, it's not a question I could answer, simply because after examining it seriously, we couldn't match the time frame we had. I couldn't give you an estimate as to how long it would take, or what the resources required would be.

    Do you have a sense of how much profit Google takes out of Canada versus what's invested in Canada?

    I think I'd describe that in two separate ways. It's actually broader than that, because the way our platforms and services work, we are often an enabler for Canadian businesses and Canadian companies, whether providing free services or paid services that support their infrastructure.

    As well, with something as broad as advertising, it certainly.... With revenue that's generated by online advertising when you're dealing with Google, there's often revenue sharing with platforms and sites, so the conversation is not a one to one. Canadian businesses are using our technology to place ads, pay for their services and drive revenue themselves.

    We're very proud of the investment being undertaken in Canada, on behalf of Google, in terms of the growth of our engineering and R and D teams, and our offices themselves.

     Mr. Kee, I think you were recently at Industry's study on copyright. One of the questions that came up was on rights management systems, and whether they have followed fair dealing to apply to Canadian law. The answer from both Google and Facebook was no. They don't follow fair dealing. They apply their own policies.

    How do we get to a point where Google says, “Canada is important enough that we will make it a priority to follow local laws on matters that affect this company with great responsibility”?

    That is in no way a reflection on the relative view of the importance of Canada or local law, in terms of the application of our content identification system.

    For the benefit of the other committee members who aren't necessarily familiar with it, Content ID is our copyright management system on YouTube. The way it works is that a rights holder will provide us with a reference file—a copy of the file they want managed online—and then we apply a policy on every single copy that we detect has been uploaded to YouTube.

    Because the system is automated, it doesn't handle context or exceptions very well. Fair dealing and Canadian copyright are actually exceptions to the law, which says they are exceptions to general infringement, because of a certain basic line of reasoning, and require a contextual analysis. It's why we respond to those exceptions by having a robust appeal system. In the event that Content ID flags content inappropriately, because you feel you have a very strong argument that fair dealing applies, you then appeal that decision and assert that fair dealing applies, and then an assessment is made.

    In the case of the copyright system, you are going ahead with a system that doesn't follow Canadian law, but has an appeal system. In the case of Bill C-76, you're saying, “We're not going to do it, because it doesn't make practical sense.”

    In the copyright experience, you're not worrying about Canadian law, quite frankly, because if somebody does have a fair-dealing exemption, it shouldn't be incumbent on them to prove they have the right to do something that they absolutely have the right to do.

    I'm trying to get my head around why you're going ahead with the copyright, and not following it with Bill C-76. To me, it seems like a difficult but entirely doable system to resolve. As Nate said earlier, if it was the United States, I'm sure it would be fixed already.

    Essentially, the main difference, at the risk of getting slightly technical, is that fair dealing in Canada is an exception to infringement. It is a defence that one raises in response to a claim that you have engaged in an act of infringement, so the way that is managed is very different.

    In this instance, Bill C-76 was introducing positive obligations—not only that you had to introduce an ads registry, but about the way it had to be done—that we simply couldn't accommodate in the time frame allowed. That's the main difference in the two.

    Thank you.

    Next up, for five minutes, is Monsieur Gourde.

    Thank you, Mr. Chair.

    Thank you to the witnesses for being here this morning.

    We know that there is only a short period of time left before the elections. You mentioned the technical problems, the problems in complying with the Canada Elections Act.

    In Canada, the reality is that we can put advertising on your platforms, but Canadian legislation limits election spending limits and the diversity of places where political parties and candidates can spend their money. This determines a certain market share, which you may have evaluated.

    First, did you evaluate this market share?

    Second, it may have been such a small amount of money, considering all the projects you are running simultaneously on your platform, that it was simply not worth complying with the act this year because of the amount it could bring you.

    Has an evaluation been done internally by your managers? For example, to have an additional income of $1 million would have cost $5 million. It may not have been worth it to comply with the act for the 2019 election. Is that correct?

    That was not a calculus that entered into the determinations at all. It was fundamentally down to how we would engage the requirements, whether it was technically feasible for us, given the way our systems currently work, given the time frames and, frankly, our risk tolerance, with respect to what would happen if we ended up getting it wrong. That was entirely it.

    It never came down to a calculation of a cost benefit. If anything, it's worthwhile noting, we have opted out of engaging in the only thing elections-related that actually would earn us revenue. Instead, we are investing in things that do not earn us revenue, such as our engagement with Elections Canada on promoting election information through search and knowledge panels, and so forth, with YouTube and in various other measures, not the least of which was a $1-million grant to CJF on news literacy, in advance of the Canadian election.

    Essentially, we have doubled down on the non-revenue-earning components of it, to compensate for the fact that we simply could not accommodate the requirements of Bill C-76.

    In your answer, you talked about risk tolerance. Was it too risky, given that some of the provisions of the bill amending the Canada Elections Act were relatively vague and didn't allow you to be sure how to implement new platforms?

     It had to do with some of the wording, which was vague. Again, there were the concerns I flagged with respect to the issue of ads, but as well, it was the time frame that would be required to update the registry based on this idea, as stated in the act, that it had to be done within the same day. Does that mean literally real time or not, and what was the flexibility there?

    We have had robust engagements with both the commissioner of Canada elections and Elections Canada on matters of interpretation. You may have seen Elections Canada issue guidance about a week and a half ago with respect to that, which actually largely confirmed some of the concerns we had, and it was more of a reflection of that.

    It may happen that some ads made by third parties directly or indirectly are not necessarily considered election ads, but that they attack certain parties. You talked about human control of your platforms. Do these people have the necessary expertise in Canadian politics to distinguish between advertising that actually attacks other parties and advertising that promotes a Canadian election issue? In fact, I would like to know what the expertise of the people who will be monitoring these platforms is. Do they have expertise in Canadian politics, American politics or the politics of another country?

    It's actually a combination of all of the above.

    What's happening is that in order to implement the prohibition, we will be updating our ads policies. As I said, there are entire classes of ads where basically we will not accept the ad, such as cannabis advertising. There's another class of ad that goes through registration requirements. That's all governed by our advertising policies.

    This decision will be reflected in those ads policies. We will have ads enforcement teams located in various places around the world who will be educated on these ads enforcement policies.

    With respect to the specific issues on the use of the ad, every class of ad that you described sounds as though it would likely fall within the ambit of Bill C-76 and fall within the ambit of the prohibition.

    We will have actually teams that are trained on that, but also, specifically looking at it from a Canadian perspective, informed by the advice that we have across functional teams located here in Canada.

    Thank you.

    Monsieur Picard, you're next up, for five minutes.

    Thank you.

    Did you just mention to my colleague that you have not looked at the return on investment issue in regard to this case?

    Correct. That wasn't entering into the calculus of the decision.

    Do I understand correctly when you say that one of the technical problems you have is that with third party publicity or advertising, when they use those third party ad “brokers”, or whatever, it is difficult, and maybe impossible, to know where it comes from? Therefore, not being able to divulge the name of the author, the publicity is therefore too complicated for now, so you cannot go forward and just backtrack from the initiative of having publicity in the next election.

    It was less about the source of the advertising. It was more about our ability to update whoever it is that showed the ad in real time, the information that they showed in a political ad, and then provide them the information and the ad creative they need to update their own registries.

    Is this practice in advertising the same in other countries with those sources and third party middlemen?

    It depends on the individual practice in individual countries. I'm reticent to comment on the electoral law of other countries that I'm simply not familiar with.

    I know with respect to our own policies we've applied around political advertising in countries where we've deployed a transparency report, a registry, we've actually been employing similar requirements with respect to advertiser verification and finding means to verify that the advertiser in question was authorized to run political advertising by the local electoral regulator.

    What was the difference where you were able to apply that? Based on that experience, what is the difference in our case where it's not possible?

    It's simply because the systems we've deployed in the U.S., the EU and India would not have accommodated the specific requirements in Bill C-76. If we had moved forward with that, we would have actually implemented a similar system.

    We actually have had preliminary conversations with Elections Canada with respect to this, but in the end, it just became clear that with Bill C-76 and the specific requirement that each individual publisher had to maintain its own registry, we would have a very difficult time accommodating the requirements.

     Chair, I'll give the rest of my time to Mr. Erskine-Smith.

    I'm not technical like Mr. Graham, but I can read Bill C-76 and I have it in front of me.

    It's the publication period of the registry that caused your problem. Is that correct?

    That's correct.

    Right.

    They don't say “real time”. You've said “real time” a number of times, but it's not what the act says.

    The act says, “during the period that begins on the day on which the online platform first publishes the advertising message”. That's the part that you're referring to.

    That's correct, because basically if the ad was displayed at 11:59 p.m., the registry would have to be updated by midnight.

    That's one interpretation. Another interpretation might be a 24-hour period of time, which might be a more reasonable interpretation. You have high-priced lawyers and presumably they're right. I'm just thinking off the top of my head and haven't taken the detailed analysis that you have, of course, but certainly one solution would be that you just defer the publication of the ad.

    Why couldn't you defer the publication in order to have a 24-hour or 48-hour waiting period until you first publish it? Then it would be quite easy to do, wouldn't it?

    Again, it's simply because technically it wasn't feasible for us to do that.

    Why not?

    Our engineers told us it wouldn't be feasible for us to do that.

    Okay, so when I go to post an ad on Facebook, they don't publish it right away. There's a holding period and they assess whether it's something that ought to be published.

    Your engineers couldn't figure out a deferral process of 24 hours, 48 hours, or even seven days, and you put it on us to say we have to schedule our advertising for the election knowing we've made it hard on you, with tight timelines. There is no way you could have figured this out with all the money that you got.

    It would have been a challenge for us to re-architect within six months the entire underlying systems that we have for online advertising.

    I have to say, I find the answers incredulous, as incredulous as you suggesting that $1 million is such a wonderful thing when you made $8 billion in Q4 last year. Not taking this as seriously as you ought to have is a detriment to our democracy, and you should have done better.

    Thanks.

     Monsieur Gourde, you're next up, for another five minutes.

    I'm going to come back to a practical question.

    Elections Canada will pay particular attention to all digital platforms during this election. If a problem arises due to false advertising, fake news or the like, do you have a memorandum of understanding to work directly with Elections Canada as quickly as possible to remedy the situation?

    We don't have something as formal as an MOU. We've had extremely robust engagement with both Elections Canada and the commissioner of Canada elections. The commissioner's office is actually doing the enforcement of act; Elections Canada actually administers the elections.

    Especially with Elections Canada, we're working with them to source data on candidates and information that we can actually see surface in Google Search, for example, when someone is searching for candidate information.

    Certainly with the commissioner of Canada elections, we're working on enforcement-related actions. With respect to their principal concern about advertising, because we're not taking advertising, it's less of a concern, but there are also additional measures that were introduced in the act around impersonation, for example, and we're working with them on those issues. We will work with them very closely.

    I'm a little worried about that, because today, everything moves so fast in the digital age. If there is an irregularity during my election campaign, I will file a complaint with Elections Canada, and an Elections Canada representative may check with you that same day. You tell me that you have no memorandum of understanding, that you have never negotiated in order to put in place ways of working and that you have not designated anyone. An Elections Canada representative will call you, but who will they contact?

    The complaint will be transferred from one person to another, and someone will eventually answer it? Has someone in your company already been designated for the election period to speak to an Elections Canada representative and respond to complaints?

    Yes. Essentially, there will be a team of people, depending on the specific issue, to respond to specific issues that get raised. We call them “escalation paths”.

    Essentially, when we have established regulators such as the commissioner's office or Elections Canada, they will actually have means to get immediate responses on urgent issues simply because they will escalate issues that are serious ones, and then we would actually bypass or accelerate through the normal reporting processes.

    It's also worth noting, in the event that you see anything on any of our systems, that we have a wide variety of reporting mechanisms for you to report it directly to us, so we'd encourage you to do that as well.

    Is this team already in place? Does it already exist, or will it be set up over the summer?

    The team is actually in the process of being set up.

    Thank you.

    That's it for me.

    Next up for five minutes is Mr. Baylis.

    Thank you, gentlemen.

    I want to follow up a little bit on what my colleague Nate was saying. We are clearly struggling to believe you about the aspect of you being able to meet the requirements of Bill C-76. Facebook says they can meet them. Are you aware of that?

    We are aware.

    You're aware of that.

    What skills do their engineers have that your engineers don't have?

    It's not a reflection of skill. It's a reflection of their advertising systems work very differently from ours and they could accommodate the requirements in a way that we simply could not.

    What if when you want to go post an advertisement, you go through that page, you post up your advertisement, and you just add a little box that says this is a political ad? Then you have your programmers program exactly like Mr. Erskine-Smith said, so it will delay that going up for 24 hours.

    Why don't they do that? It seems like a pretty simple fix.

    Simply because the way our advertisement systems work, the fix was not nearly as simple as it might seem on the surface. Again, these are extremely complex systems so that every single time you implement one change it actually has a cascade effect. After robust discussion over many months about the ability to implement these kinds of systems, basically it simply became clear we couldn't do it in time.

    It became clear, so you had a long discussion. Clearly when you had that discussion you came out with a timeline that was reasonable, right?

    We were working towards the June 30 timeline, which is when the requirements—

    And I got that. You said you couldn't meet that.

    When you had this robust discussion you clearly said we looked at it all, we can't get it done in this time. You had a schedule that had to be done. What was the schedule date?

    June 30 was the only scheduled date we were looking at because that's when the legal obligations came into effect.

    Let's say I've got to do a project and I've got to get it done by June 30. I say how do I get it done, I see the steps that need to be done, and I come up with a date, and I say can I meet that date or not?

    I don't just say as an engineer we can't do June 30. They had to do some form of calculations and some form of projections. That's what I'm asking you. They clearly did that to say they can't meet this, so when they did these projections, what date did they come up with?

    There wasn't a date because June 30 was the only date. It was the date when legal obligations came into force and if the system was not built and in place by then, we would not be in compliance with the law. Therefore, if it was—

    How did they determine they couldn't meet that date?

    Simply by virtue of having looked at the work that would be required—

    Okay, so they looked at the work that would be required and then they said it's a lot of work. I got that. They had to say how long that work's going to take if we start today. How long did they say that work would take?

    The didn't give us that projection.

    I want to understand something. We said we need it done by June 30 and then they went and did some kind of projection that said they couldn't do it by June 30, but the projection didn't say when they could do it, it just said it couldn't be done. But they have to, then, say I need six months, seven months, 10 months, 10 years, and then say if I start today, I can make it. What was their projection date to get it done by?

    I can't give you that information because it was a “can you do it by this?”

     Even if they say can you do it by this...? Let's say I want you to build me a house and I want you to build it in 10 days. They say they did the calculations, they could only build it in 30 days. Okay fine, so if you start today, you've got 30 days.

    You say I want to program something. It's really complicated.

     I understand that, Mr. Kee, it's very complicated. So I do a calculation, I say I need seven months, and it says that you've got to get it done in six. It can't get done. I get that. But I want the seven months number, or the 10 days number. I want the number when they did the projections of when it can be done by. It's a simple question. They can't just say it can't be done, because then they didn't do the work.

    Do you follow me or not?

    Did they do a schedule to say when it could be done? They have to have done this to say it could not be done by this schedule.

    I don't have the information. I can inquire.

    I'm not asking you to inquire only. I'm asking you to come back here and give us a specific date on the calculations that they did to say they could meet this date. Am I clear?

    You're clear.

    Do you understand what I'm asking for?

    I understand what you're asking. With respect to our ability to disclose it, because given the fact that it's an internal confidential engineering information—

    You cannot disclose—

    I'm saying that I was not made aware of any kind of projected dates.

    I understand that you per se weren't made aware of it, but it exists. Are you going to disclose it or not? Is this some kind of secret that it takes...? Is this is part of the Google secrets?

    Simply put, I cannot commit to the disclosure without having internal conversations first.

    I would like to see the exact calculations that were done when they predicted they couldn't meet this date. Does that make sense or not? Is this some kind of bad question I'm asking? I'm just curious.

    Given the fact that as of June 30, legal obligations came into force and that [Inaudible-Editor]

    They couldn't meet them, I got that.

     The question was a binary one, simply “Can you do it by this date?” The answer was no. That was where we were.

    Okay, so how did they get to the “no”? They were asked, “Can you do it in six months?” They said, “No, we can't.”

    How did they get there? Did they just say, “No, can't do it”, or did they make some type of calculation?

    It's a simple question. I'm asking you, “Can you get it done in this time frame?” “No.” Did you just say no off the top of your head, or did you do some type of work to say when you can't get it done by?

    Facebook said, “Well, our engineers are maybe a little smarter,” or “Our systems are clearly not as complex as their wonderful systems,” or “We have more money than Google to do it.” I don't know what they did, but they did some calculations and said, “Yeah, we can do it.”

    You guys did some calculations and said no, or did you just say off the top of your head, “No, we can't do it”?

    Did you make it up or did you at least do some type of work? That's what I'm asking; and if you did the work, I'd like to see it.

    As I said, we examined the requirements and it became clear that we simply couldn't comply within the time frame. To be clear, it has also been clear that Microsoft can't do it; Yahoo! is still undetermined, but likely also; and many others are not going to be able to complete the requirement.

     Facebook can do it; and Microsoft, you said....

    Yahoo! is doing some type of calculations to determine if they can do it.

    You did some calculations, or did you make it up?

    It's a simple question: Did you calculate it, or did you just decide you can't do it.

    We were advised by our engineering teams that we would not be able to meet the requirement. That is—

    Did your engineering team just say it off the top of their heads, or did they do some type of work?

    Why are you hiding from this? If you did the work, just say, “Look, Frank, we looked at it; it's going to take us 2.2 years and four months.”

    Why are you so upset about it? Just tell me. Why are you hiding from it?

    It's not a matter of being upset. It's just more the fact that there was a hard deadline and the question was a binary yes or no. The answer—

    Okay, but how did you get the answer to that question?

    That is what I'm saying we'd have to look into.

    Did they just make it off the top of their heads, “We can't do it”, or did they do some calculations? You said Yahoo! is doing some calculations.

    I've already told you that we would have to inquire.

    Are you going to come back with the dates—

    I have to inquire.

    —and the calculations?

     I have to inquire. I can't commit to coming back with calculations.

    If they've done—

    Thank you, Mr. Baylis. We are going to have more time, so if you want some more time, you can ask for that.

    Next we'll have Mr. Dusseault.

    Folks, we do have quite a bit of time. We have 55 minutes. We have these witnesses for the rest of the time. Anyway, just let me know. Typically, we try to end by five on Thursdays, but we'll see where it goes.

    Mr. Dusseault, you have three minutes.

    Thank you, Mr. Chair.

    This time I will focus more on YouTube, a very popular and influential platform, just like Google, of which it is a part. I was talking about it earlier. YouTube sometimes directs users to extreme, unreliable content that reports or occasionally praises conspiracy theories. YouTube makes this content look like real information.

    I was wondering if you had any details about the algorithm used for users, once they are on a web page displaying, say, political content, since that is the subject of our discussion today. The algorithm will give them suggestions for other videos on the right of the page they are viewing, or under the video if they are using a mobile phone. What algorithm is used and what is the degree of transparency of this algorithm that suggests content to users when they are on a particular web page?

    What mechanism is there to ensure that this content does not praise conspiracy theories or give fake news, unreliable information or, perhaps, unbalanced information, in other words information that may just promote an idea or vision, a political party?

    What degree of transparency and what mechanism have you put in place to ensure that the content that is suggested to users is quality content, that it is balanced in terms of public policy, political parties and political ideas as well?

     There are a number of factors that go into the recommendation system. It's worthwhile noting that the weighting that occurs around news and information content is different from, say, entertainment content. Initially, the recommendations were actually built more for entertainment content such as music, etc. It actually works extremely well for that.

    When it was applied to news and information, it became more apparent that there were some challenges, which is actually why we changed the weighting system. What that means, again, is looking at the factor once we evaluate the video and then at what's the authoritativeness: overweighting for authoritativeness and under-weighting for information that isn't necessarily going to be authoritative or trustworthy.

    That is very contextual. It depends on the specifics of whether you're signed in or not. The information is available on your watch time. It's based on information about the video you're watching, on the kind of video that other people have liked to watch and on what are the other videos that people who have liked this video like to watch, etc. This is why it actually is dynamic and will constantly evolve and change.

    In addition to making tweaks and changes to that system over time to ensure that we're actually providing more authoritative information in the case of news and information, we're also adding additional contextual pieces, whereby we'll actually have clear flags, labels and contextual boxes to indicate when there is subject matter or individuals that are frequently subject to misinformation. For example, there's the conspiracy theory issue that you raised.

     Essentially, if you see a video that may be suggesting vaccine hesitation or so on and so forth, you'll get information saying that this is not actually confirmed by science and that gives more contextual information about what that video is covering. The same thing applies to things like 9/11 conspiracy theories and so on and so forth. This will be an ongoing process.

     Mostly, we want to make sure that even if a user is seeing information, they're actually given context so they can properly evaluate it themselves.

    I would like to use the seconds I have left to discuss the transparency of this algorithm. Earlier, I think Mr. McKay talked about this issue of transparency with respect to advertising, which is why you are shown certain ads. There even seems to be a new feature in the Chrome browser that allows users to see why such advertising has been offered to them.

    Is it possible to have the same functionality for content recommended to YouTube users? This would give them a better understanding of why certain content is suggested to them rather than another.

    Certainly, finding means by which we can actually increase transparency and users can understand the context in which they're being served information is something that we're constantly working on.

    We actually produced a report—I think it was 25 pages—on how Google fights this information. That includes an entire dedicated section on YouTube that explains much of what I described to you, as well as, again, the general factors that go in. It's something that we'll strive to work towards, like we're doing on ads on the YouTube platform as well.

    Just to add a supplemental to what Mr. Kee just said, on your Google account writ more largely, you can go into “myaccount” and it will identify what we've identified as your interests across all of our products and services. You can go to myaccount and it will say in general terms that you like 1980s music and racing videos. It will give you that general observation, which you can then correct. You can delete that information or you can add in additional interests so that across our services we have a better understanding of what you're interested in and, as well, what you don't want us to serve.

    On the page itself, there is a little three-dot bar beside the videos, the specially recommended videos—on mobile, as you mentioned—where you can signal that you're not interested in that content or that you would like more of it. There is granular control to not seeing that in your video feed, your newsfeed or across Google services.

    Thank you.

    We will go into the next round, which will be our final one. I think we have four more questions to be asked, or four more time periods. On the list, we have Mr. Graham, Mr. Erskine-Smith, Mr. Nater and Mr. Baylis, each for five minutes.

     Is that okay?

    Go ahead, Mr. Graham.

    Thanks.

    I want to build on other questions on timelines. At the very beginning, in the first round of questions, we were trying to get a sense of whether you would be ready for the next election in 2023. That seemed to be a difficult question to answer. I never got an answer that said “yes, Google will be ready to implement Bill C-76, by the 2023 federal election”, assuming it happens at that time.

    If we know that it's going to be ready for 2023 and we know it's not going to be ready for June 30, 2019, do you know? Are you actively working on it now? Do you know if it's going to be ready at some point between those two dates?

    Simply put, it would be clear to say that we are going to strive to have it ready by 2023. I couldn't commit to you specifically about when it may or may not be ready.

    The other thing that is worthwhile noting is that as a global company we work on global elections, so the teams that are working on this are deploying the transparency report from place to place to place. It will continue to evolve and grow in terms of that. As well, our own advertising systems will continue to evolve and grow.

    That's why providing a hard date—that it will take two years or however—is very difficult. Between now and then, there will be a number of changes that have already been introduced in the system that would actually impact that.

     Understood.

    When GDPR came in fairly recently, how long a lead time did you have on that and how long did it take for you to put it in place?

    I believe the conversations around GDPR took upwards of four years to deliberate on the legislation itself and then its implementation. It's still going through implementation. The focus we had on GDPR from the outset was both on participating in the discussion about the content, the tone and the objectives of the legislation, working closely with the European Commission and their staff, and then on also ensuring we had the systems in place to be able to comply with it. That's still an ongoing process.

    If you're drawing an analogy, there's an extreme distinction between the way the amendments to Bill C-76 were considered and implemented and the way legislation normally is considered and implemented.

    I appreciate your point.

    Mr. Kee, I talked in the last round of questions about the 15 word changes. Between rounds, I've been looking through my notes, because I do sit on PROC and I was involved with the Bill C-76 process from beginning to end. We had numerous witnesses and numerous submissions, but I cannot find any from Google. On those 15 words, how would—

    That would be because the changes were introduced during clause-by-clause after the witness list had closed, so we didn't make representations to the committee because the provisions in question were not being considered at the time.

    As I said, I'm more than happy to circulate them now. They were provided to the Senate committee when it was at that stage.

    Okay. Thank you. That's good to know.

    That's all I have for the moment. I appreciate your being here. It's been an interesting meeting, so thank you very much for this.

    Thank you.

    Go ahead, Mr. Baylis.

    You're not going to accept political advertisements. That was the decision that was made, right?

    Correct.

    Okay.

    How are you going to stop me from advertising? You say that you don't want to accept my ad, but I'm, like, nefarious. I'm not a good actor, so I'm going to try to put an ad up anyway. How are you going to stop me?

    As I mentioned in my opening remarks, there will be a combination of automated systems that will be evaluating advertising that comes in, as well as, basically, ad enforcement teams that will be also reviewing all the ads.

    Okay. To stop me, you're going to have this system that you're going to put in place to identify an ad, right?

    Mr. Jason Kee: Correct.

    Mr. Frank Baylis: Then you're also going to have a separate team to identify ads as well—one automated, one not automated—and they're going to identify these ads, right?

    Yes. Often what happens is that the automated systems will review the ads at the beginning and look for flags or tags that indicate that it's probably a political ad. Then, in some instances, that will go to a human team for review, because it may require a contextual analysis that the machines simply can't provide.

    Right. You are not able within this time frame to make the registry, but you're able to put in the programming, the people and the resources necessary to stop it, right?

    Correct.

    Okay. I come along and I have this ad. Either you can take this ad and put it on the registry or you can just stop it, but you clearly can identify it, right? We've agreed on that. You've just said that you can identify it.

    We will be using our systems to identify and enforce our ad policies, yes.

    You said to me that you agree that you can identify that this is the ad.

    Correct.

    Okay.

    Once you can identify that it's the ad, instead of saying “I have all my technology and people to block it”—you've got that—why can't you just say, “Okay, I've identified it, and I'm just going to put it on the registry”?

    Simply because, again, the real-time time frame in order to do that would actually be tricky for us to comply with, to update a registry, and, as I said, it also would be delivering that information to third parties.

    You can't do it in real time.

     Let's say I come along. You have your automated system. Here's the ad. It's no good, but you don't know that because you can't do it in real time, right? It's up on your page or your Google platform for however time it takes you.... You can't do it in 24 hours, so in how long can you do it...? Two days? Give me a number.

    Well, like I said, it would get tagged and then reviewed—

    I know, but how long would it take for you to identify it? You couldn't make the 24 hours, right? I got that. In how long could you have done it...? A week? A day? Two days?

    Again, I couldn't provide you with specific time frames.

    Let's say 48 hours. Can we say that just for argument's sake?

    I come along. Here's my ad—boom. I have a system to identify it. It takes 48 hours, though, right? It's up on your system for 48 hours before you go, “Oh my gosh, this ad has taken us 48 hours to identify and now we have to take it down.” Is that what's going to happen?

    As I said, we will have both the automated systems and our ads enforcement team that will identify it and—

    But will they be able to identify it in real time?

    Mr. Jason Kee: —that will basically be applying—

    Mr. Frank Baylis: In real time?

    As fast as we can manage.

    Okay, but you have to do it within 24 hours, right?

    There's a difference between being able to identify and remove the ad versus being able to update an ad registry.

    Okay. I want to make sure I understand. There's a difference between being able to identify it and remove it within 24 hours, no problem.... I have an automated system to identify it and remove it, and I have automated system people. Say I've identified it and removed it—whew, that was close. Identifying it and removing it in 24 hours, that I can do, but identifying it and putting it on a registry, like a big database—it's not even that big, I'd imagine, by your standards—that I can't do. Is that what I understand?

     Due to other aspects of the complexities with respect to the registry requirements, that is why we couldn't actually deploy. As I said, we do this in other jurisdictions, right?

    Yes, I'm sure you do.

    Let's be clear on what you're saying. You can identify it and block it within 24 hours, using a combination of software and people.

    You can identify and block it within 24 hours. Is that correct?

    As I said, we will be enforcing very vigorously and with respect to detecting and removal.

    You said that you have the systems. Your engineers did that.

     Say I've identified it and blocked it within 24 hours. Once you've identified it, whatever programming...because I have to do some programming to say, “Poof, take it down within 24 hours.” I imagine you can do it instantaneously or is it going to be up for 24 hours? I don't even know. How long is it going to be up for? That's my question to you.

    As we do with all classes of advertising, we'll basically be trying to remove it immediately upon detection.

    Oh, immediately. What's that? Instantaneously?

    Immediately upon detection.

    Okay. So you can catch it and immediately bring it down, but you can't catch it and within 24 hours you have to program it and move it all the way into this database? Now, that's asking too much, frankly; your programmers can't do that, but they can sure as heck program it, find it, get a backup team and get it down instantaneously. That ad is not getting up there. Say I know the ad, and I found it instantaneously, but you want me to program a database? I mean, we're just Google....

     How many billions of hours do you get watched a day, did you say? Was it 500? What was your number?

    On YouTube, there are a billion hours of content watched every day.

    A day? So you have a pretty big database. Can I assume that?

    There are a multitude of databases, but again—

    For a little database for a bunch of legal advertisements to put on...I can't imagine that database is one-tenth of one-tenth of one-tenth of your hardware/software net, but your engineers can't program...? They can do everything. They can catch it and identify it instantaneously, but they just can't put it in a database because that's too complicated? Isn't one of the expertises of Google database management? Am I off here?

    As I said previously, the specific requirements were simply something that we couldn't accommodate.

    Can you put—

    Mr. Baylis, that finishes the combined time for you and Mr. Erskine-Smith.

    We have somebody else, and then we're going to come back to you.

    All right. That's fair.

     It looks as if that's what you're asking for.

    We'll go next to Mr. Nater for five minutes.

    Thank you, Mr. Chair.

    It's good to be here on this committee. I usually sit on the procedure and House affairs committee, so it's nice to be here for a bit of a change of scenery.

    When it comes to YouTube, you have policies in place in terms of what can and cannot be shown on YouTube. Often, though, there are situations where some of the most viral videos, or those that approach that line and may not quite step over the line.... How do you determine how close to that line you get? What type of contextual analysis is there? What types of safeguards are in place to establish where that line is and when someone may get right up to that line without necessarily stepping over it? What types of procedures do you have in place?

    Essentially, any video that gets identified as such, either through reporting or through our automated systems, which also, again, review the videos to detect for compliance, will go to manual review for a human analysis. We'll also basically examine it, saying, okay, this doesn't actually cross the line into actual hate speech that's inciting violence, but it's derogatory in some other way, and they will basically classify it in its borderline context.

     I want to go back to Bill C-76 for a minute.

    When we talk about technology, innovation, the digital economy, the first people I think of are not public servants at the Privy Council Office; they are not public servants at Elections Canada. When I think of the digital economy and things like that, I think of companies like Google, Facebook, Twitter—those that are innovating.

    In your testimony, you mentioned that you weren't consulted by the drafters of the legislation. I find that troubling. Government is probably one of the worst offenders for keeping up with technology, especially when drafting legislation, when they're not consulting with those within the industry.

    You mentioned that the decision not to advertise was largely a technical decision, based on the requirements of obtaining that by June 30, July 1. I can accept that, and I want to put on the record the reason why I accept that.

     Elections Canada themselves said that the provisions in Bill C-76 ought to have been in law with royal assent by April 30, 2018. On April 30, 2018, the legislation was only just tabled in the House of Commons. It did not receive royal assent until December 2018.

    If Google, YouTube, your private businesses decide tomorrow that you no longer want to stream cute cat videos, there is nothing that the Government of Canada can do force you to do so. I would assume it would be the same with any type of advertising. If you decide not to advertise for any reason outside of human rights violations, there is nothing requiring you to do that.

    What I find fascinating, though—and it is more of a rant than a question—is that the Government of Canada, in their rush to implement this legislation at the very last minute of the time period they're able to do it within, never consulted with those who would be implementing a large portion of this legislation.

    The changes were done in clause-by-clause. There were 200-plus amendments in clause-by-clause. I was part of those discussions. I missed a few of them for the birth of a child, but I was there for most of the discussions. Then, they were table-dropped at the very last minute, after the witnesses had the opportunity to discuss....

    It's not a question, but you're welcome to comment on that. I am just incredulous that the government would rush this legislation—the very last possible period of time to have it implemented before the election—and then expect every private business to comply with the rules for which they have had no opportunity to, (a) be consulted or (b) make suggestions during the period that clause-by-clause happened.

     I'd be happy if you have any comments on that.

    I can reply with an observation. We're having a conversation here where the discussion is effectively that one company has discussed how they may possibly implement this, based on a promise to do it by June 30, and how we have provided information on why we cannot introduce tools that reflect the obligations in the Elections Act as amended.

    The reality, as you described it, is that many companies have many different products and services across this industry, whether it's with advertising or delivering news, and there are many different ways that they're interpreting the obligations and their capacity to meet the obligations. So far, we've had only one company that has said they in fact will be implementing a tool as described in the legislation.

    I am not saying that as a defence; I am saying that as a reflection of the complexity issues. As you observed, it needs an intense amount of study and discussion to be able to roll out tools that are effective, and that have longevity and provide the information in a predictable and reliable way to users.

    In no way do we want to be in a position where we're not doing that. Our business is to provide information and answers to questions from our users.

    However, we are in an uncomfortable situation where we are complying with the legislation by not taking political advertising. The reality, as you said, is that with further consultation and a discussion of the amendments that Mr. Kee tabled, as well as broader consideration of the industry, from both our point of view as well as Canadian businesses and Canadian sites, there may have been a different conversation around the appropriate tool for Canadians that would deliver information to them during this electoral cycle.

    Next up, we're going to go with a combination. We're going to go with Mr. Erskine-Smith first and then Mr. Baylis to finish. We have about four minutes left.

    Go ahead, Mr. Erskine-Smith.

    Thanks very much.

    First I want to ask about the review team. Obviously a lot depends upon the honesty of users, in a way. You had a certain experience in Washington state that was reported where political ads were placed in some instances.

    What is the size of the team in Canada that will be reviewing this? How many staff—?

     It will be a globally distributed team that will basically be ramped up as required to respond and enforce the ads policy.

    Is there a number of staff that you've targeted?

    Again, I don't know the specific number of staff: enough to comply with the policy.

    Okay.

    I ask about numbers, because with respect to trust and safety.... Again, this is from an article in Bloomberg. I understand that there were a modest number of staff—only 22, I think—who were to review content for trust and safety. That number is higher now.

     What is the exact number of people who are now reviewing content for trust and safety?

    Over 10,000.

    So you've gone from 22 to over 10,000. That's great.

    Maybe you could enlighten me.... In 2016, one of your employees proposed that content right up to that line ought not to be recommended. YouTube did not take that advice seriously and rejected that advice.

    You said now that as of January of this year, they have accepted the advice. What changed?

    Again, I wouldn't characterize it that way.

     I think it's part of an ongoing development and evolution of the policies, and basically how we approach borderline, challenging, controversial, unauthoritative, low-quality content on the platform. There is currently an evolution in our approach with respect to this, and basically a deployment of additional technologies, especially through algorithms, etc., to try to make sure we're serving up authoritative and quality content to our users.

    So it's not perhaps because no one was paying attention previously and you were making lots of money, and now people are paying attention and there's negative press and you've decided to change your practices as a result.

    I wouldn't agree with the interpretation.

    Okay.

    I'm going to take a different tack now.

    Google, my colleague said that you made $8 billion in last quarter. I looked it up. It's actually $8.5 billion in U.S. dollars, so congratulations. That's really good in Canadian dollars. You make it by selling advertisements, right? I like a certain music video or I like to read a certain journalist, so you show me that information and then you throw an ad up there and you make a lot of money on it. You make a tremendous amount of money.

    That has worked very well for you. In fact, what has happened is that the journalist, the musician, photographer, writer, actor, the movie producer—all those people—make nothing. But that's okay. Even though that copyright is taken from them and they make nothing and you make all the money, that's okay. Why is it okay? As you well said, Mr. Kee, you're a platform not a publisher, and as long as you remain a platform, you have something called “safe harbour” which protects you.

    What happens is that all of our artists, anybody who has anything copyrighted, a photojournalist.... It used to happen that they'd get an amazing picture, and they'd sell that picture and make a lot of money. Now you take that picture for free, but you didn't do anything. You show me the picture; you throw up an advertisement, and you make all the money. That's where all this content is coming from, and you don't pay for it and you don't want to.

    The danger you have—why you don't want to do this—is that the minute you start controlling these ads, you move from being a platform to proof positive that you're a publisher. Once you're a publisher, you're subject to copyright and all that.

    Is that not the real reason....? With this technical mumbo-jumbo you just fed me about how you can catch it instantaneously, stop it, but you can't get it on a database, is that not really that you're trying to protect this business model that allows you to make $8.5 billion U.S. in a quarter while all of these copyrighted people can't make a nickel? They've have been screaming to high hell that they can't make a nickel, and you're taking all that money. You just don't want to be a publisher, because once you're a publisher, you're no longer covered by safe harbour.

     Isn't that the real reason?

    Not remotely.

    That's what I thought—

    There are several things there.

    Number one, if that were the case, it's puzzling as to why we would take a decision that is costing us money, insofar as we're no longer earning revenue from a class of ads. More importantly, vis-à-vis our publisher partners, vis-à-vis YouTube creators, we operate under a partnership model where we actually share revenue. In the case of websites, for example, that use Google's infrastructure—which is what has generated our challenges for complying with Bill C-76, because they show Google ads against their content—they earn more than 70% of the revenue for every single ad that shows because they're the ones providing the content.

    Mr. Kee, you're paying a few dollars to protect your business model. You're not paying it...and don't ever try to pass that off as being good citizens.

    We paid over $13 billion out to websites last year alone.

     You're paying it because you do not want to become a publisher. Are you ready to say here and now that you're ready to be a publisher, and call yourself a publisher, and be subject—

    We're not a publisher.

    I know you're not a publisher. I know you're very well protecting yourself that way, because the minute you become a publisher, all that money that you're stealing from everybody else you have to start paying for.

    As I said, because we operate under a partnership model, where we're actually paying out to creators and we're paying out to these publishers—

    And you have negotiated.... You are paying out, and which artists did you negotiate how much for their—

    We have thousands of music licence agreements in place with all the major collectives, record labels, etc., where we paid out over $6 billion to the music industry last year.

    You paid $6 billion, and you only made $8.5 billion a quarter.

    That was through YouTube alone. That's not actually talking about our other services.

    That's it; you made even more money, I know that.

    That's in addition to the over $13 billion we paid out to publishers who are showing our ads. As we said, our fundamental model is a partnership model.

     Are you telling me here and now that all these people have copyright, are happy...?

    Mr. Baylis—

    This is my last question.

    Are you telling me you're such a good corporate citizen that if I talked to any of the musicians, the journalists, the photographers, the writers, the actors or the movie producers, they would say, “We're very satisfied with what Google and YouTube are paying us”? That's a yes-or-no question.

    We will continue to engage with them. We treat them as partners, and we'll do that.

    Can you just say yes or no? Will they say, yes, they are happy, or no?

    We will continue to engage and we have a partnership with them.

    It's a simple question. Are they going to be happy?

    We have a partnership model with them.

    Thank you.

    That brings us to the end of our questions.

    Mr. Erskine-Smith, have you one last comment, or are we good? Okay.

    Mr. Graham.

    I want to make a comment on a separate topic, on something that Google did say, because it's not all bad.

    I want to congratulate Google, because today you announced that all Chromebooks will be Linux-compatible out of the box starting soon. So not everything is negative. I'm very happy with some of the things you are doing. Thank you for that.

    Once again, Mr. McKay and Mr. Kee, thank you for coming. We look forward to your appearance and Google's appearance at the IGC. I hope the requests are taken seriously, as we've requested your CEO; we hope you're still considering that.

    Thanks for coming today.

    The meeting is adjourned.