Evidence of meeting #37 for Status of Women in the 42nd Parliament, 1st Session. (The original version is on Parliament’s site, as are the minutes.) The winning word was content.
A recording is available from Parliament.
3:30 p.m.
Conservative
The Chair Conservative Marilyn Gladu
I'll call the meeting to order.
I'm very excited to see Ruby Sahota back with us today.
Welcome.
Bryan, welcome to the committee.
This is going to be a good topic.
We have with us today, from MediaSmarts, Matthew Johnson, who is the Director of Education, and Jane Bailey, who is a professor with the Faculty of Law at the University of Ottawa. I'm going to invite them to open with their comments, beginning with Jane.
3:30 p.m.
Jane Bailey Professor, Faculty of Law, University of Ottawa, As an Individual
Thank you very much for inviting me back.
I understand that today one of the things we've been asked to focus on is this notion of algorithmic curation. I'm making these remarks as the co-leader of the eQuality Project, which is a project that in fact is focused on the big data environment and its impacts on online conflict between young people. I'm also a member of the steering committee of the National Association of Women and the Law.
Big data, or the big data environment, where each of us trade our data online for the services we get, is a mechanism for sorting all of us, including young people, into categories in an attempt to predict what we will do based on what we've done in the past and also to influence our behaviour in the future, especially around marketing, to encourage us to purchase certain goods or to consume in certain ways.
In terms of our concerns at the eQuality Project with the big data model, and with algorithmic sorting in particular, there are three that I want to touch on.
The first is this assumption that the past predicts the future. This can become a self-fulfilling prophecy, which in the context of youth is particularly concerning. The assumption is not only that what we do predicts what we will do individually in the future, but that what people who are assumed to be like us will do or have done in the past somehow predicts what we as individuals will do in the future.
We can begin with an example that will appear soon in the eQuality Project annual report, courtesy of my co-leader Valerie Steeves. Think about online advertising and targeting. If you are a racialized male online and the algorithmic sort sorts racialized males as people who are more likely to commit crimes, then the advertising targeted to those people in that category—the young racialized male—might lean more toward names of criminal lawyers and ads for searching out people's criminal records, as opposed to advertising for law schools, which might be the kind of advertising that a middle-class white young person might get. There's a study by Latanya Sweeney about this.
The shaping of our online experience, that information to which we have access, according to our algorithmic sorting into groups, then can become a bit of a self-fulfilling prophecy because it's assumed that there's certain information that's relevant to us, and that's the information that we have access to. I don't know if you have ever sat side by side with someone and done a Google search and have seen that you get different results. That's one thing. The assumption that the past predicts the future is problematic in a very conservative way. It's problematic when the groups that we're using are based on discriminatory categories as well.
The second problem obviously is the constraint that this imposes on change, the constraint that it imposes on people's equal capacity to participate and to grow. In the context of young people, our concern is around whether young people will be influenced in ways such that they internalize the stereotypes that are wallpapering their online spaces, how internalization of that stereotype may affect their self-presentation, their self-understanding, and their understanding of their possibilities for future growth and participation, and in what ways this may set youth up for conflict with one another and set youth up to judge each other according to the stereotype's marketed standards that are part of the algorithmic sort in an online environment.
The third problem that we're particularly concerned with is the lack of transparency, of course, around this algorithmic sort. We cannot question it. Most people, even people who are computer programmers, don't necessarily understand the outcomes of the algorithmic sort. When important decisions are getting made about people's lives, such as what information they have access to and what categories they're being sorted into, and we have a system that we are not allowed to question, that isn't required to be transparent, and that isn't required to provide us with an explanation of why it is we've been sorted in this particular way, there are obviously serious democratic issues.
Again, our concern in the eQuality Project is to focus on the impact that this has on young people, particularly young people from vulnerable communities, which includes girls.
What to do about this?
One of the important points, which came from earlier work that I did with Professor Steeves in the eGirls Project, is that more surveillance is not the solution. The big data algorithmic environment is a surveillance environment. It's a corporate surveillance environment and, of course, the corporate collection of this data spills over into the public environment, because it creates opportunities for public law enforcement access to this data.
What the girls in the eGirls Project told us about their experiences in the online environment was that surveillance was a problem and not a solution. Algorithmic sort solutions that purport to categorize young people according to surveillance of their data instill greater distrust for young people and adults, and greater distrust of young people in the systems they're using.
I think it's really important to think about refocusing and reshaping our concerns on corporate practices here, rather than on training children to accept an algorithmic model, to accept that they're going to be sorted in this particular way. We should take a step back and ask corporations to better explain their practices—the how, the why, the when—and to consider regulation if necessary, including to require that explanations be provided where decisions are being made about a young people's life chances according to algorithmic curation and sorting.
Those are my remarks for now.
3:35 p.m.
Conservative
3:35 p.m.
Matthew Johnson Director of Education, MediaSmarts
Thank you to the committee for inviting MediaSmarts to testify on this issue.
Our research suggests that algorithms and the collection of the data that make them work are poorly understood by youth. Only one in six young Canadians feel that the companies that operate social networks should be able to access the information they post there, and just one in 20 think advertisers should be able to access that information, but almost half of youth appear to be unaware that this is how most of these businesses make money.
With support from the Office of the Privacy Commissioner, we've been creating resources to educate youth about this issue and to teach them how to take greater control of their online privacy.
Algorithmic content curation is relevant to cyber-violence and youth in a number of ways. When algorithms are used to determine what content users see, they can make it a challenge to management one's online privacy and reputation. Because algorithms are typically mostly opaque in terms of how they work, it can be hard to manage your online reputation if you don't understand why certain content appears at the top of searches for you. Algorithms can also present problems in terms of how they deliver content, because they an embody their creator's conscious or unconscious biases and prejudices.
I believe Ms. Chemaly testified before this committee about how women may be shown different want ads than men. There are other examples that are perhaps more closely related to cyber-violence. Auto-correct programs that won't complete the words “rape” or “abortion”, for example, or Internet content filters, which are often used in schools, may prevent students from accessing legitimate information about sexual health or sexual identity.
This is why it remains vital that youth learn both digital and media literacy skills. One of the core concepts of media literacy is the idea that all media texts have social and political implications, even if those weren't consciously intended by the producers. This is entirely true of algorithms as well and may be particularly relevant because we're so rarely aware of how algorithms are operating and how they influence content that we see.
Even if there is no conscious bias involved in the design of algorithms, they can be the product and embodiment of our unconscious assumptions, such as one algorithm that led to a delivery service not being offered in minority neighbourhoods in the United States. Similarly, algorithms that are designed primarily to solve a technical problem, without any consideration of the possible social implications, may lead to unequal or even harmful results entirely accidentally.
At the same time, a group that is skilled at gaming algorithms can amplify harassment by what's called “brigading”: boosting harmful content in ways that make it seem more relevant to the algorithm, which can place it higher in search results or make it more likely to be delivered to audiences as a trending topic. This was an identified problem in the recent U.S. election, where various groups successfully manipulated several social networks' content algorithms to spread fake news stories. Also, it could be easily used to greatly magnify the reach of an embarrassing or intimate photo, for example, that was shared without the subject's consent.
Manipulating algorithms in this way can also be used to essentially silence victims of cyber-violence, especially in platforms that allow for downvoting content as well as upvoting.
In terms of digital literacy, it's clear that we need to teach students how to recognize false and biased information. Our research has found that youth are least likely to take steps to authenticate information that comes to them via social media, which, of course, is where they get most of their information. We need to educate them about the role that algorithms play in deciding what information they see. We also need to promote digital citizenship, both in terms of using counter-speech to confront hate and harassment, and in terms of understanding and exercising their rights as citizens and consumers. For example, there have been a number of cases where consumer action has successfully led to modifying algorithms that were seen to embody racist or sexist attitudes.
Thank you.
3:40 p.m.
Conservative
The Chair Conservative Marilyn Gladu
That was excellent. We will go to our first round of questioning.
We're going to start with you, Ms. Vandenbeld, for seven minutes.
3:40 p.m.
Liberal
Anita Vandenbeld Liberal Ottawa West—Nepean, ON
Thank you.
Thank you very much for returning to help us delve a little deeper into some of these topics. I have a few questions for clarification, and then I'd like to talk a bit about the regulation of corporations and how we can have more transparency.
First of all, this is the first time I've heard about brigading and about manipulating the algorithms, which of course is quite alarming. Could I hear a bit more about that from Mr. Johnson?
Also, for Ms. Bailey, my understanding has always been that what you put into search terms determines the kinds of things that you see. From your testimony, it sounds like it's bigger than that. It's also more predictive. It's the group that you're in. I'm not entirely sure how that can be programmed in. Could you both clarify that?
Then I'd like to hear more about the corporate regulation.
Mr. Johnson.
3:40 p.m.
Director of Education, MediaSmarts
I'm not really sure what to add on the topic. We've seen a number of cases in which, either through what you might call savvy gaming of algorithms or, in some cases, just brute force, people have been able to manipulate things such as trending topics.
One study that was done of fake news leading up to the U.S. election found that there was actually a small group of writers in Macedonia producing content who managed to get it spread to a tremendous number of people, because they understood how the initial readers were going to interact with it and how that would influence how the platforms promoted that content. To my knowledge, I haven't seen that used yet on a wide scale for harassment of individuals, but the same technique certainly could easily be used.
3:40 p.m.
Liberal
Anita Vandenbeld Liberal Ottawa West—Nepean, ON
You suggested that it can also be used not just to amplify certain news items, but also to silence others. How does that happen?
3:40 p.m.
Director of Education, MediaSmarts
In part, it happens because any time one message is being amplified, others get lowered. Someone's message is less likely to be trending if someone else's is. There's a limited number of spots.
Some platforms also have what's called “downvoting”, in which users not only can boost a signal of one thing but can also say that another thing is less relevant. If you have a savvy group that is boosting the signal of the harassment and, when the victim and the victim's allies are trying to get a message out, they are downvoting that, they can essentially be working in both directions. We did see that to a certain extent with the “Gamergate” situation, in which women were being harassed in the games industry. Some of those techniques were being used, although they weren't quite as technically sophisticated as what we've seen recently with the fake news situation.
3:45 p.m.
Liberal
Anita Vandenbeld Liberal Ottawa West—Nepean, ON
Basically, those who understand how the algorithm works and anticipates human behaviour can then manipulate what gets amplified, so that if something is repeated often enough, people think it's true. Is that it? Okay.
3:45 p.m.
Voices
Oh, oh!
3:45 p.m.
Liberal
Anita Vandenbeld Liberal Ottawa West—Nepean, ON
I'd like to move to Ms. Bailey and talk a bit about what you were saying in terms of the transparency. Of course, we're looking at where the federal government would be able to find remedies for these sorts of things, so are there ways in which we can force corporations to be more transparent—whether that would actually solve the problem—and how would we do that?
3:45 p.m.
Professor, Faculty of Law, University of Ottawa, As an Individual
There are models in the EU in particular, in the EU directives around data privacy, that focus more on bringing human decision-making into the loop. Where a decision is made that affects someone's life chances, for example, there needs to be some sort of human element in the determination of the result.
Again, this adds a certain level of accountability or transparency, where neither you nor I—or maybe even some computer scientist—could actually explain what the algorithm did in terms of how it came to the conclusion that you were in a particular group, or that certain information should come to you or not. Thus, we can have some other form of explanation about what is actually being taken into account in determining what kind of information it is that we're seeing and why a particular decision is being made about us. This is becoming more and more important as we move toward machine-made decision-making in all kinds of atmospheres.
I think people or countries are beginning to think about ways in terms of how to put the “public” in public values and public discourse back into decision-making in this sphere, which, although it is largely privately controlled is really a public infrastructure, in terms of a necessity for people to have access to it increasingly for work, for social life, and for education. It's about how to think about righting the balance between the decisions being made from a private sector perspective—not for nefarious reasons, but for profit reasons, because that's what they're in business to do—and how we re-inject public conversation and public discourse around the issues in terms of what's happening, what kinds of decisions are being made, how people are being profiled, and how they're being categorized. I think this is a really important start.
3:45 p.m.
Liberal
Anita Vandenbeld Liberal Ottawa West—Nepean, ON
So essentially there's a multiplier effect. It's beyond just an echo chamber of all your friends on social media liking things and therefore you'll see only the same things. This is as opposed to the old days, when you'd flip through the newspaper and be exposed to all kinds of things. Now the algorithm is picking that up and then reinforcing it. Is that...?
3:45 p.m.
Professor, Faculty of Law, University of Ottawa, As an Individual
It could be that some do that. In other words, they close our circle instead of opening our circle.
In some cases, people who look at this may say, well, that's an advantage, because if I don't want to see hate speech, I don't have to see hate speech. But let's take Twitter's mute button as an example. I can mute somebody so that I don't see that they are attacking me online, but the fact of the matter is that they are attacking me online and I don't know about that.
The way our worlds are being curated is that in some instances we might say that at least it relieves my pain in the initial moment. But in the long term, in terms of what violence is being done, what harassment is happening, and what issues we really need to be engaging in, it's a problem if we are closing ourselves off.
3:50 p.m.
Conservative
The Chair Conservative Marilyn Gladu
I'm sorry. That's your time.
I'll go to Ms. Vecchio now, for seven minutes.
3:50 p.m.
Conservative
Karen Vecchio Conservative Elgin—Middlesex—London, ON
Thank you.
Hi, and thank you very much for coming.
I want to start off with a personal story. Maybe you can share with me how this came about. You can say, no, these were algorithms—or maybe I had a bad past I don't know about—but what happened is this. I was on a flight the other day, and I watched two YouTube videos, parts one and two from the international advertising awards. I'll share with everybody that they were regarding men's underwear. It was a funny clip—very funny; two testicles; great.
After the first two videos, the third video, which automatically went to play, was pornography. It was a young man and a young woman. Unfortunately, I was sitting there with my 13-year-old son, and I went, “Oh, my gosh”, because the video itself that I was watching with my son wasn't too inappropriate—somewhat, but not too inappropriate—but I can tell you that the third thing absolutely should not have been there.
Would that have been an algorithm? Would that have been something from a previous search history, although I can tell you that I've never searched for pornography on YouTube? How would that have come up? Can you share with me your thoughts on how you start with something that's getting a national award for advertising and the third thing is pornography?
3:50 p.m.
Professor, Faculty of Law, University of Ottawa, As an Individual
First of all, let me give a disclaimer. I'm a lawyer, not a computer scientist, so to say specifically what was happening, I'm not sure. It could be many things. It could be an example of the sort of thing that Matthew was talking about, that there's an algorithm that's calculating what people who look at these two videos tend to like. It could reflect other users' preferences and be associating things together. It could reflect some less sophisticated algorithm that's searching for a term or a content like underwear, or testicles, or whatever the case may be. It's aggregating like content in that way. Or it could be, although you've told us that it's not, based on your own personal search history.
3:50 p.m.
Voices
Oh, oh!
3:50 p.m.
Professor, Faculty of Law, University of Ottawa, As an Individual
The more sophisticated algorithms get, the better they are supposed to be at predicting what we would actually want to see. If I look at my search history on Amazon, Amazon did this fairly early. I bought a lot of books about feminism, so Amazon constantly gave me ad suggestions that had anything to do with women, especially diet and exercise books. I was like, “I think your algorithm doesn't get it. An old, rad feminist is not looking for Suzanne Somers' diet and exercise book. You guys are way off.”
3:50 p.m.
Voices
Oh, oh!
3:50 p.m.
Professor, Faculty of Law, University of Ottawa, As an Individual
What you can see is that the algorithms are getting more sophisticated, and the more data you give them, the more predictive they become in terms of thinking about emulating your behaviour to the extent that your behaviour is premised on what you did in the past.
3:50 p.m.
Conservative
Karen Vecchio Conservative Elgin—Middlesex—London, ON
Okay. I just found it extremely.... Oh, my goodness, I didn't expect it, especially from the documentary that it came from. Actually, some of the other things were Pampers commercials, so you didn't expect something to fall into this. I was really quite surprised.
When we talk about this, I think that brings us into what protective measures we could also use. For something so simple as a commercial that within two plays gives us this sort of result, what can we do to protect in the long run? Is there a way we can downvote something when it comes to pornography on the Internet? Is there anything we could do there, not necessarily legislative, to make sure we're being more cautious? Is anything like that currently being done here in Canada?
December 5th, 2016 / 3:50 p.m.
Director of Education, MediaSmarts
Yes, absolutely.
One of the things that we educate young people about is, again, digital citizenship: their ability to make a difference online. We teach them, for instance, that when they see inappropriate content, particularly when it's something like cyber-bullying or hate content, there are a lot of steps they can take. Almost every platform, whether it's a video platform or a social network, has ways of reporting content. Many of them do have downvoting. That's one of the reasons downvoting exists, even though it can be misused. We teach them that they have a responsibility to do that, and that they have a right to have an online experience where they're not exposed to harassment and hate.
We also advocate and provide resources for parents to talk to their kids and for teachers to teach students about all of these different issues. We know that kids are going to be exposed to them, whether intentionally or unintentionally. We know it happens. We know that even the best filters don't block out all of this content, and often, when it comes to things like hate or cyber-bullying, filters don't do a good job.
It's important that we talk about these things, so that by the first time someone encounters pornography, they already know that it's not real, and they already know not to take it as a realistic or healthy view of sexuality.