May 20, 2018

Archives for March 2018

When The Choice Is To Delete Facebook Or Buy A Loaf Of Bread

By Julieanne Romanosky and Marshini Chetty

In the last week, there has been a growing debate around Facebook and privacy. On Twitter, the newly formed #deletefacebook movement calls for users who are upset over the data breach of over 50 million Facebook accounts by Cambridge Analytica to rid themselves of the platform altogether. But like others have stated, deleting Facebook may not be the easy option for everyone on the platform because in some countries, Facebook is the Internet. In fact, in 63 countries around the world, Facebook has introduced the Free Basics platform which includes Facebook and offers marginalized users limited “free” browsing on the Internet. More importantly, our recent study, jointly conducted with the University of Maryland [5], suggests that deleting Facebook and Free Basics for low income users could be the difference between saving enough money to afford a loaf of bread or not.

What is Facebook’s Free Basics and why is it being used by low income users?: Free Basics was founded in 2013 by Facebook with the goal of connecting rural and low-income populations to the Internet for the first time. While Free Basics appears as a single app, it is actually a platform for hosting a variety of data-charge free or “zero-rated” applications and the available content changes depending on the country and unpaid partnerships with local service providers, i.e., no two Free Basics offerings are the same. However, all versions provide access to a lite version of Facebook (with no images or video) and select other third party apps such as Bing and Wikipedia. Educational materials, news, weather reports dominate the application topics in Free Basics across countries. Other apps cover health care, job listings, search engines, and classifieds. Here is what the app interface looks like in South Africa:

Free Basics in South Africa

What did we do to investigate Facebook and Free Basics usage?: We interviewed 35 Free Basics users in South Africa, one of the countries that the platform is offered in. We spoke to a combination of current low-income users and non-regular student users. Including both groups in our study allowed us to form a more comprehensive understanding of the impact of zero-rated services, the factors that affect the adoption of these services, and the possible use of these services in more developed countries than if we studied users or non-users alone or those who were unconnected and low-income only. Both groups were asked to talk about their online habits (i.e. time spent online, what websites or apps they used etc), how much money they typically spent on Internet access, and how, if at all, they worked to keep their mobile Internet costs down.

How do low income users use Facebook’s Free Basics?: We found, particularly, the low income users on Free Basics were able to cut their mobile data costs significantly, with one participant in our study exclaiming that they could now afford a loaf of bread with the money saved from being online for “free”. The service also drove users to the “free” apps included in the platform even when they preferred other apps that were not “free” to use. Interestingly, all the participants who used Free Basics regularly were not “unconnected” users who had never been online prior to using the platform. Instead, these participants had been using the Internet as paying customers but they had heard about the platform from others (often through word of mouth) as a way to save on Internet costs. For these users, deleting Facebook and its relevant resources would be like deleting a lifeline in an already expensive data landscape. The platform was not without limitations for our participants however. Since our participants were already online, they were also very conscious of the fact that the apps included in the platform were, in their perception, “second-rate” – for instance, the Facebook app on the platform does not include images or video unless users pay for them. [Read more…]

Is affiliate marketing disclosed to consumers on social media?

By Arunesh Mathur, Arvind Narayanan and Marshini Chetty

YouTube has millions of videos similar in spirit to this one:

The video reviews Blue Apron—an online grocery service—describing how it is efficient and cheaper than buying groceries at the store. The description of the video has a link to Blue Apron which gets you a $30 off your first order, a seemingly sweet offer.

The video’s description contains an affiliate link (marked in red).

What you might miss, though, is that the link in question is an “affiliate” link. Clicking on it takes you through five redirects courtesy of Impact—an affiliate marketing company—which tracks the subsequent sale and provide a kickback to the YouTuber, in this case Melea Johnson. YouTubers use affiliate marketing to monetize their channels and support their activities.

This example is not unique to YouTube or affiliate marketing. There are several marketing strategies that YouTubers, Instagrammers, and other content creators on social media (called influencers in marketing-speak) engage in to generate revenue: affiliate marketing, paid product placements, product giveaways, and social media contests.

Endorsement-based marketing is regulated. In the United States, the Federal Trade Commission requires that these endorsement-based marketing strategies be disclosed to end-users so they can give appropriate weightage to content creators’ endorsements. In 2017 alone, the FTC sent cease and desist letters to Instagram celebrities who were partnering with brands and reprimanded YouTubers with gaming channels who were endorsing gambling companies—all without appropriate disclosure. The need to ensure content creators disclose will likely become all the more important as advertisers and brands attempt to target consumers on consumers’ existing social networks, and as lack of disclosure causes harm to end-users.

Our research. In a paper that is set to appear at the 2018 IEEE Workshop on Consumer Protection in May, we conducted a study to better understand how content creators on social media disclose their relationships with advertisers to end-users. Specifically, we examined affiliate marketing disclosures—ones that need to accompany affiliate links—-which content creators placed along with their content, both on YouTube and Pinterest.

How we found affiliate links. To study this empirically, we gathered two large datasets consisting of nearly half a million YouTube videos and two million Pinterest pins. We then examined the description of the YouTube videos and the Pinterest pins to look for affiliate links. This was a challenging problem, since there is no comprehensive public repository of affiliate marketing companies and links.

However, affiliate links do contain predictable patterns, because they are designed to carry information about the specific content creator and merchant. For instance, an affiliate link to Amazon contains the tag URL parameter that carries the name of the creator who is set to make money from the sale. Using this insight, we created a database containing all sub-domains, paths and parameters that appeared with a given domain. We then examined this database and manually classified each entry either as affiliate or non-affiliate by searching for information about the organization owning that domain and sometimes even signing up as affiliates to validate our findings. Through this process, we compiled a list of 57 URL patterns from 33 affiliate marketing companies, the most comprehensive publicly available curated list of this kind (see Appendix in the paper, and GitHub repo).

How we scanned for disclosures. We could expect to find affiliate link disclosures either in the description of the videos or pins, during the course of the video, or on the pin’s image. We began our analysis by manually inspecting 20 randomly selected affiliate videos and pins, searching for any mention about the affiliate nature of the accompanying URLs. We found that none these videos or pins conveyed this information.

Instead, we turned our attention to inspecting the descriptions of the videos and pins. Given that any sentence (or phrase) could contain a disclosure, we first parsed descriptions into sentences using automated methods. We then clustered these sentences using hierarchical clustering, and manually identified the clusters of sentences that represented disclosure wording.

What we found. Of all the YouTube videos and Pinterest pins that contained affiliate links, only ~10% and ~7% respectively contained accompanying disclosures. When these disclosures were present, we could classify them into three types:

  • Affiliate link disclosures: The first type of disclosures simply stated that the link was an “affiliate link”, or that “affiliate links were included”. On YouTube and Pinterest these type of disclosures were present on ~7% and 4.5% of all affiliate videos and pins respectively.
  • Explanation disclosures: The second type of disclosures attempted to explain what an affiliate link was, on the lines of “This is an affiliate link and I receive a commission for the sales”. These disclosures—which are of the type the FTC expects in its guidelines—only appeared ~2% each of all affiliate videos and pins.
  • Support channel disclosures: Finally, the third type of disclosures—exclusive to YouTube—told users that they would be supporting the channel by clicking on the links in the description (without exactly specifying how). These disclosures were present in about 2.5% of all affiliate videos.

In the paper, we present additional findings, including how the disclosures varied by content type, and compare the engagement metrics of affiliate and non-affiliate content.

Cause for concern. Our results paint a bleak picture: the vast majority of affiliate content on both platforms has no accompanying disclosures. Worse, Affiliate link disclosures—ones that the FTC specifically advocates against using—were the most prevalent. In future work, we hope to investigate the reason behind this lack of disclosure. Is it because the affiliates are unaware that they need to disclose? How aware are they of the FTC’s specific guidelines?

Further, we are concluding a user study that examines the efficacy of these disclosures as they exist today: Do users think of affiliate content as an endorsement by the content creator? Do users notice the accompanying disclosures? What do the disclosures communicate to users?

What can be done? Our results also provide several starting points for improvement by various stakeholders in the affiliate marketing industry. For instance, social media platforms can do a lot more to ensure content creators disclose their relationships with advertisers to end-users, and that end-users understand the relationship. Recently, YouTube and Instagram have taken steps in this direction, releasing tools that enable disclosures, but it’s unlikely that any one type of disclosure will cover all marketing practices.

Similarly, affiliate marketing companies can hold their registered content creators accountable to better standards. On examining the affiliate terms and conditions of the eight most common affiliate marketing companies in our dataset, we noted only two explicitly pointed to the FTC’s guidelines.

Finally, we argue that web browsers can do more in helping users identify disclosures by means of automated detection of these disclosures and content that needs to be disclosed. Machine learning and natural language processing techniques can be of particular help in designing tools that enable such automatic analyses. We are working towards building a browser extension that can detect, present and explain these disclosures to end-users.

Artificial Intelligence and the Future of Online Content Moderation

Yesterday in Berlin, I attended a workshop on the use of artificial intelligence in governing communication online, hosted by the Humboldt Institute for Internet and Society.

Context

In the United States and Europe, many platforms that host user content, such as Facebook, YouTube, and Twitter, have enjoyed safe harbor protections for the content they host, under laws such as Section 230 of the Communications Decency Act (CDA), the Digital Millenium Copyright Act (DMCA), and in Europe, Articles 12–15 of the eCommerce Directive. Some of these laws, such as the DMCA, provide immunity to platforms for copyright damages if the platforms remove content based on knowledge that it is unlawful. Section 230 of the CDA provides broad immunity to platforms, with the express goals of promoting economic development and free expression. Daphne Keller has a good summary of the legal landscape on intermediary liability.

Platforms are now facing increasing pressure to detect and remove illegal (and, in some cases, legal-but-objectionable) content. In the United States, for example, bills in the House and Senate would remove safe harbor protection for platforms that do not remove illegal content related to sex trafficking. The European Union has also considering laws that would limit the immunity of platforms who do not remove illegal content, which in the EU includes four categories: child sex abuse, incitement to terrorism, certain types of hate speech, and intellectual property or copyright infringement.

The mounting pressure on platforms to moderate online content coincides with increasing attention to algorithms that can automate the process of content moderation (“AI”) for the detection and ultimate removal of illegal (or unwanted) content.

The focus of yesterday’s workshop was to explore questions surrounding the role of AI in moderating content online, and the possible implications of AI for the moderation of online content and how online content moderation is governed.

Setting the Tone: Challenges for Automated Filtering

Malavika Jayaram from Digital Asia Hub and I delivered the two opening “impulse statements” for the day. Malavika talked about some of the inherent limitations of AI for automated detection (with a reference to the infamous “Not Hot Dog” app) and pointed out some of the tools that platforms are being pressured to use automated content moderation tools.

I spoke about our long line of research on applying machine learning to detect a wide range of unwanted traffic, ranging from spam to botnets to bulletproof scam hosting sites. I then talked about how the dialog has in some ways used the technical community’s past success in spam filtering to suggest that automated filtering of other types of content should be as easy as flipping a switch. Since spam detection was something we knew how to do, then surely the platforms could also ferret out everything from copyright violations to hate speech, right?

In practice Evan Engstrom and I previously wrote about the difficulty of applying automated filtering algorithms to copyrighted content.  In short, even with a database that matches fingerprints of audio and video content against fingerprints of known copyrighted content, the methods are imperfect. When framing the problem in terms of incitement to violence or hate speech, automated detection becomes even more challenging, due to “corner cases” such as parody, fair use, irony, and so forth. A recent article from James Grimmelmann summarizes some of these challenges.

What I Learned

Over the course of the day, I learned many things about automated filtering that I hadn’t previously thought about.

  • Regulators and platforms are under tremendous pressure to act, based on the assumption that the technical problems are easy.  Regulators and platforms alike are facing increasing pressure to act, as I previously mentioned. Part of the pressure comes from a perception that detection of unwanted content is a solved problem. This myth is sometimes perpetuated by the designers of the original content fingerprinting technologies, some of which are now in widespread use. But, there’s a big difference between testing fingerprints of content against a database of known offending content and building detection algorithms that can classify the semantics of content that has never been seen before. An area where technologists can contribute to this dialog is in studying and demonstrating the capabilities and limitations of automated filtering, both in terms of scale and accuracy. Technologists might study existing automated filtering techniques or design new ones entirely.
  • Takedown requests are a frequent instrument for censorship. I learned about the prevalence of “snitching”, whereby one user may request that a platform take down objectionable content by flagging the content or otherwise complaining about it—in such instances, oppressed groups (e.g., Rohingya Muslims) can be disproportionately targeted by large campaigns of takedown requests. (It was not known whether such campaigns to flag content have been automated on a large scale, but my intuition is that they likely are.) In such cases, the platforms err on the side of removing content, and the process for “remedy” (i.e., restoring the content) can be slow and tedious. This process creates a lever for censorship and suppression of speech.The trends are troubling: according to a recent article, a year ago, Facebook removed 50 percent of content that Israel requested be removed; now that figure is 95 percent.  Jillian York runs a site where users can report these types of takedowns, but these reports and statistics are all self-reported. A useful project might be to automate the measurement of takedowns for some portion of the ecosystem or group of users.
  • The larger platforms share content hashes of unwanted content, but the database and process are opaque. About nine months ago, Twitter, Facebook, YouTube, and Microsoft formed the Global Internet Forum to Counter Terrorism. Essentially, the project relies on something called the Shared Industry Hash Database. It’s very challenging to find anything about this database online aside from a few blog posts from the companies, although it does seem in some way associated with Tech Against Terrorism.The secretive nature of the shared hash database and the process itself has a couple of implications. First, the database is difficult to audit—if content is wrongly placed in the database, removing it would appear next to impossible. Second, only the member companies can check content against the database, essentially preventing smaller companies (e.g., startups) from benefitting from the information. Such limits in knowledge could ultimately prove to be a significant disadvantage if the platforms are ultimately held liable for the content that they are hosting. As I discovered throughout the day, the opaque nature of commercial content moderation proves to be a recurring theme, which I’ll return to later.
  • Different countries have very different definitions of unlawful content. The patchwork of laws governing speech on the Internet makes regulation complicated, as different countries have different laws and restrictions on speech. For example, “incitement to violence” or “hate speech” might mean a different thing in Germany (where Nazi propaganda is illegal) than it does in Spain (where it is illegal to insult the king) or France (which recently vowed to ferret out racist content on social media). When applying this observation to automated detection of illegal content, things become complicated. It becomes impossible to train a single classifier that can be applied generally; essentially, each jurisdiction needs its own classifier.
  • Norms and speech evolve over time, often rapidly. Several attendees observed that most of the automated filtering techniques today boil down to flagging content based on keywords. Such a model can be incredibly difficult to maintain, particularly when it comes to detecting certain types of content such as hate speech. For one, norms and language evolve; a word that was innocuous or unremarkable today could take on an entirely new meaning tomorrow. Complicating matters further, sometimes people try to regain control in an online discussion by co-opting a slur; therefore, a model that bases classification on the presence of certain keywords can produce unexpected false positives, especially in the absence of context.

Takeaways

Aside from the information I learned above, I also took away a few themes about the state of online content moderation:

  • There will likely always be a human in the loop. We must figure out what role the human should play. Detection algorithms are only as good as their input data. If the data is biased, if norms and language evolve, or if data is mislabeled (an even more likely occurrence, since a label like “hate speech” could differ by country), then the outputs will be incorrect. Additionally, algorithms can only detect proxies for semantics and meaning (e.g., an ISIS flag, a large swath of bare skin) but have much more difficulty assessing context, fair use, parody, and other nuance. In short, on a technical front, we have our work cut out for us. It was widely held that humans will always need to be in the loop, and that AI should merely be an assistive technology, for triage, scale, and improving human effectiveness and efficiency when making decisions about moderation. Figuring out the appropriate division of labor between machines and humans is a challenging technical, social, and legal problem.
  • Governance and auditing is currently challenging because decision-making is secretive. The online platforms currently control all aspects of content moderation and governance. They have the data; nobody else has it. They know the classification algorithms they use and the features they use as input to those algorithms; nobody else knows them. They also are the only ones who have insight into the ultimate decision-making process. This situation is different from other unwanted traffic detection problems that the computer science research community has worked on, where it was relatively easy to get a trace of email spam or denial of service traffic, either by generating it or by working with an ISP. In this situation, everything is under lock and key.This lack of public access to data and information makes it difficult to audit the process that platforms are currently using, and it also raises important questions about governance:
    • Should the platforms be the ultimate arbiter in takedown and moderation?
    • Is that an acceptable situation, even if we don’t know the rules that they are using to make those decisions?
    • Who trains the algorithms, and with what data?
    • Who gets access to the models and algorithms? How does disclosure work?
    • How does a user learn that his or her content was taken down, as well as why it was taken down?
    • What are the steps to remedy an incorrect, unlawful, or unjust takedown request?
    • How can we trust the platforms to make the right decisions when in some cases it is in their financial interests to suppress speech? History has suggested that trusting the platforms to do the right thing in these situations can lead to restrictions on speech.

Many of the above questions are regulatory. Yet, technologists can play a role for some aspects of these questions. For example, measurement tools might detect and evaluate removal and takedowns of content for some well-scoped forum or topic. A useful starting point for the design of such a measurement system could be a platform such as Politwoops, which monitors tweets that politicians have deleted.

Summary

The workshop was enlightening. I came as a technologist wanting to learn more about how computer science might be applied to the social and legal problems concerning content moderation; I came away with a few ideas, fueled by exciting discussion. The attendees were an healthy mix of computer scientists, regulators, practitioners, legal scholars, and human rights activists. I’ve worked on censorship of Internet protocols for many years, but in some sense measuring censorship can feel a little bit like looking for one’s key under the lamppost—my sense is that the real power over speech is now held by the platforms, and as a community we need new mechanisms—technical, legal, economic, and social—to hold them to account.