From 80eb18b998d605d5b6a694370dbca49995205ad2 Mon Sep 17 00:00:00 2001 From: Owen Mundy Date: Thu, 12 Sep 2024 10:12:23 -0400 Subject: [PATCH] Adds Botelho essay --- papers/botelho/game/RadRecommenders.html | 3774 +++++++++++++++++++++ papers/botelho/game/rad-recommenders.html | 3764 -------------------- papers/botelho/index.php | 315 +- 3 files changed, 4073 insertions(+), 3780 deletions(-) create mode 100644 papers/botelho/game/RadRecommenders.html delete mode 100644 papers/botelho/game/rad-recommenders.html diff --git a/papers/botelho/game/RadRecommenders.html b/papers/botelho/game/RadRecommenders.html new file mode 100644 index 0000000..f293120 --- /dev/null +++ b/papers/botelho/game/RadRecommenders.html @@ -0,0 +1,3774 @@ + + + + +Rad Recommenders Botelho 2022 + + + + + + + + + + + + + + + + + + + + + + +
+
+

Browser lacks capabilities required to play.

Upgrade or switch to another browser.

+
Loading…
+
+ + + + diff --git a/papers/botelho/game/rad-recommenders.html b/papers/botelho/game/rad-recommenders.html deleted file mode 100644 index 16ec748..0000000 --- a/papers/botelho/game/rad-recommenders.html +++ /dev/null @@ -1,3764 +0,0 @@ - - - - -Rad Recommeners - - - - - - - - - - - - - - - - -
-
-

Browser lacks capabilities required to play.

Upgrade or switch to another browser.

-
Loading…
-
- - - - diff --git a/papers/botelho/index.php b/papers/botelho/index.php index 524823d..c958c65 100644 --- a/papers/botelho/index.php +++ b/papers/botelho/index.php @@ -15,12 +15,75 @@
-
-
+
+
+ + + + + +
+ thumb +
+ Image by screamonei +
+
+ + + +
Introduction
+ +

+ Companies are constantly vying for consumer attention, optimizing how much time + consumers spend on their platform, thus increasing advertising revenue. Technologists can make + hundreds of thousands of dollars per year to design these addictive systems. YouTube was one of + the first companies to implement this technology in 2005. The company has since grown and + boasts over two billion users (Singh, 2022). Pew Research Center found that 94% of Americans + between 18 and 24 use YouTube (Ibid). One of the main draws of YouTube is that people spend + more time than initially desired. Recommended videos and autoplay encourage this behavior. + Advertisers cash in on the captured attention of the audience, playing ads before, during, or even + within videos. +

+ + +

+ YouTube captures, coerces, and commodifies the attention of its users by learning what + they like and influencing user preferences. To understand how YouTube captivates its audience, + one must understand how recommender systems work. I will introduce the main types of + recommender systems: Content-based, Collaborative-filtering, and Knowledge-based (Ibid). + Then, I analyze the strategies and design decisions of YouTube and assess its effectiveness. + After, I will dissect how these tactics are manipulative by encouraging people to spend more + time on the platform. I then examine how recommender systems circulate harmful content like + pedophilia and far-right radicalization, encouraging people to tend toward radicalized extremes. + Finally, I propose various solutions for platform improvement to resist these persuasion tactics. +

+ + + + +
Recommender Systems
+

+ The first recommender system, Grundy, was created in 1979. Its purpose was to suggest + library books to readers. By the early 1990s, companies released commercial recommender systems, and in the late 1990s, Amazon invented the collaborative filtering technique, which has + since become widely popular. More recently, in 2006, Netflix hosted a contest with a + million-dollar prize for whoever could implement the best recommendation system, and finally, + in 2010, YouTube implemented a recommendation system (Qomariyah, 2020). +

+

+ Recommender systems can track user activity history or create a profile of similar users to + predict the type of content a user may want to see in the future. Content-based systems use + metadata on a user, such as clicks and likes to hone in on what they want to view. + Collaborative-filtering systems base suggestions on the interests of similar users. These systems + base recommendations on which content tends to be purchased or viewed in conjunction. Since + these systems look at other users, they make predictions based on a substantially larger dataset. + Knowledge-based recommender systems look at the attributes of both the user and the product. + "For example, such a system could identify attribute similarities between a job seeker’s resume + and a job description" (Singh, 2022). Singh points out that most algorithms employ all the above + systems, not just one. +

-
Introduction
@@ -28,8 +91,29 @@ +
Persuasion Tactics
+

+ YouTube’s ingenious algorithm represents a fundamental shift of the embrace of + classical machine learning methods like tree and linear-based methods to deep learning, + which can hold the complexity, yet adaptability and robustness required for modern-day + problems. Yet deep learning models need to be carefully used — not as a bashable + cookie-cutter solution to everything but as a delicate tool — which is purposed cleverly + by YouTube to their unique problems (Ibid). +

+

+ YouTube's tactics have led to $7 billion in ad revenue in 2021 alone. The pandemic has + expedited the growth of online shopping, making online advertisements a gold mine. YouTube + employs several design strategies to induce users into spending more time. Measures include + notifications to grab user attention or autoplay; a recommended video will play automatically + without a user having to do anything. Creators have tools to boost engagement, such as user data + on the exact moment a watcher clicked out of a video. YouTube creators also know that + subscriptions drive engagement. The company benefits tremendously from in-video advertising. + “A marketing study by ThinkByGoogle found that brand collaborations with YouTube creators + are ‘4x more effective at driving lift in brand familiarity than those with celebrities” (Wong, + 2021). +

@@ -44,39 +128,238 @@
- -
xxxxx
+ +
+ + +
+ Rad Recommenders by Blake Botelho + +

+ For as long as I have had access to a smartphone, I have been an avid user of YouTube. I am mesmerized by the algorithm's ability to recommend enjoyable videos. In my Digital Studies minor, I have wrestled with the many drawbacks that modern-day technology and software can have on society and the benefits. I am particularly passionate about understanding the effects of recommender systems, namely YouTube. +

+ +

+ After deciding that I wanted to work on a game that captures the experience someone may have using a recommender system, classmate Evan Rothman suggested I create a game where users click on which videos to watch. The game, now titled Rad Recommenders, has since blossomed into its final version, where users can simulate the experience watching YouTube of one of two playable characters: Billy or Danny. Once a person chooses their character, I provide narrative background on each. I then pick an initial YouTube page based on the pre-assigned demographic information and interests. The rest of the game is a mix of YouTube pages and narration. The player can choose from which videos they want more recommended content. This game enables players to take many different paths. It can be played at a slow pace, watching each of the videos on each page, or relatively quickly. +

+ +

+ Rad Recommenders tackles pressing topics, notably Xenophobia and Sexism. Depending on the path they take, users can encounter upsetting videos. I hope that people realize the unintentional effects recommender systems can have when not carefully designed to account for them. I did not create this game to demonize YouTube but to offer criticism to better the platform. From the player's perspective, I hope that people are more mindful of the effects recommender systems can have on them. I offer solutions on how legislators can improve recommender systems. I encourage people to wrestle with these proposed solutions, expanding upon and improving them. +

+ +
+
-
+ + +
+
+
+ +

+ According to the Chief Product Officer, people spend 70% of their time watching + recommended videos on YouTube. The company must overcome the following difficulties: Scale + (recommendation systems can be harder to implement on YouTube given the sheer volume of + videos), videos are constantly being uploaded, so there needs to be a way to balance new content and old content, and user behavior can be complicated. To overcome this, YouTube starts by + creating a list of a few hundred videos the user may enjoy and then using a second neural + network to display the top three videos as recommendations. They show users videos the + company has assessed that the user is most likely to watch for the most time. The algorithm is + more likely to spread controversial content that sparks a strong reaction and keeps users engaged. + Human monitors remove videos that violate community standards, but the mental toll this takes + on the checkers can often be severe (Ye, 2020). +

+ + + + +
How YouTube Manipulates Users
+ + +

+ Nick Seaver compares recommender algorithms to animal traps. Companies capture user + retention the same way a hunter traps an animal (Seaver, 2019). Seaver’s analogy serves as a + framework for viewing the manipulative aspect of recommender systems. The user molds the + algorithm, and the algorithm molds the user. The algorithm can be self-fulfilling since + recommended videos can have a placebo effect. Adomavicius et al. conducted an experiment + where consumers listened to and rated songs between 1 and 5 stars. Experimenters randomly + assigned recommendations, yet consumers were willing to pay between 12% and 17% more for a + one-star increase in a song rating. Researchers persuaded consumers to spend more simply + because the machine indicates they would like something, even if the recommendation is based + on nothing (Adomavicius, et al 2019). +

+ -
-
-
-

Bibliography

+
Radicalization
-
    +

    + Beyond the time manipulation on YouTube lies an even more serious problem: + radicalization. Sinead Bovell from Wired experimented to see how YouTube polarizes its + audience. Bovell created three accounts. She watched and liked a left-leaning video from + MSNBC, a center-leaning video from ABC on another account, and a right-leaning Fox News video on a third account. After liking each video, her recommended videos were drastically + different between each. The MSNBC account recommended liberal content, the ABC account + moderate content, and the Fox News account conservative content. (WIRED, 2020). +

    -
  1. xxxx
    2. +

    + Dr. Tim Kessler conducted a similar experiment. He posed as his Dad by creating profiles + on different platforms. After entering that he liked to eat meat and pasta on a recipe website, the + recommender continued to suggest meat and pasta recipes. To prove how these recommender + systems can negatively impact physical health in addition to mental health, Dr. Kessler ate what + he was recommended for a month. This experiment resulted in him gaining weight and becoming + ill. On Facebook, after entering his elderly Dad’s demographic information and interests (history, + golf, documentaries, etc.), Facebook recommended over 50 neo-Nazi groups in a month (TEDx + Talks, 2019). +

    -
+

+ This startling revelation extends to YouTube as well. Caolan Robertson is a prime + example. Robertson is a gay man from Ireland who has faced prolonged hostility due to his + sexuality. When Robertson moved to London, he experienced the most homophobia when + walking in predominately Muslim neighborhoods. Robertson began watching YouTube videos + about the Orlando nightclub shooting at a gay club in 2016. He became exposed to far-right + perspectives on the tragic event. After watching these videos, Robertson became staunchly + Islamophobic and anti-immigrant. YouTube's curated chorus of hateful voices affirmed his + right-winged views. His perspective changed after the mosque shooting in New Zealand. He saw + how online radicalization could affect different groups, not just his own, and he has since worked + to de-radicalize others. His story shows the dangers of unchecked algorithms (Metz, 2021). +

+

+ YouTube’s recommender system has also enabled pedophiles to use the platform to view + videos of young children. Christianne posted a YouTube video of her young daughter and a friend playing in a pool. Within days, the view count skyrocketed to 400,000. Users were + purposefully seeking out content that showed young children, and YouTube connected them with + these and other similar videos. This video spread months after people criticized YouTube for this + exact issue. A group of researchers found that on sexually themed videos, as videos became + more abnormal, they placed a greater emphasis on youth. For example, videos of women + discussing sex eventually led to videos of teenage moms breastfeeding. (Fisher and Taub, 2019). + Innocent videos of children were being recommended to pedophiles, and some of these videos + had view counts in the millions (Ibid). YouTube’s algorithm suffers from and has suffered from + issues that enable egregious behaviors and viewpoints to flourish. Lawmakers, and others alike, + have proposed legislation and solutions to combat these issues. +

-
+ +
Solutions
+ +

+ People have proposed solutions to mitigate the dangers of profit-maximizing + recommender systems. Stray et al. suggest that imitation learning can be a more predictable + method. They also recommend a feature that shows a receipt of where a person has clicked or + shows they have watched. By having a clearer understanding of how an algorithm sways + decision-making, people can avoid feeling manipulated. Through the ability to review their + course of action, a person has more autonomy to avoid previous traps that lead them down + content viewing paths. Profit-maximizing companies would not likely apply this method without + being required. Legislative enforcement can ensure they do and with it more algorithmic + transparency. +

+ +

+ A new Chinese law has been pioneering stricter regulations on recommender systems. + This law includes the following: companies must inform users how a recommender system + works, users need to be able to opt-out, users must have the option to get rid of tags used to make recommendations, and companies must protect the elderly from recommender system scams. + Laws like these increase the transparency and agency of consumers by allowing users to take + ownership over what they see. Consumers can still benefit from recommendation systems' ability + to connect them with relevant content while lessening the potential risks (Kharpal, 2022). +

+ + + + + +
Conclusion
+ +

+ This piece has given an overview of how recommendation algorithms work to maintain + user attention. In commodifying its users, platforms walk a dangerous line. Provocative content + can be the most reactionary and the most successful at accomplishing the purpose of the + algorithm. YouTube is known to encourage radicalization. Studies have confirmed how + recommendation systems coerce people and provide a re-enforcement loop that can lead to bad + outcomes. People watch more extreme videos over time until they have reached the point of + radicalization. Without regulation, these systems can and will hurt people in the process. + Thankfully, there are codifiable suggestions discussed in the Solutions section that safeguard + against these pitfalls and improve recommender systems. +

+ + + + + + +
+
+
+ + + + + + + +
+
+
+ +

Bibliography

+ + +
    + +
  1. Adomavicius, Gediminas, Jesse Bockstedt, Shawn P. Curley, Jingjing Zhang, and Sam + Ransbotham. 2019. “The Hidden Side Effects of Recommendation Systems.” MIT + Sloan Management Review 60 (2): 1.
    2. +
  2. Fisher, Max, and Amanda Taub. 2019. “On YouTube’s Digital Playground, an Open Gate + for Pedophiles.” The New York Times, June 3, 2019, sec. World. + https://www.nytimes.com/2019/06/03/world/americas/youtube-pedophiles.html.
    3. +
  3. Kharpal, Arjun. 2022. “China’s next Regulatory Target — Algorithms, the Secret of + Many Tech Giants’ Success.” CNBC. January 7, 2022. + https://www.cnbc.com/2022/01/07/china-to-regulate-tech-giants-algorithms-in-un + precedented-move.html.
    4. +
  4. Metz, Cada “Feeding Hate with Video: A Former Alt-Right YouTuber Explains His + Methods.” 2021. The Denver Post (blog). May 22, 2021. + https://www.denverpost.com/2021/05/22/youtube-alt-right-racism-internet/.
    5. +
  5. “Our Reality Shaped by Recommendation Algorithms | Dr. Tim Kessler | TEDxTUBerlin + - YouTube.” n.d. Accessed February 9, 2022. + https://www.youtube.com/watch?v=bPdRBXpjENI.
    6. +
  6. Qomariyah, Nunung Nurul . “Definition and History of Recommender Systems.” n.d. + Computer Science (blog). Accessed March 1, 2022. + https://international.binus.ac.id/computer-science/2020/11/03/definition-and-histo + ry-of-recommender-systems/.
    7. +
  7. Seaver, Nick. 2019. “Captivating Algorithms: Recommender Systems as Traps.” Journal + of Material Culture 24 (4): 421–36. https://doi.org/10.1177/1359183518820366.
    8. +
  8. Singh, Spandana. “Why Am I Seeing This?” n.d. New America. Accessed February 16, + 2022. http://newamerica.org/oti/reports/why-am-i-seeing-this/.
    9. +
  9. Stray, Jonathan, Ivan Vendrov, Jeremy Nixon, Steven Adler, and Dylan Hadfield-Menell. + 2021. “What Are You Optimizing for? Aligning Recommender Systems with + Human Values.” ArXiv:2107.10939 [Cs], July. http://arxiv.org/abs/2107.10939.
    10. +
  10. WIRED. 2020. Is the YouTube Algorithm Controlling Us? + https://www.youtube.com/watch?v=XuORTmLhIiU.
    11. +
  11. Wong, Michael. “[UX Case Study] How YouTube Keeps You Watching and Addicted.” + 2021. The Designership. September 23, 2021. + https://thedesignership.com/how-youtube-keeps-you-watching-addicted/.
    12. +
  12. Ye, Andre. 2020. “The Algorithm Worth Billions: How YouTube’s Addictive Video + Recommender Works.” Medium. May 22, 2020. + https://faun.pub/the-algorithm-worth-billions-how-youtubes-addictive-video-reco + mmender-works-d75646dac6a3.
    13. + +
+ + +
+
-
- + \ No newline at end of file