Recommendation Algorithms and Ethics

Recommendation algorithms are only ever increasing in prevelance and complexity.Recommendation algorithms are used in nearly every mainstram application. Examples include YouTube, Twitch, Etsy, Ebay, Amazon, TikTok, Instagram, Twitter, Netflix, Snapchat, Hulu, Disney Plus, Facebook, Google News, and MSN. Being used in so many popular applications, it would almost be weird if someone with access to modern technology did not interact with these recomemendation algorithms every day. Given that, I believe they are an important subject to analyze the ethics of. Keep note that what you are about to read is my opinion and is lacking sources. This article is written for the first module of Tech Roulette 2021, Justice Matrix - Project 4.

The goal of recommendation algorithms

The stated goal of recommendation algorithms shifts from industry to industry. However, one thing stays constant - they are built to maximize profits by maximizing user engagement. I am not against the idea of companies trying to maximize their profits. However, it is important to analyze the means by which they do it and the effects to ensure that it does not violate the rights of any person. It is also important to analyze if eeking out those extra profits hurt other areas of the business.

Respecting the user's time

Recommendation algorithms for entertainment/social media are designed to serve the user the content that it thinks will keep the user engaged all the time - they never to give the user content they think will cause them to log off. This makes it the user's decision for whether they should continue to use the platform or to log off. This seems reasonable. Why should the platform decide whether it's best the user log off? They can make their own decisions, right? Obviously, people aren't literally scrolling through TikTok until they die (I think). At some point, outside factors like fatigue, personal needs, and growing boredom from the often reptitiveness generally get users to log off at some point. However, I think another strong motivator is guilt. Everyone has goals for things that they want to accomplish in a day. Spending as much time as they can on Facebook usually isn't one of those things. So, to those concious of the time they're wasting, they feel guilty for continuing to use it. This leads to some treating quitting social media like quitting a bad addiction (because it can be). It's important to note that the best recommendation algorithm that keeps users engaged for as long as possible (per session) isn't the one that always serves the best content. It's the one that starves the user - giving them of satisfaction over time. It makes the user feel that they're "on a hunt" for satisfaction and that if they scroll long enough, eventually, they will get "the big one" and can log off feeling entirely satisfied. This strategy for engagement is not just from recommendation algorithms, but platforms generally makking the content users want harder to find - like hiding posts or making them non-chronological from friends or subscriptions. In doing this, platforms completely eliminate nearly all phycological benefit from the app that the user could have gotten. The best-case scenario for keeping a user happy and productive is to serve them exactly what they want right from the start, the best of the best content. They leave the app as soon as possible, feeling satisfied. Now, that would be great and I'm sure designers have realized that. However, it comes at sacrifice. This approach leads to less engagement in the short term. Many entertainment and social media apps are barely making a profit already and the space is very competitive. Getting users to leave one's app when they're satisfied with what they got from it doesn't always mean that that user is then going to go do something else entirely. Likely, they're going to go to another app. That app may not follow the same practice and instead go for the more invasive approach of the user but that means more engagement for a competitor, and less engagement for the more "friendly one" - until the user decides that the more invasive one is not worth visiting and drops it entirely. This motivates companies looking purely at engagement for being more invasive because their engagement graphs, that likely do not accurately predict the future, show that it's more effective. This leads me to think that this is an issue of an ad-based profit motive. When a business gets their money through advertisements, it leads to a seemingly clear relationship of time engaged to money made. For a purely ad supported platform, it's true - in the short term, until users start to quit. The time-based incentive needs to be dropped which means no advertisements within the app unless the company can restrain themselves from eeking out the tiny profits they get from it by tweaking their recommendation algorithm to take more time from the user. I think this could be done by developing a profit structure centered around rewarding the platform for gaining the user's respect. Doing that is harder to do, harder to maintain. But, I think it would help to solve, or at least improve the relationship that social media platforms have with their users in respect to respecting their time.

Respecting the user's money

Recommendation algorithms for shopping platforms run into similar issues as ones for entertainment and social media. Their main objective is to get the user to spend as much money as possible. This is a little less excusable than wasting the user's time (though, time is money - usually) but, it comes with the same defense that the user should be able to limit themselves. It also has the same tactic - make things just hard enough to find so that when they do find the thing they want, they are willing to pay more as it makes it come across as more of a or product due to it being on the second page of Amazon. As a price-concious consumer, my goal is to spend as little money as possible for the best product and I'm generally looking for one specific thing. This means that my goals and that of recommendation algorithm are more or less, not aligned. But are my goals misaligned with the platform itself? Generally not, the platform should want my purchase no matter how much I spend (as long as it's not zero) - they make more than enough margin for server upkeep. Amazon seems to think that no matter how annoying they make their shopping process, that I'll continue to purchase from them - I won't. Their lack of proper filters, easy to find product information, putting the reviews not at the bottom of a never-ending page isn't because they can't do it. They want the user to spend more time, spend more money per session. Another issue arises when one looks at the topics of addictions. What people see when they first go to a store influences their decisions and what they're going to buy. Say someone has been buying Oreos and icecream every time they've purchased from Amazon in the past 6 months. Understandably, the recommendation algorithm would begin to recommend them Oreos and icecream as well as other related treats. Recently, this person has noticed that they're quickly gaining weight and decide it's time to quit Oreos and icecream. Having used Amazon nearly exclusively for a while, they continue to use it when they go to purchase other, more healthy foods instead. However, the first thing they see when they open it is Oreos and icecream, putting it back in their mind. They stick through, they serach for what they want. Alongside of their search, they see Oreos and icream. They continue and put what they want in their cart. They go to their cart, it says, "purchase again: Oreos and icecram" with an simple, easy-to-press button saying "add to cart" next to it. The person decides that maybe a little won't hurt to "ease the transition". Just like that, the person's previous will has been broken. Hiding recomendations, at least specific ones, would be a fair, and simple way to help alleviate this problem (Amazon states they have this option, but it is not where they say it is and I can't find it - it was either removed, moved, or never existed). It would take more will from the person to toggle this option, but it could stop users from having to quit using a service that has their recommendation algorithm "stuck" on products they don't want to see.

Exacerbating biases

Recommendation algorithms start out "feeling out" the user - recommending them different things that they may or may or may not enjoy. This is so that it can eventually give them personalized recommendations. This is usually fine when looking at broad topics like gaming, music, makeup, vlogs, horror movies, etc. However, a lot of recommendation algorithms are picking from a pool of content that include propoganda and generally highly opinionated topics. This stuff is usually not recommended right away, but when it is, most recommendation algorithms latch on very hard when someone engages with it because it's very likely that someone who engages with, and seemingly enjoys a specific side of a devicive topic, will want to see more of it. It pulls people into a sort of rabbit hole where everything they see on the platform seems to agree with them and ideas. The more similar content they consume, the more the algorithm holds onto it, giving it "sticking power". What people consume isn't just a reflection of the person, it also influences and changes them. The more and more extreme the content gets, the more the person loses sight of the other side. Especially because it's not unlikely that social media is where they intereact with the most people, and if it seems, that from their perspective, that most people agree with them and that everyone else, must be wrong or off-base. This doesn't make me think that platforms should not host or that recommendation algorithms should not host content with "devisive" topics or even extremist topics (as long as it follows other guidelines and is not illegal). Seeing a variety of perspectives and opinions is what is important to further discussions and bring people closer together. I do believe something has to be done about the segmentation and disunity that platforms with heavy use of recommendation algorithms seem to brew. I'm unsure of whether this should be the jobs of the platforms themselves. Something that may be able to be done that could help is for algorithms to detect when something is devisive in nature and try to recommend content alongside of it with differeing opinions. That solution however, I find makes people angry - like when Twitter and YouTube put warnings on what they determined to be misinformation about Covid19. I've been thinking for a little bit about a project of my own to help solve this problem but there are too many details to include on here - maybe on another blog post (or product launch w).

Determining who has a job, and who doesn't

Recommendation algorithms play a huge part into where user engagement is targeted towards. For platforms whose content is provided by multiple people, who are trying to make a profit or even just a living on the content they upload, a new ethics question is brought into play. If people make their livlihoods off of content that is filtered by these algorithms, that makes the algorithm almost the ultimate decider of whether that person's career on the platforms succeeds or fails. It puts a lot of pressure on those creators who may put hours, days, weeks, months, into making something and having its fate not even decided by a curator, boss, or even the people watching it. For algorithms that are not open source for people to analyze how to optimize content for it (most algorithms to "prevent abuse" - I don't believe that point, I think it's to hide certain aspects of it that may or may not be beneficial for everyone, but may make some uncomfortable - it's also because recommendation algorithms are one of a platform's most valuable assets), fear of it changing is also highly prevelant. Creators spend significant amounts of time trying to optimize for a theoretical algorithm that they think is how the one the platform their basing their livlihood's depend on. The stakes here also create some concern for algorithms that are made using machine learning as those can gain unintended biases. Most concerning of these biases include race, sexuality, region, and political slant of the creators and advantages/disadvantges brought by the algorithm for those in specific groups that are not inherently indicitive to the quality of the content they produce but may be shown in statistical analyses due to other, unseen factors (which is what most ML algorithms do).

Are they necessary?

All this concern over what recommendation algorithms make and the design considerations when creating them may lead some to want to give up on them entirely. It's a fair thought, before recommendation algorithms, we just asked friends about the latest movie to watch or what newspaper to read. The issue arises when one looks at just how much content is being produced nowadays. One person cannot have near enough time or knowledge to be able to filter through all the new content being made to recommend the best content for someone. Ok, that's kind of a pointless example, so why don't we hire a bunch of people to look through it? There's multiple issues there. First being that those people don't know the person on the other end - it can't be personalized. There are also other ethical questions when having real people go through user-uploaded content. Not everyone can be vetted and sometimes, people post very illegal content - one example being ISIS, who tends to upload some not-very-family-friendly content - as one could imagine. Ok, make an algorithm to filter that out (not getting into that ethical discussion) and then have "recommenders" tag the content for what kind of person would like to watch it and maybe users could identify themselves or take a test and then they'd get the content tagged for them. At this point, it's back at recommendation algorithm with nearly all the same problems - except now, it's calculated by workers probably being paid next to nothing. Hopefully, this has illustrated that recommendation algorithms are a necessary that we have to deal with. But, I don't think they have to be evil. They just need some - or a lot - more improvement.