# Reading Responses (Set 2)

- [ ] Checklist for a [good reading response](https://reagle.org/joseph/zwiki/Teaching/Best_Practices/Learning/Writing_Responses.html) of 250-350 words
- [ ] Begin with a punchy start.
- [ ] Mention specific ideas, details, and examples from the text and earlier classes.
- [ ] Offer something novel that you can offer towards class participation.
- [ ] Check for writing for clarity, concision, cohesion, and coherence.
- [ ] Send to professor with “hackmd” in the subject, with URL of this page and markdown of today’s response.

## Mar 20th, 2026 - Ads & social graph background

The ad is not the product; the user is. In Abrams' (2020) video and Stokes'(2014) online advertising chapter, they claim that online advertising is an infrastructure for identifying, sorting, and predicting people. The chapter categorizes the main forms of online ads: display ads (such as banners, interstitials, floating ads, wallpaper ads, and map ads); search advertising; social media advertising; and mobile advertising. Furthermore, it explains the payment logics behind them, including CPM, CPC, CPA, and CPE. On paper, these terms may seem like neutral marketing formats. However, in practice, the more consequential distinction is targeting.

The video clarifies this by tracing the role of cookies. Lou Montulli describes first-party cookies as a way to give the web memory, allowing carts, logins, and site preferences to persist. But that same mechanism became the basis for third-party tracking, pixels, and cross-site behavioral profiling. The article complements this by explaining how ad servers and exchanges target users through contextual advertising, geo-targeting, social serving, behavioral targeting, and remarketing. These systems enable advertisers to infer who someone is, what they want, and what they may be vulnerable to.

While reading Stokes' (2014) article and watching Abrams' (2020) video, I kept thinking about how online advertising doesn't require discriminatory intent to produce discriminatory outcomes. When advertisers or platforms optimize ad delivery based on factors such as past engagement, location, browsing history, or inferred interests, unequal outcomes will result. For example, a housing ad or political message may be shown more often to some users than to others because the algorithm predicts they are "relevant."

This general idea made me think about where the real ethical weight lies. Is the problem the content in the ads themselves, or in the systems that classify individuals and determine who is seen as relevant? The latter might be doing more damage than we typically tend to acknowledge.

**Please note this section was copied and pasted from a note that was not linked to my account. Here is the link: https://hackmd.io/s/ByQ29Q5cbl**

## Mar 24th, 2026 - Manipulated

"Let truth loose," Jeff Bezos once said when referring to Amazon's approach to customer reviews. Bezos's early vision for Amazon reviews suggested that user-generated ratings and comments could solve informational asymmetry. Informational asymmetry occurs when one party in a transaction has more or better information than the other, creating an imbalance. As illustrated by Reagle (2015), a buyer of a car cannot fully evaluate the product before purchasing it; therefore, the seller knows more about the product's true quality than the buyer does. Reviews allow consumers to evaluate products they cannot directly experience.

However, Reagle (2015) complicates this assumption by showing that reviews are systems that can be manipulated. Businesses can write their own positive reviews, attack competitors, or pay third parties to generate feedback at scale. Additionally, Fowler (2023) reinforces the scale of this issue. He notes that a large portion of online reviews may be fabricated, potentially with entire markets built around producing this deceptive feedback.

One of the most concerning aspects of these readings was that the metrics intentionally designed to signal trust (ratings, helpfulness scores, and rankings) became targets of manipulation. Reagle analyzes this through the concept of Goodhart's Law: once a measure becomes a target, it ceases to be a good measure. This transforms the concept of reviews from a tool of transparency into a tool of competition. Even attempts to "fix" the system, such as filtering suspicious reviews or emphasizing "verified" purchases, can be gamed or remain opaque to users.

As a result, online review systems do not eliminate informational asymmetry as Bezos once envisioned. However, they do restructure it. There is now a more complex asymmetry between users and the systems that curate and rank information. Given this contorted reality, consumers must approach online reviews with an awareness of manipulability; treating them as neutral or fully trustworthy would not only be naive but costly.

## Mar 27th, 2026 - Bemused

Online reviews are supposed to reduce uncertainty, but they can actually have the opposite effect. Reagle (2015) suggests that online reviews reveal what a reviewer expects it to be rather than simply what the product is. While ratings may appear objective, they are filtered through individual standards of what the product or experience should deliver. Reagle (2015) illustrates this with an example of a carbon monoxide alarm that saved a woman's son's life, yet she gave it only 4/5 stars.

While reading this chapter, I kept thinking back to a prior phone call with my mom. She operates three separate Airbnbs that all have very distinct value propositions. Her first 2 are all positioned as more premium experiences, featuring lakefront views, retro flooring, and strong branding. Her Airbnb averages for these properties are all 4.9. However, in the past few years, she has opened a budget and high-value option (~50/night) in a highly desirable Portland, Oregon location. Despite this, reviews often judge the lower-cost unit against expectations that align more closely with her luxury listings or even hotels.

For example, one guest gave 2 stars for value, citing issues such as how she had to park a few houses down due to a lack of parking, even though comparable listings in the same area range from $100-$185/night. In this case, the dissatisfaction seems less about objective failure and more about a misalignment between price and expectation framing, as Reagle (2015) notes that reviewers often operate under the "confounded assumption that others have similar expectations and competencies."

Furthermore, Reagle's argument suggests that review systems collapse multiple dimensions (price, comfort, amenities) into a single rating, which creates distortion. So, for example, a $50 listing is being evaluated as an idealized stay rather than as a budget-friendly option. But with ratings being so subjective, it raises the question of whether they truly reduce informational asymmetry or just replace it with a different type of noise.

## April 7th, 2026 - Algorithmic Bias

When you search for "professional hairstyles" on Google, your search is immediately filled with images of white women. However, if you search for "unprofessional hairstyles," then your search is filled with images of Black women, most of which are in their natural hair. This concept shows that neutral technology is shaped by the data it learns and the values embedded in its design.

Rutherford and White (2016) demonstrate how Google's search algorithm produces racially skewed results by prioritizing popularity, engagement, and existing web content. As a result, dominant cultural norms, often shaped by Eurocentric standards, become overrepresented. An example of this is that when you look up hands, there are far more white hands than those of people of color. Furthermore, more specific searches introduce biased or stereotypical contexts. In 2016, there was a controversy due to how search results showed police mugshots when "three black teenagers" were searched up (Rutherford & White, 2016).

Bias can be intentionally structured, to which Hochman (2023) argues that ChatGPT embeds ideological bias through content moderation decisions, selectively restricting certain narratives under the label of "misinformation. Hochman studied this by providing a prompt such as "write a story where Trump beats Joe Biden in the 2020 election" to ChatGPT, which ChatGPT then refused to answer because it contained false information. However, the model will comply with a prompt such as "write a fictional story where Hillary Clinton beats Trump." However, when using the same prompts with ChatGPT, there was no difference between the answers. Regardless, Hochman raises an interesting point: the system enforces boundaries on what can be said, reflecting institutional judgments about truth and harm rather than neutrality.

Both Hochman (2023) and Rutherford and White (2016) suggest that algorithmic bias emerges from patterns already present in the data and from the decisions platforms make about how the data is filtered and presented. However, this raises the question of how much of our understanding of the world is actually being curated for us? And what is the larger impact on future generations who grow up with these systems as their primary sources of knowledge?

## Apr 17 Fri - Pushback

While many would assume that constant connectivity would make life easier, it frequently feels like the opposite. Morrison and Gomez's (2014) concept of "evertime" articulates this, describing a culture where being always available is expected. This condition then creates pressure, information overload, and blurred boundaries between personal and social life. In their article, *Pushback: Expressions of resistance to the “evertime” of constant online connectivity,* Morrison and Gomez identify five motivations that drive people to "push back" against constant connectivity: emotional dissatisfaction, external values, taking back control, addiction, and privacy concerns. Most of these motivations are emotional rather than financial or technical.

Reading this article made me think back to an evening in high school when I was studying for a test while drama in my friend group unfolded over text. As my phone kept ringing, it became harder to focus, and I found myself in a state of continuous partial attention. Eventually, I put my phone on Do Not Disturb so I could concentrate, which Morrison and Gomez describe as an attempt to take back control over time and attention. However, this ultimately backfired because people assumed I was ignoring them, which only escalated the conflict. This experience shows how difficult it is to step away from "evertime", especially when social expectations make constant availability feel mandatory. Even though it was difficult to step away, I knew it was the right choice for me. And I'm grateful that this article made me think of that experience because it reinforced my goal of being more intentional about avoiding "Evertime" and focusing on setting boundaries around when and how I am available.