In §7.4, I argue that the "don't feed the trolls" rule shifts responsibility to the targets. On Discord, I help keep a balance, not focusing solely on doxing threats that have escalated; what finally worked was friction and compliance: maIembe mode during spikes, cooldowns for new members, removal of prohibitive thresholds, and robust reporting. I also interpret §7.3 as the ethical distinction between the “increase” of protest trolling (e.g., flooding K-pop apps) and the “reduction” of cruelty (RIP). It would be helpful if the chapter explicitly identified conformity and power dynamics as key dividing lines between ridicule, resistance, and contempt.

[g32] argues that ignoring rarely stops harassment; abusers escalate to force a reaction. The piece advocates platform-level moderation and removals, shifting responsibility from victims to systems. This directly reinforces §7.4’s critique of “don’t feed the trolls” and grounds the chapter’s recommendation in reported cases rather than abstract principle.

Let’s answer this question: Performance itself does not compromise authenticity. As defined in this chapter, a connection is authentic when the service provided matches the actual service. If the audience can identify with a carefully crafted persona and feel the intimacy promised, the connection is authentic. If important facts are hidden (for example, through undisclosed sponsorships or deceptive practices), it becomes fake. In my research on the Bluesky bot, labeling posts as coming from a bot maintained trust and correlated the connection provided with reality.

[f10] Regarding the “unidentified professor”: This report reveals how community skepticism (strange memorial service phone calls, no Arizona State University records) confirmed that @Sciencing_Bi was a McLaughlin scam and how the scam caused multiple harms: identity theft, community manipulation, and the discrediting of a genuine Title IX whistleblower. The name reference and institutional confirmation are powerful sources. Combined with BuzzFeed [f9] and The Verge [f13], this strengthens the timeline and mitigates the bias of a single source.

Reading [e33], the inventor apologizes for infinite scroll because it removes stopping points and lengthens use. I propose an ethical design rule: build in “natural stops”—page ends, session timers, and a one-tap “take a break” card—enabled by default and easy to keep on. Platforms should have to justify any removal of these stops.

I support adding "friendly friction" to the UI. The pop-up before retweeting makes me hesitant. I also suggest offering adjustable friction levels (read timers, cooldowns for late-night posts). This would reduce impulsive spreading while still maintaining freedom of choice. Do you agree?

[d28] Hoffman, "Data Violence" (2018): Harm arises not only from biased outcomes but also from categorization/interface choices that force people to "adapt." Key takeaway: If the affected community doesn't shape the architecture, technical fairness improvements can still perpetuate violence. This reframes the formal failure in §4.3: a flawed drop-down menu isn't a failure; it's a governance failure.

While reading §4.2 and the debate over Twitter bots, I was struck by how definitions can quietly gain influence. For a class project, I modified a spam heuristic (number of URLs + account age), and my bot estimate went from 3% to 14%—the same data, just simplified. This convinced me that the "<5%" number wasn't a fact, but a governance decision about what's considered criticaI. Platforms should publish ranges based on uncertain scenarios and disclose the underlying assumptions. Question: Why don't we require confidence intervals and other definitions for platform metrics, as we do in epidemioIogy?

I disagree with completely excluding "click farms" from the "bot" discussion. The key question for platform ecosystems isn't whether they're computerized, but whether interactions can be manipulated at superhuman scale, at low cost, and through scripted automation. If the goal of governance is to maintain the information environment, why not include a dimension of "functional automation" in the definition and include manual, batch operations within the framework of disclosure and restrictions? Otherwise, the same manipulation effects will be overlooked simply because the implementation method has changed.

Sarah pointed out that Tay's out-of-control behavior can be prevented by design: content fiItering, frequency limiting, adversariaI testing, and narrowing learning objectives should be implemented. It is recommended that boundary-based Iearning be adopted by default (no online self-learning without manual review) and that "modeI cards" be made public to explain data sources and security boundaries, transforming vigilance into boundary-based governance.