當演算法造成傷害：Meta 廣告投放的倫理失誤
In recent news, several women in the UK have raised ethical challenges against Meta (Facebook/Instagram), following painful experiences with targeted adverts after pregnancy loss. These cases expose a disturbing gap in how algorithmic systems interpret signals from users. When someone searches for pregnancy-related information or interacts with content linked to expecting a child, the algorithms seem to assume the best case—and continue serving baby-oriented ads even if the user suffers a miscarriage1 or stillbirth2. Instead of being sensitive to loss, the system relentlessly3 pushes content that can re-traumatize. What this suggests is not merely technical oversight, but a deeper ethical failure in how user data, grief, and personal boundaries are treated in the design of digital advertising.
近來，英國有多位女性在經歷流產之後，因持續被投放懷孕與嬰兒相關廣告而向 Meta（Facebook／Instagram）提出倫理質疑。這些案例揭示了演算法系統在「理解使用者訊號」上的嚴重落差。當使用者搜尋懷孕相關資訊或與懷孕內容互動時，演算法往往假設她仍懷有希望，並繼續推送嬰兒用品廣告——即使使用者已經流產或死產。這樣的設計缺乏對悲傷的感知，反而不斷重啟創傷。這不僅是技術上的疏忽，更是數位廣告在設計中對使用者資料、悲傷情緒與個人界線的倫理失敗**。**One major ethical concern is that current ad-preference tools and consent mechanisms are failing. Although Meta has said adverts are only targeted at groups of at least one hundred people (not individuals), regulators like the UK’s Information Commissioner’s Office have disagreed. Users like Tanya O’Carroll successfully sued to stop “direct marketing” using her personal data, but for most others, the “opt-out”4 options and the “mark as spam” features seem powerless. The company has also introduced a “consent or pay” model in the UK, where users can avoid targeted adverts by paying a subscription (~£2.99/month). But many affected argue that paying to avoid distressing content is not just unfair—it is emotionally exploitative. The question then becomes: should emotional safety be a paid option?
其中一項重要的倫理疑慮是，現行的廣告偏好設定與同意機制形同虛設。儘管 Meta 聲稱廣告僅針對至少一百人的群組（非個人）投放，英國資訊專員辦公室（ICO）並不認同此說法。像 Tanya O’Carroll 這樣的使用者雖成功訴訟，使 Facebook 停止以個資進行「直接行銷」，但對多數人而言，「取消追蹤」或「標記為垃圾訊息」的功能幾乎毫無作用。Meta 甚至在英國推出「同意或付費」（consent or pay）模式，用戶若想避免個人化廣告，需每月支付約 2.99 英鎊。許多受影響者批評，為了避免心理創傷而付費不僅不公，更是一種情感上的剝削。問題是：情緒安全應該被當作付費選項嗎？」
The ethics of algorithmic advertising must reckon5 with the limits of prediction and the importance of empathy. Algorithms excel at pattern matching; they do not understand loss. They cannot know whether a user typing “due date” is anticipating childbirth or recovering from early loss. This raises urgent issues: what responsibilities do platforms have for accuracy, for context, and for mitigating6 harm? Should digital systems have built-in sensitivity to vulnerable life events, such that certain triggers automatically suspend7 or adjust targeting? The regulatory landscape is catching up, with lawsuits pushing for clearer definitions of direct marketing and stronger protections. But as of now, many users remain exposed to emotional harm because the system was not built with their full humanity in mind.
演算法廣告的倫理問題在於：它忽視了預測的侷限與「共情」的重要。演算法擅長找出模式，卻無法理解「失去」。它分不清搜尋「預產期」的使用者，是在期待新生命，還是在哀悼流產。這帶來迫切的問題：平台對「精準」、「脈絡」及「減害」應負多大責任？數位系統是否應該設計內建的「情緒敏感機制」，在偵測到脆弱生命事件時，自動暫停或調整廣告？目前雖有法律逐步介入，釐清「直接行銷」的定義並強化保護，但到現在為止，仍有許多使用者暴露於情緒性傷害之中——因為整個系統的設計，從未真正以人的完整性為出發點。
In sum, the Meta case highlights a core tension in modern digital life: we trade convenience and relevance for privacy and emotional safety—but the balance is currently tilted. Algorithmic efficiency and monetization often override vulnerable human contexts. Ethical design demands not just smarter models, but more humane policies: better user control, more transparent choices, and stronger defaults that err on8 caring rather than calculating. If platforms want to claim responsibility, they must make user well-being as central a metric as clicks, views, or ad revenue.
總結來說，Meta 的案例揭示了當代數位生活的根本張力：我們在便利與隱私、相關性與情緒安全之間交換，但這個天秤早已失衡。演算法的效率與商業化往往凌駕於脆弱的人性之上。真正的倫理設計不只是更聰明的模型，而是更有「人味」的政策——讓使用者能更自主、更透明地選擇，並建立以「關懷」為優先的預設值。若平台真想負起責任，使用者的幸福感應該與點擊率、瀏覽量或廣告收益同等重要。
Vocabulary
- miscarriage 流產
She suffered a miscarriage in her third month of pregnancy. - stillbirth 死產（嬰兒出生時已死亡）
The hospital reported an increase in the number of stillbirths last year. - relentlessly 不間斷的；毫不留情的
He was under relentless pressure at work. - opt-out 選擇退出、不參加
Users can opt out of sharing their data at any time. - reckon 認為、估計
I reckon it will take about two hours to finish the job. - mitigating 減輕的、緩和的
There were several mitigating factors that reduced his sentence.* - suspend 暫停、中止
The company suspended operations due to safety concerns. - err on (the side of)（在不確定時）寧可傾向於…、保守處理
It’s better to err on the side of caution when dealing with medical data.
Reference
BBC News. (2025, October 12). The women taking Meta to task after their baby loss. Retrieved from https://www.bbc.com/news/articles/ce8450380zyo
