雅思阅读 52: The Algorithm's Gaze（算法的凝视）

改编由 ISACA / Amnesty International / Software Freedom Law Center（2025-2026）。雅思阅读 Section 3 难度，约 1050 词。 素材来源：https://www.isaca.org/resources/news-and-trends/isaca-now-blog/2025/facial-recognition-and-privacy-concerns-and-solutions-in-the-age-of-ai

Reading Passage

A. When a camera on a London high street identifies your face, checks it against a database of millions, and alerts a police officer within half a second, you have not been stopped by a person. You have been stopped by a probabilistic match made by software that has never seen you before. That technology, once confined to the laboratories of defence contractors, has now spread to airports, stadiums, apartment buildings and school entrances in more than sixty countries. It is marketed as a convenience — face unlock on a phone, a "familiar faces" feature on a doorbell camera, a fast-track lane at an airport — but the underlying infrastructure is indistinguishable from mass surveillance. In 2025, Amnesty International documented that Argentina's Ministry of Security spent at least 1.2 million dollars on social-media monitoring, facial-recognition software and thermal-camera drones within a single year, with no parliamentary oversight of how the resulting data were used. The same report noted that Hungary had used facial recognition to monitor activists at demonstrations, raising questions about the technology's compatibility with European free-expression law.

B. The technical problem is not that facial recognition is inaccurate in general. A 2024 evaluation by the US National Institute of Standards and Technology, testing 189 algorithms, found that the best systems correctly identified a white male face with vanishingly few false alarms. The problem is that the error rate is not evenly distributed. For women, and in particular for women and men with darker skin, the false-positive rate was between ten and one hundred times higher than for white males, depending on the algorithm. When such systems are deployed in a city centre, they therefore tend to misidentify the same small group of people over and over — and those people are disproportionately poor, disproportionately young, and disproportionately drawn from ethnic minorities. The UN Human Rights Office has repeatedly warned that automated decision systems, unless audited, will reproduce the biases of the data on which they were trained. Because most training photographs of well-lit, front-facing faces show young white men, the algorithms are simply better at recognising people like the ones they were shown.

C. The consequences have moved from abstract bias to concrete harm. By 2025, at least twelve wrongful arrests in the United States had been traced to misidentification by facial-recognition algorithms, and in each documented case the person arrested was a person of colour. The database behind many of these identifications belongs to Clearview AI, a company that scraped more than fifty billion facial images from social media, news photographs and public websites without the consent of the people pictured, and then sold access to law-enforcement agencies. A single false positive means hours in a police cell, a criminal record, and a court case that may take years to clear. Civil-liberties groups argue that treating a probabilistic match as probable cause is a fundamental violation of the presumption of innocence, because the algorithm cannot be cross-examined and its internal weights cannot be inspected in open court. Several of those wrongfully arrested have only been exonerated after months or years of legal proceedings, and many have lost jobs, housing or custody of their children in the interim.

D. The commercial rollout is proceeding in parallel. Amazon's Ring doorbell company launched a "Familiar Faces" feature in 2026 that identifies individuals across residential cameras and uploads the resulting biometric templates to a cloud database. Illinois, Texas and the city of Portland blocked the feature under existing biometric-privacy laws, but forty-seven other US states have no equivalent legislation. The result is that the same doorbell camera, installed by the same household, may be legal in one suburb and illegal in the next. In India, airports rolled out a "DigiYatra" face-recognition boarding system marketed as a time-saving convenience, while the country's software-freedom movement warned that the same infrastructure doubles as a national identity database. In China, new regulations governing the security of facial-recognition technology took effect in June 2025, prohibiting the installation of such cameras in hotel rooms and bedrooms and requiring opt-in consent in most public contexts — a rare example of a government drawing a line between convenience and surveillance. In the absence of federal rules, the result is a patchwork: the same feature may be legal in one state, illegal in the next, and invisible to the consumer who never consented to have their face stored in the first place.

E. The debate is no longer whether the technology can be built, but whether it can be governed. Three approaches are now on the table. The first is prohibition: some cities, including San Francisco and Boston, have banned the use of facial recognition by local police. The second is audit: requiring vendors to publish error rates by demographic group and to disclose the size and origin of their training sets. Without such disclosure, a city council cannot know whether the system it is buying will misidentify a Black teenager in the same proportions that the 2024 NIST evaluation predicted. The third is participation: giving individuals the right to know when their face is in a database, to request its deletion, and to see the algorithm's reasoning when it has contributed to a decision about them. None of these approaches has yet been tested at national scale. The risk, as Amnesty International put it in its 2026 report on Argentina, is that governments will build the infrastructure of a surveillance state before the legal safeguards are in place — and that by the time the safeguards arrive, the data already collected will be impossible to recall. Once a face has been scraped, matched and stored, it can be used in ways that no one, including the person it belongs to, can foresee. The choice now, all three reports argue, is not between surveillance and no surveillance, but between surveillance with democratic safeguards and surveillance without them.

Questions 1-4

Choose the correct heading for paragraphs B, C, D and E from the list of headings below.

List of Headings i. Uneven accuracy across demographic groups ii. From algorithmic bias to wrongful arrests iii. The parallel commercial rollout iv. Three approaches to governance v. A brief history of camera manufacture vi. How doorbells are sold online vii. The cost of airport security

1. Paragraph B: ____
2. Paragraph C: ____
3. Paragraph D: ____
4. Paragraph E: ____

Questions 5-8

Choose the correct letter, A, B, C or D.

5. What did the 2024 NIST evaluation find? A. All 189 algorithms had equal error rates. B. False positives were 10-100 times higher for people with darker skin than for white males. C. No algorithms worked at all. D. The evaluation tested only European faces.

6. By 2025, how many confirmed wrongful arrests in the US were traced to facial recognition? A. Zero. B. At least twelve. C. About two. D. Over a thousand.

7. What is unusual about the Chinese regulations that took effect in June 2025? A. They banned all cameras. B. They prohibit facial recognition in hotel rooms and require opt-in consent in public. C. They require cameras in every school. D. They apply only to foreign tourists.

8. Which of the following is NOT one of the three governance approaches described? A. Prohibition by cities. B. Algorithmic audits. C. Individual participation rights. D. Mandatory installation in all homes.

Questions 9-13

Do the following statements agree with the claims of the writer?

Write:

• TRUE if the statement agrees with the information
• FALSE if the statement contradicts the information
• NOT GIVEN if there is no information on this

9. Amnesty International documented that Argentina spent at least $1.2 million on surveillance technology in 2024-2025.
10. The NIST evaluation found that false-positive rates were equal across all demographic groups.
11. Clearview AI built its database by asking people for permission before scraping their photos.
12. Illinois, Texas and Portland blocked Amazon's Ring "Familiar Faces" feature.
13. The European Union has proposed a complete ban on all facial-recognition technology.

Questions 14-15

Complete the summary below using NO MORE THAN TWO WORDS from the passage.

NIST found that (14) __________ positives were 10-100 times higher for people with darker skin, and civil-liberties groups argue that treating a probabilistic match as probable cause violates the presumption of (15) __________.

答案与解析

← 上一篇 | 返回雅思焦点 | 下一篇 →