Section 03 · SAT-Style Education
Righteous AI Edition
Academic Skills Through Models of Righteousness
This interactive zone connects The Righteousness Digest‘s article on Project Maven and Gender Shades to academic skills tested on the SAT and in college-level coursework.
How to use this module: Work through the three skill areas in order, or jump to any one. Your answers are saved privately in this browser. You can check your score and export a summary at any time.
Critical Reading — “Two Acts of Conscience”
Skill focus: Identify the central claim, examine supporting evidence, interpret vocabulary in context, and draw inferences from a nonfiction passage. The passage below is adapted from this section’s article. Read it carefully before answering the questions.
Two Acts of Conscience
1In 2018, two very different events changed the direction of artificial intelligence. One was a protest inside a technology company. The other was a research paper published by two scientists.
2In April 2018, more than 3,000 Google employees signed an open letter demanding that their company cancel its contract with the U.S. Department of Defense for a program called Project Maven. The program used machine learning to analyze drone footage. The employees wrote plainly: “We believe that Google should not be in the business of war.” Two months later, Google announced it would not renew the contract.
3In the same year, a researcher named Joy Buolamwini and a scientist named Timnit Gebru published a study called “Gender Shades.” They tested three commercial facial recognition systems using 1,270 faces from three African and three European countries. The results were unmistakable: all three systems performed worst on darker-skinned women. IBM’s system had an error rate of 34.7% on darker-skinned women but only 0.3% on lighter-skinned men.
4Neither the protest nor the paper was required by any law. No regulator demanded that Google cancel its contract. No court ordered IBM and Microsoft to fix their systems. In both cases, people saw something wrong and said so — publicly, with evidence, at personal cost.
5This is what makes them models of righteousness. They did not wait for permission. They did not wait for the law. They acted on conscience — and the industry changed because of it.
Central Ideas and Details
Which choice best states the central claim of the passage?
- AGoogle employees and Joy Buolamwini both worked in the technology industry.
- BActs of individual conscience, rather than regulation, drove two major changes in AI in 2018.
- CFacial recognition systems are inherently biased against darker-skinned women.
- DProject Maven was more significant than the Gender Shades study.
Explanation
Correct answer: B. The central argument appears in Paragraph 4: “Neither the protest nor the paper was required by any law… people saw something wrong and said so.” Option A is a factual detail. Option C is a finding, not the central claim. Option D is a comparison the passage never makes.
Command of Evidence
Which detail from the passage best supports the claim that the Gender Shades study produced measurable, objective findings?
- A“They tested three commercial facial recognition systems using 1,270 faces”
- B“IBM’s system had an error rate of 34.7% on darker-skinned women but only 0.3% on lighter-skinned men”
- C“Joy Buolamwini and a scientist named Timnit Gebru published a study”
- D“No regulator demanded that Google cancel its contract”
Explanation
Correct answer: B. This is the specific numerical evidence — a quantifiable disparity. Option A describes the method, not the finding. Option C identifies the authors. Option D is about Project Maven, not Gender Shades.
Words in Context
As used in Paragraph 5, the word “conscience” most nearly means:
- Aa legal obligation
- Ban inner sense of what is right and wrong
- Ca formal complaint
- Da written agreement
Explanation
Correct answer: B. “Conscience” in this context refers to the internal moral sense that drives people to act without external compulsion. Option A is the opposite — the passage explicitly says neither act was required by law.
Inference
Based on the passage, what can be most reasonably inferred about the relationship between conscience and regulation?
- AConscience is a substitute for regulation.
- BRegulation is more effective than conscience.
- CConscience can drive change even when regulation does not require it.
- DConscience is only effective in large groups.
Explanation
Correct answer: C. The passage states that “neither the protest nor the paper was required by any law” — yet “the industry changed.” Option A overstates the case. Option B contradicts the passage. Option D contradicts the fact that Joy Buolamwini acted with just one collaborator.
Text Structure and Purpose
What is the primary rhetorical purpose of Paragraph 4?
- ATo criticize the technology industry for acting too slowly
- BTo establish that the two events shared a common moral character
- CTo list the legal requirements that were ignored
- DTo compare Project Maven with Gender Shades on technical grounds
Explanation
Correct answer: B. Paragraph 4 explicitly states “In both cases, people saw something wrong and said so.” It links the two events. Option C is the opposite — the passage says there were no legal requirements. Option D is not the purpose of the paragraph.
Data Analysis — “The Numbers Behind Gender Shades”
Skill focus: Interpret quantitative information, draw inferences from data, and compare values across categories. Use the table below to answer the questions that follow.
Table 3.4 — Gender Shades: Error Rates by Skin Tone and Gender
Error rates of three commercial facial recognition systems (IBM, Microsoft, Face++) across four demographic groups. Lower percentages indicate better performance. The highest error rates appear in bold.
| System | Lighter Males | Lighter Females | Darker Males | Darker Females |
|---|---|---|---|---|
| IBM | 0.3% | 7.1% | 12.0% | 34.7% |
| Microsoft | 0.0% | 4.2% | 8.3% | 20.8% |
| Face++ | 0.8% | 5.2% | 13.3% | 34.5% |
Source: Buolamwini & Gebru, 2018, Proceedings of Machine Learning Research 81:1–15
Quantitative Interpretation
For IBM’s system, approximately how many times higher was the error rate on darker-skinned females compared with lighter-skinned males?
- AAbout 10 times
- BAbout 50 times
- CAbout 115 times
- DAbout 200 times
Explanation
Correct answer: C. Calculation: 34.7 ÷ 0.3 ≈ 115.7. The IBM system was over 100 times more likely to misclassify a darker-skinned woman than a lighter-skinned man. Option D is a trap — it overstates the figure.
Comparative Analysis
Which system performed best on darker-skinned females?
- AIBM
- BMicrosoft
- CFace++
- DAll three performed equally
Explanation
Correct answer: B. Microsoft’s error rate on darker-skinned females was 20.8% — lower than IBM’s 34.7% and Face++’s 34.5%. This is a meaningful difference, though still high in absolute terms.
Data Interpretation
What pattern does the table most clearly suggest?
- AError rates decrease as skin tone darkens.
- BAll systems perform equally across all groups.
- CError rates increase as skin tone darkens, and are highest at the intersection of dark skin and female gender.
- DError rates are determined by the system, not the demographic group.
Explanation
Correct answer: C. Every system shows the same pattern: lighter males (lowest error), lighter females, darker males, darker females (highest error). The disparity is intersectional — both skin tone and gender contribute.
Drawing Conclusions
Which conclusion is best supported by the data?
- AFacial recognition systems should never be used in any commercial application.
- BThe disparity was isolated to a single company.
- CThe disparity was systematic across all three commercial systems tested.
- DThe systems were poorly designed.
Explanation
Correct answer: C. All three systems tested showed the same pattern of disparity. Option A is extreme. Option B is contradicted by the data. Option D may be a cause, but is not supported by the table alone.
Vocabulary Builder — “Words of Courage and Fairness”
Skill focus: Build academic vocabulary through context, definition, and usage. Each card below presents a key term from the article. Click any card to reveal its definition, pronunciation, example sentence, synonyms, and antonyms. A fill-in-the-blank quiz follows.
conscience
/ˈkɒn.ʃəns/ · noun
Click to reveal →
An inner sense of what is right and wrong.
complicity
/kəmˈplɪs.ə.ti/ · noun
Click to reveal →
Involvement in wrongdoing, especially by remaining silent.
disparity
/dɪˈspær.ə.ti/ · noun
Click to reveal →
A great difference or inequality.
benchmark
/ˈbentʃ.mɑːk/ · noun
Click to reveal →
A standard used for comparison.
audit
/ˈɔː.dɪt/ · verb
Click to reveal →
To examine carefully and systematically.
accountability
/əˌkaʊn.təˈbɪl.ə.ti/ · noun
Click to reveal →
The condition of being responsible and answerable.
intersectional
/ˌɪn.təˈsek.ʃən.əl/ · adjective
Click to reveal →
Relating to overlapping categories, such as race and gender together.
moratorium
/ˌmɒr.əˈtɔː.ri.əm/ · noun
Click to reveal →
A temporary suspension of an activity.
Vocabulary Check — Fill in the Blank
Type the vocabulary word that best completes each sentence. Spellings are checked exactly.