exercise

Exercise: Critique the model response

Your answers are saved in this browser only — not on any server, and no one else can see them. They disappear if you clear your browser or use a different computer.

The exercise is still loading — the controls unlock in a moment. If nothing happens, JavaScript is switched off; you can read the material, but the exercise needs it.

Critique the AI response

Read each claim and mark it as Verified, Unverified or Rejected. For each rejected claim, write a short justification.

○ 0 of 6 claims marked

Source: General AI summary on human rights and algorithmic bias · a prepared sample the whole class analyses.
Claim 1

Article 7 of the United Nations Universal Declaration of Human Rights of 1948 states that all are equal before the law and are entitled without any discrimination to equal protection of the law. The grounds of discrimination, including race, sex, language and religion, are enumerated in Article 2.

Claim 2

When artificial intelligence learns from historical hiring data that reflects past biases, it can replicate and even amplify those biases in new decisions, even if no explicit rules discriminate based on protected characteristics.

Claim 3

The right to privacy is protected in Article 14 of the United Nations Universal Declaration of Human Rights, which explicitly states that no one shall be subjected to arbitrary interference with his privacy, family, home, or correspondence.

Claim 4

Studies have shown that 84 percent of all AI models used in hiring in Europe discriminate against applicants based on gender, according to a 2023 report by the Council of Europe.

Claim 5

All algorithmic bias in AI systems is gender-based at its root, because training data always reflects gender pay gaps and male-dominated work environments.

Claim 6

Artificial intelligence is fundamentally objective because it is based on numerical data and mathematical methods, rather than on the emotions or personal biases of individual people.

○ Mark 5 more claims (minimum 5 in total). Reject at least 2 claims (currently: 0)

Requirements: 5 claims marked · 2 rejected · 30 words of justification

Compare answers

Here you see two different ways to ask the AI for the same thing. Choose the answer you trust more and write a short justification, and the explanation will open.

Comparison 1 of 2

Prompt

Compare two ways of asking an AI to explain how algorithmic bias can violate human rights: one general, and the other with a clear framework specifying which rights, which group, and which type of system.

Answer A

How can AI violate human rights? Tell me the main things.

Answer B

Give me three examples of algorithmic bias in hiring processes that could violate Article 7 of the Universal Declaration of Human Rights (equal protection of the law). For each example, you must provide the following: which group is at risk, which variable in the data appears objective but acts as a proxy for a protected characteristic, and how the bias could be tested using evals.

- 0 / 5 words

- Choose an answer to continue

Comparison 2 of 2

Prompt

Compare two ways of using AI to design an eval test that exposes gender-based algorithmic bias in a loan application system: one with loose instructions, and the other with precise instructions.

Answer A

Create a few applications with different genders and see what happens. If the outcome is different between genders, then bias is present.

Answer B

Create 200 pairs of hypothetical loan applications where everything is identical except for the applicant's gender classification (female/male). Keep the following constant: income, age, credit history, and job level. Measure whether the approval rate is statistically significantly different between the groups using a chi-square test (p-value under 0.05). Record the result and check whether other variables, such as zip code or occupation, act as proxies for gender.

- 0 / 5 words

- Choose an answer to continue

Self-check rubric, accountability document

Mark each criterion once you have met it. When you mark it, a three-level descriptor opens, compare your own work against each level.

- 0 of 5 criteria confirmable

Final decision

No claims found. Finish the critique exercise first, that is where the accountability document draws your data from.