Guide-yakabatana mibvunzo · Pakati Nhanho
Bias in AI Grading Quiz
Tests scoring validity, subgroup consistency, rubric alignment, human review, disagreement analysis, evidence limits, and student correction.
Mubvunzo 1 ye 8
A scoring modhi inobvumirana zvakanyanya nevadzidzisi zvachose. Chii chinosimbisa pamusoro peboka rega rega revadzidzi?
Tarisa mubvunzo wese8 mibvunzo
Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.
- A scoring modhi inobvumirana zvakanyanya nevadzidzisi zvachose. Chii chinosimbisa pamusoro peboka rega rega revadzidzi?
- That all groups receive identical scores
- That the model is fair for every writing task
- That teachers can stop reviewing scores
- It does not by itself establish equal performance or error rates across subgroups
- Chivakashure chemudzidzi wemitauro yakawanda chinosiyana nemienzaniso mukudzidziswa. Chii chinofanira kuongorora vaongorori?
- Whether the rubric measures the intended learning goal or overweights language form
- Whether the student should be excluded from scoring
- Whether the model can rewrite the essay
- Whether the assignment used a computer
- What does a subgroup score difference mean by itself?
- Proof that the model is discriminatory
- A signal to investigate the assessment and context, not a complete finding of bias or fairness
- Proof that the model is fair
- Proof that the student used AI
- Sei uchiongorora zviitiko apo AI uye matekisheni evadzidzisi asingabvumirani?
- They should be deleted as outliers
- They prove the teacher is wrong
- They can reveal rubric, prompt, or model errors hidden by an average
- They identify the correct grade automatically
- Ndeipi kuenzanisa inotsigira zvakanyanya kuongororwa kwekururama?
- Compare different prompts with different students
- Use only the AI score distribution
- Check only the model’s training loss
- Compare model and trained human scores on the same prompt and rubric
- Chii chinogona kuwirirana kwepamusoro pakati peAI nechiyero chevanhu chichiri kuvanza?
- Systematic score offsets or different error patterns
- That the scores are random
- That the rubric was not used
- That no student has a writing sample
- What should a student have if a score is consequential?
- Only a model confidence score
- A way to ask questions and request review or correction
- A second model that cannot be appealed
- A rule that the first score is final
- A fairness study evaluates a scoring model on one essay corpus. What can a school conclude?
- Every AI grader has the identical bias
- The tool is approved for all grades
- The result informs that setting, but the school still needs to evaluate its own tool and learners
- Hapana kumwe kuongorora kunodiwa