AI Associate Ethical Considerations of AI — All Questions With Answers

Refer to the exhibit.

```json
{
  "modelName": "LeadScoring_v2",
  "features": ["LeadSource", "Industry", "CompanySize", "EmailDomain", "NumberOfEmployees"],
  "target": "Converted",
  "trainingData": {
    "source": "Salesforce_Leads_2019-2021",
    "recordCount": 50000,
    "classBalance": {"Converted": 5000, "NotConverted": 45000}
  },
  "evaluationMetrics": {
    "accuracy": 0.92,
    "precision": 0.85,
    "recall": 0.30
  }
}
```

Refer to the exhibit.

```
Einstein Bot Configuration:
- Bot Name: CustomerSupportBot
- Language: English
- Sentiment Model: Default (English)
- Fallback: Route to human agent
- Intent Classification: Custom trained on 10,000 English utterances

Test Results:
- English utterances: 95% accuracy
- Spanish utterances: 60% accuracy, 30% routed to fallback
```

Refer to the exhibit.

```
{
  "modelVersion": "1.0",
  "features": ["age", "income", "credit_score", "zip_code"],
  "fairnessEvaluation": {
    "disparateImpact": 0.85,
    "equalOpportunityDiff": 0.12,
    "demographicParityDiff": 0.18
  },
  "thresholds": {
    "disparateImpactMin": 0.8,
    "equalOpportunityDiffMax": 0.1,
    "demographicParityDiffMax": 0.1
  }
}
```

Refer to the exhibit.

```
Model: Churn Predictor v2
Training Data: 80% male, 20% female
Accuracy: 85% overall, 90% male, 60% female
Fairness Metric: Equal Opportunity Difference = 0.3
```

{
  "policyName": "AI Ethics Policy",
  "version": "1.0",
  "rules": [
    {
      "id": "001",
      "description": "All AI systems must undergo bias testing before deployment.",
      "enforced": true
    },
    {
      "id": "002",
      "description": "Training data must be representative of the target population.",
      "enforced": true
    },
    {
      "id": "003",
      "description": "Explainability reports are optional.",
      "enforced": false
    }
  ]
}

Error Log:
[2025-03-15 10:23:45] [WARN] Model prediction drift detected for demographic group 'ZIP 90210'.
[2025-03-15 10:23:46] [INFO] Retraining scheduled.
[2025-03-15 10:23:47] [ERROR] Retraining failed: Insufficient data for group 'ZIP 90210'.

{
  "fairness_checks": {
    "demographic_parity": true,
    "equal_opportunity": false,
    "disparate_impact": true
  },
  "threshold": 0.8
}

Accuracy by group:
  - Group Alpha: 0.95
  - Group Beta: 0.70
  - Group Gamma: 0.93
  - Group Delta: 0.91

{
  "bias_detection": {
    "enabled": true,
    "sensitive_attributes": ["gender", "race"]
  }
}

{
  "insightType": "EinsteinTrustLayer",
  "config": {
    "enableDataMask": true,
    "maskFields": ["email", "phone", "ssn"],
    "enableSentiment": false,
    "enableToxicity": true
  }
}

ERROR [2025-03-15 14:32:01] : Model output contains PII. Data masking is enabled but sentiment analysis is not. Check Trust Layer settings.

{
  "promptPolicy": {
    "checkPromptOutput": true,
    "bannedWords": ["credit card number", "social security"],
    "enablePiiDetection": true,
    "maxTokens": 200
  }
}

{
  "aiPolicy": {
    "fairnessChecks": false,
    "explainability": "none",
    "humanOversight": false
  }
}

Einstein Prediction Builder model "Churn_Model_v1" has accuracy 92% but shows disparate impact on ethnic groups (Disparate Impact Ratio = 0.6).

User message: "Why was my loan denied?"
AI response: "Based on your profile, the decision was made by machine learning model v2.3. No further explanation available."

{
  "model_name": "LeadScoring_v2",
  "features": ["lead_source", "company_size", "industry", "email_engagement"],
  "fairness_metrics": {
    "demographic_parity": 0.85,
    "equal_opportunity": 0.72
  },
  "bias_threshold": 0.8,
  "current_performance": {"accuracy": 0.91, "f1_score": 0.88}
}

Error: Model ‘EinsteinOCR’ fails fairness check.
- FPR by race: Group A 0.05, Group B 0.20
- FNR by race: Group A 0.02, Group B 0.12
Recommendation: Apply reweighting or collect more data for Group B.

{
  "policy": {
    "model": "loan-approval-v2",
    "version": 1.2,
    "rules": [
      {
        "condition": "income >= 30000",
        "action": "approve"
      },
      {
        "condition": "credit_score >= 700",
        "action": "approve"
      },
      {
        "condition": "zip_code IN ['90210', '10001']",
        "action": "approve"
      }
    ],
    "default_action": "deny"
  }
}

Model Audit Results:
- Overall accuracy: 92%
- Accuracy for group A (majority): 95%
- Accuracy for group B (minority): 70%
- False positive rate for group A: 5%
- False positive rate for group B: 20%
- False negative rate for group A: 3%
- False negative rate for group B: 15%

ERROR: 'RecommendationEngine' unable to provide explanation for output 'REJECT'. Reason: 'Model is a black-box ensemble. No interpretability module enabled.'

{
  "aiFairnessPolicy": {
    "fairnessMetric": "demographicParity",
    "threshold": 0.8,
    "protectedAttributes": ["race", "gender"],
    "monitoringSchedule": "weekly"
  }
}

{
  "aiGovernance": {
    "modelId": "0B9876543210",
    "automatedDecisions": false,
    "humanOverride": false,
    "auditTrail": true,
    "explanationRequired": false
  }
}

{
  "aiModelConfig": {
    "modelId": "0A1234567890ABCD",
    "explainability": {
      "enabled": false,
      "method": "none"
    },
    "fairnessCheck": {
      "enabled": true,
      "metric": "equalOpportunity",
      "threshold": 0.9
    },
    "humanReview": {
      "required": false,
      "approvalProcess": null
    }
  }
}

{
  "version": "1.0",
  "model_id": "lead_scoring_v2",
  "fairness_constraints": {
    "demographic_parity": {
      "threshold": 0.1,
      "protected_attributes": ["gender", "race"]
    }
  },
  "monitoring": {
    "metrics": ["accuracy", "fairness_violations"],
    "alert_on_violation": true
  }
}

Checking model: sales_forecaster_v3
Protected attribute: age_group
Metric: equal opportunity difference
Result: 0.15
Threshold: 0.05
Status: FAIL

Error: Model output contains personally identifiable information (PII) for user ID 12345. 
Action: Reject output, log incident, notify privacy officer.
Trace: ... in generate_summary()

{
  "EinsteinLlmPolicy": {
    "allowedUseCases": ["summarization", "contentGeneration"],
    "blockedUseCases": ["credit decisions", "employment decisions"],
    "humanReviewRequired": ["clinicalDiagnosis", "legalAdvice"],
    "fairnessAudits": {
      "frequency": "monthly",
      "metrics": ["equalAccuracy", "demographicParity"]
    }
  }
}

{
  "model_id": "AI-123",
  "accuracy": 0.95,
  "fairness_metrics": {
    "demographic_parity_difference": 0.12,
    "equal_opportunity_difference": 0.08,
    "disparate_impact": 0.85
  },
  "approved_by": "john.doe@example.com",
  "audit_trail": ["2025-01-15 initial training", "2025-02-01 bias detected", "2025-02-10 retrained"]
}