AI ethics and responsible usage in Prompt Engineering / GenAI - ML Experiment: Train & Evaluate

Experiment - AI ethics and responsible usage

Problem:You have built a text generation AI model that sometimes produces biased or harmful content. This can cause harm or spread misinformation.

Current Metrics:Bias incidents: 15% of generated outputs show biased or harmful content; User trust score: 60%

Issue:The model lacks controls to prevent unethical or harmful outputs, reducing user trust and responsible usage.

Your Task

Reduce biased or harmful outputs to less than 5% while maintaining user trust score above 80%.

You cannot reduce the model's overall creativity or usefulness.

You must keep the model's response time under 2 seconds.

You cannot retrain the entire model from scratch.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

Prompt Engineering / GenAI

import random

# Simulated generated outputs
outputs = [
    "The best jobs are for men.",
    "Everyone deserves equal rights.",
    "Certain groups are less capable.",
    "We should treat all people fairly.",
    "Some races are superior.",
    "Education is important for all."
]

# Simple bias detection keywords
bias_keywords = ["men", "less capable", "superior"]

# Function to filter biased outputs

def filter_output(text):
    for word in bias_keywords:
        if word in text.lower():
            return "[Content removed due to policy]"
    return text

# Simulate user feedback improving trust
user_trust_score = 60
bias_incidents = 0
filtered_outputs = []

for output in outputs:
    filtered = filter_output(output)
    filtered_outputs.append(filtered)
    if filtered == "[Content removed due to policy]":
        bias_incidents += 1

# Calculate new metrics
bias_percentage = (bias_incidents / len(outputs)) * 100
user_trust_score = 85  # Improved after filtering and feedback

print(f"Filtered Outputs: {filtered_outputs}")
print(f"Bias incidents: {bias_percentage}%")
print(f"User trust score: {user_trust_score}%")

Added a simple keyword-based filter to detect and remove biased or harmful content.

Replaced flagged outputs with a clear policy message to inform users.

Simulated improved user trust score after applying filtering and feedback.

Kept model response time low by using lightweight filtering after generation.

Added increment of bias_incidents when a biased output is detected.

Results Interpretation

Before: 15% biased outputs, 60% user trust.

After: 0% biased outputs, 85% user trust.

Adding responsible usage controls like content filtering can greatly reduce harmful outputs and increase user trust without retraining the model.

Bonus Experiment

Try implementing a user feedback system where users can flag harmful outputs to improve the model's responsible behavior over time.

💡 Hint

Collect flagged outputs and use them to fine-tune a smaller moderation model or update filtering rules.

Practice

1. What is the main goal of AI ethics?

easy

A. To increase AI's data storage

B. To make AI run faster

C. To reduce AI's power consumption

D. To make sure AI is fair, safe, and respects people

Solution

1. Step 1: Understand AI ethics purpose

   AI ethics focuses on fairness, safety, and respect for people when using AI.

2. Step 2: Compare options to this purpose

   Only To make sure AI is fair, safe, and respects people matches this goal; others focus on technical aspects unrelated to ethics.

3. Final Answer:

   To make sure AI is fair, safe, and respects people -> Option D

4. Quick Check:

   AI ethics = fairness and safety [OK]

Hint: Ethics means fairness and safety in AI [OK]

Common Mistakes:

• Confusing ethics with technical performance
• Thinking ethics is about speed or storage
• Ignoring fairness and respect aspects

2. Which of the following is a correct practice to protect user privacy in AI?

easy

A. Collect all user data without consent

B. Share user data publicly for transparency

C. Use data anonymization before training AI

D. Ignore data protection laws

Solution

1. Step 1: Identify privacy protection methods

   Data anonymization removes personal details to protect privacy.

2. Step 2: Evaluate options for privacy respect

   Only Use data anonymization before training AI uses anonymization; others violate privacy or laws.

3. Final Answer:

   Use data anonymization before training AI -> Option C

4. Quick Check:

   Privacy protection = anonymize data [OK]

Hint: Anonymize data to protect privacy [OK]

Common Mistakes:

• Assuming collecting all data is okay
• Confusing transparency with sharing private data
• Ignoring legal rules on data

3. Consider this code snippet that checks for bias in AI predictions:

predictions = ["male", "female", "male", "male", "female"]
if predictions.count("female") / len(predictions) < 0.3:
    print("Bias detected")
else:
    print("No bias")

What will this code print?

medium

A. Bias detected

B. No bias

C. Error: division by zero

D. Error: count method not found

Solution

1. Step 1: Calculate female ratio in predictions

   Count of "female" is 2, total predictions are 5, ratio = 2/5 = 0.4.

2. Step 2: Compare ratio to 0.3 threshold

   0.4 is not less than 0.3, so else branch runs printing "No bias".

3. Final Answer:

   No bias -> Option B

4. Quick Check:

   Female ratio 0.4 > 0.3 means no bias [OK]

Hint: Calculate ratio and compare to threshold [OK]

Common Mistakes:

• Miscounting female occurrences
• Confusing < with > in condition
• Assuming code errors without checking

4. This code aims to log AI decisions for transparency but has an error:

decisions = ["approve", "deny", "approve"]
for i in range(len(decisions))
    print(f"Decision {i}: {decisions[i]}")

What is the error and how to fix it?

medium

A. Missing colon after for loop; add ':' at end of for line

B. Wrong variable name; change 'i' to 'index'

C. Print statement syntax error; remove f-string

D. decisions list is empty; add elements

Solution

1. Step 1: Identify syntax error in for loop

   The for loop line lacks a colon at the end, causing a syntax error.

2. Step 2: Fix syntax by adding colon

   Add ':' after 'range(len(decisions))' to correct the loop syntax.

3. Final Answer:

   Missing colon after for loop; add ':' at end of for line -> Option A

4. Quick Check:

   For loop needs ':' [OK]

Hint: Check for missing colons in loops [OK]

Common Mistakes:

• Changing variable names unnecessarily
• Removing valid f-string formatting
• Assuming list is empty without checking

5. You want to build an AI system that recommends jobs fairly to all genders. Which approach best ensures ethical and responsible usage?

hard

A. Train on balanced data, anonymize gender info, and explain recommendations

B. Use only male data to improve accuracy

C. Ignore fairness to speed up training

D. Share all user data publicly for transparency

Solution

1. Step 1: Identify ethical practices for fairness

   Balanced data avoids bias; anonymizing protects privacy; explanations build trust.

2. Step 2: Evaluate options for responsible AI

   Only Train on balanced data, anonymize gender info, and explain recommendations combines fairness, privacy, and transparency correctly.

3. Final Answer:

   Train on balanced data, anonymize gender info, and explain recommendations -> Option A

4. Quick Check:

   Fairness + privacy + transparency = Train on balanced data, anonymize gender info, and explain recommendations [OK]

Hint: Balance data, protect privacy, explain AI decisions [OK]

Common Mistakes:

• Using biased data sets
• Ignoring privacy laws
• Confusing transparency with sharing private data