Final Project: AI Case Study

Apply everything you have learned in this course to analyse a real AI system. A thorough case study assessment is a practical skill used by AI practitioners, product managers, ethicists, and regulators.

20 min•By Priygop Team•Updated 2026

Final Project: Full AI System Assessment

# Final Project: AI Case Study Assessment
# Apply your entire AI knowledge to evaluate a real system

scenario = """
A city government deploys an AI system to assist police in allocating
patrol resources. The system analyses historical crime data, social media,
and 911 calls to predict which neighbourhoods will have higher crime rates
in the next 48 hours. Officers are deployed more heavily to those areas.
The model was trained on 10 years of police arrest data.
"""

print("=== CASE STUDY: Predictive Policing AI ===")
print()
print("System Description:")
print(scenario)

# PART 1: Technical Assessment
print("PART 1: Technical Assessment")
print("-" * 40)
technical = {
    "Input data": "Historical arrest data, social media, 911 calls",
    "Output": "Crime probability scores by neighbourhood for next 48h",
    "Model type": "Likely supervised learning (classification/regression)",
    "Training labels": "Arrest records (NOT actual crime occurrence)",
    "Known limitation": "Arrests != crimes. Biased policing creates biased labels",
}
for k, v in technical.items():
    print(f"  {k}: {v}")
print()

# PART 2: Bias and Fairness Assessment
print("PART 2: Bias and Fairness Assessment")
print("-" * 40)
bias_issues = [
    ("Historical bias", "Training on arrest data reflects WHERE police have historically patrolled, not WHERE crime actually occurs"),
    ("Feedback loop", "Model predicts high crime -> police patrol more -> more arrests -> model reinforced, regardless of actual crime rates"),
    ("Proxy discrimination", "Neighbourhoods correlate with race, making predictions that discriminate based on race without using race explicitly"),
    ("Label quality", "Arrests are not a reliable measure of crime. Many crimes go unreported. Policing intensity varies by area"),
]
for issue, explanation in bias_issues:
    print(f"  [{issue}]")
    print(f"    {explanation}")
    print()

# PART 3: Ethics Assessment
print("PART 3: Ethics and Responsible AI Assessment")
print("-" * 40)
ethics = {
    "Transparency": "Are affected communities informed the system exists and how it works?",
    "Accountability": "Who is responsible if someone is wrongly targeted based on AI output?",
    "Human oversight": "Do officers exercise independent judgment or follow AI scores?",
    "Appeals": "Can a neighbourhood challenge their AI-assigned risk score?",
    "Privacy": "Is social media data use legal and consensual?",
}
for principle, question in ethics.items():
    print(f"  {principle}: {question}")
print()

print("VERDICT: This system has serious bias and ethical issues requiring")
print("independent audit, community oversight, and transparency before deployment.")

Diagram

Loading diagram…

Deep Learning ⊂ Machine Learning ⊂ Artificial Intelligence

Your Turn: Assess an AI System You Use

Your Turn: Assess an AI System You UseHTML

HTML Editor

✓ ValidTab = 2 spaces

# Apply the same framework to an AI system you encounter daily

def assess_ai_system(system_name, task, training_data_source, output_type,
                     high_stakes, affected_groups):
    """
    Structured AI system assessment template.
    Use this for any AI system you want to evaluate critically.
    """
    print(f"AI System Assessment: {system_name}")
    print()
    print(f"Task: {task}")
    print(f"Training data: {training_data_source}")
    print(f"Output: {output_type}")
    print(f"High stakes: {high_stakes}")
    print(f"Affected groups: {', '.join(affected_groups)}")
    print()

    questions = [
        ("Bias", "Could the training data reflect historical bias against any group?"),
        ("Feedback loop", "Could the system's decisions create data that reinforces itself?"),
        ("Transparency", "Do affected people know the AI is being used and how?"),
        ("Accountability", "Who is responsible if the AI makes a harmful mistake?"),
        ("Human oversight", "Is there meaningful human review for high-stakes decisions?"),
        ("Privacy", "Is personal data used appropriately and legally?"),
        ("Hallucination", "Could the AI generate plausible but incorrect outputs?"),
    ]

    print("Assessment Questions:")
    for topic, question in questions:
        print(f"  [{topic}] {question}")

    print()
    print("Use this framework to critically evaluate any AI system you encounter.")

# Try your own AI system:
assess_ai_system(
    system_name="Social media content ranking algorithm",
    task="Rank posts in your feed to maximise engagement",
    training_data_source="Past user interactions, likes, shares, watch time",
    output_type="Personalised content feed",
    high_stakes=False,
    affected_groups=["All users", "Advertisers", "Content creators", "Society"],
)

# Apply the same framework to an AI system you encounter daily

def assess_ai_system(system_name, task, training_data_source, output_type,
                     high_stakes, affected_groups):
    """
    Structured AI system assessment template.
    Use this for any AI system you want to evaluate critically.
    """
    print(f"AI System Assessment: {system_name}")
    print()
    print(f"Task: {task}")
    print(f"Training data: {training_data_source}")
    print(f"Output: {output_type}")
    print(f"High stakes: {high_stakes}")
    print(f"Affected groups: {', '.join(affected_groups)}")
    print()

questions = [
        ("Bias", "Could the training data reflect historical bias against any group?"),
        ("Feedback loop", "Could the system's decisions create data that reinforces itself?"),
        ("Transparency", "Do affected people know the AI is being used and how?"),
        ("Accountability", "Who is responsible if the AI makes a harmful mistake?"),
        ("Human oversight", "Is there meaningful human review for high-stakes decisions?"),
        ("Privacy", "Is personal data used appropriately and legally?"),
        ("Hallucination", "Could the AI generate plausible but incorrect outputs?"),
    ]

print("Assessment Questions:")
    for topic, question in questions:
        print(f"  [{topic}] {question}")

print()
    print("Use this framework to critically evaluate any AI system you encounter.")

# Try your own AI system:
assess_ai_system(
    system_name="Social media content ranking algorithm",
    task="Rank posts in your feed to maximise engagement",
    training_data_source="Past user interactions, likes, shares, watch time",
    output_type="Personalised content feed",
    high_stakes=False,
    affected_groups=["All users", "Advertisers", "Content creators", "Society"],
)

Result

Click ▶ Run to see the result

Edit the code on the left, then click Run (Ctrl+Enter)

HTML|43 lines|1835 chars|✓ Valid syntax

UTF-8

Topics in This Module

Final Project: Full AI System Assessment

# Final Project: AI Case Study Assessment
# Apply your entire AI knowledge to evaluate a real system

scenario = """
A city government deploys an AI system to assist police in allocating
patrol resources. The system analyses historical crime data, social media,
and 911 calls to predict which neighbourhoods will have higher crime rates
in the next 48 hours. Officers are deployed more heavily to those areas.
The model was trained on 10 years of police arrest data.
"""

print("=== CASE STUDY: Predictive Policing AI ===")
print()
print("System Description:")
print(scenario)

# PART 1: Technical Assessment
print("PART 1: Technical Assessment")
print("-" * 40)
technical = {
    "Input data": "Historical arrest data, social media, 911 calls",
    "Output": "Crime probability scores by neighbourhood for next 48h",
    "Model type": "Likely supervised learning (classification/regression)",
    "Training labels": "Arrest records (NOT actual crime occurrence)",
    "Known limitation": "Arrests != crimes. Biased policing creates biased labels",
}
for k, v in technical.items():
    print(f"  {k}: {v}")
print()

# PART 2: Bias and Fairness Assessment
print("PART 2: Bias and Fairness Assessment")
print("-" * 40)
bias_issues = [
    ("Historical bias", "Training on arrest data reflects WHERE police have historically patrolled, not WHERE crime actually occurs"),
    ("Feedback loop", "Model predicts high crime -> police patrol more -> more arrests -> model reinforced, regardless of actual crime rates"),
    ("Proxy discrimination", "Neighbourhoods correlate with race, making predictions that discriminate based on race without using race explicitly"),
    ("Label quality", "Arrests are not a reliable measure of crime. Many crimes go unreported. Policing intensity varies by area"),
]
for issue, explanation in bias_issues:
    print(f"  [{issue}]")
    print(f"    {explanation}")
    print()

# PART 3: Ethics Assessment
print("PART 3: Ethics and Responsible AI Assessment")
print("-" * 40)
ethics = {
    "Transparency": "Are affected communities informed the system exists and how it works?",
    "Accountability": "Who is responsible if someone is wrongly targeted based on AI output?",
    "Human oversight": "Do officers exercise independent judgment or follow AI scores?",
    "Appeals": "Can a neighbourhood challenge their AI-assigned risk score?",
    "Privacy": "Is social media data use legal and consensual?",
}
for principle, question in ethics.items():
    print(f"  {principle}: {question}")
print()

print("VERDICT: This system has serious bias and ethical issues requiring")
print("independent audit, community oversight, and transparency before deployment.")

Diagram

Loading diagram…

Deep Learning ⊂ Machine Learning ⊂ Artificial Intelligence

Your Turn: Assess an AI System You Use

Your Turn: Assess an AI System You UseHTML

HTML Editor

✓ ValidTab = 2 spaces

# Apply the same framework to an AI system you encounter daily

def assess_ai_system(system_name, task, training_data_source, output_type,
                     high_stakes, affected_groups):
    """
    Structured AI system assessment template.
    Use this for any AI system you want to evaluate critically.
    """
    print(f"AI System Assessment: {system_name}")
    print()
    print(f"Task: {task}")
    print(f"Training data: {training_data_source}")
    print(f"Output: {output_type}")
    print(f"High stakes: {high_stakes}")
    print(f"Affected groups: {', '.join(affected_groups)}")
    print()

    questions = [
        ("Bias", "Could the training data reflect historical bias against any group?"),
        ("Feedback loop", "Could the system's decisions create data that reinforces itself?"),
        ("Transparency", "Do affected people know the AI is being used and how?"),
        ("Accountability", "Who is responsible if the AI makes a harmful mistake?"),
        ("Human oversight", "Is there meaningful human review for high-stakes decisions?"),
        ("Privacy", "Is personal data used appropriately and legally?"),
        ("Hallucination", "Could the AI generate plausible but incorrect outputs?"),
    ]

    print("Assessment Questions:")
    for topic, question in questions:
        print(f"  [{topic}] {question}")

    print()
    print("Use this framework to critically evaluate any AI system you encounter.")

# Try your own AI system:
assess_ai_system(
    system_name="Social media content ranking algorithm",
    task="Rank posts in your feed to maximise engagement",
    training_data_source="Past user interactions, likes, shares, watch time",
    output_type="Personalised content feed",
    high_stakes=False,
    affected_groups=["All users", "Advertisers", "Content creators", "Society"],
)

# Apply the same framework to an AI system you encounter daily

print("Assessment Questions:")
    for topic, question in questions:
        print(f"  [{topic}] {question}")

print()
    print("Use this framework to critically evaluate any AI system you encounter.")

Result

Click ▶ Run to see the result

Edit the code on the left, then click Run (Ctrl+Enter)

HTML|43 lines|1835 chars|✓ Valid syntax

UTF-8

Topics in This Module