AB-731, AI, AI Strategy, Generative AI, Microsoft Certification

Identify the challenges of using Generative AI solutions, including fabrications, reliability, and bias (AB-731 Exam Prep)

 
This post is a part of the AB-731: AI Transformation Leader Exam Prep Hub.
This topic falls under these sections:
Identify the business value of generative AI solutions (35–40%)
   --> Identify the foundational concepts of generative AI
      --> Identify the challenges of using Generative AI solutions, including fabrications, reliability, and bias

Note that there are 10 practice questions (with answers) at the end of each section to help you solidify your knowledge of the material. Also, there are 4 practice tests with 30 questions each available from the hub's main page below the exam topics section.

Introduction

Generative AI offers tremendous opportunities for organizations, including improved productivity, enhanced customer experiences, and accelerated innovation. However, AI Transformation Leaders must recognize that generative AI also introduces challenges and risks.

Unlike traditional software systems that follow predefined rules, generative AI produces probabilistic outputs. This means responses may vary and are not always completely accurate. Organizations must therefore implement governance, oversight, and responsible AI practices to ensure that AI systems are trustworthy and aligned with business objectives.

For the AB-731 certification exam, understanding the limitations and risks of generative AI is just as important as understanding its capabilities.

Why Generative AI Has Limitations

Generative AI models do not “understand” information in the same way humans do.

Instead, they:

  • Learn patterns from training data.
  • Predict likely outputs.
  • Generate responses based on probabilities.

Because they rely on patterns rather than true understanding, AI systems can sometimes:

  • Produce incorrect information.
  • Generate inconsistent responses.
  • Reflect biases found in training data.
  • Omit important context.
  • Produce misleading outputs.

These limitations highlight the need for human oversight and responsible AI practices.

Fabrications (Hallucinations)

One of the most widely discussed challenges of generative AI is the possibility of fabrications, often called hallucinations.

A fabrication occurs when an AI model generates information that:

  • Appears convincing,
  • Sounds credible,
  • But is incorrect, misleading, or entirely invented.

Examples

The AI may:

  • Cite nonexistent sources.
  • Invent statistics.
  • Generate incorrect facts.
  • Create fictional events.
  • Provide inaccurate references.

Example Scenario

An employee asks AI:

“Provide sources supporting these market statistics.”

The AI produces references that look legitimate, but some of the sources do not actually exist.

Why Fabrications Occur

Generative AI predicts likely sequences of text rather than verifying facts.

The model may prioritize producing a fluent response over ensuring factual accuracy.

Factors that can increase hallucinations include:

  • Ambiguous prompts
  • Missing context
  • Questions outside the model’s knowledge
  • Lack of supporting data
  • Complex or highly specialized topics

Reducing Fabrications

Organizations can reduce hallucinations by:

Providing Better Prompts

Specific prompts generally produce better results.

Using Retrieval-Augmented Generation (RAG)

RAG retrieves trusted organizational data before generating responses.

Incorporating Human Review

Employees should validate important outputs.

Using Reliable Data Sources

Current and authoritative information improves response quality.

Restricting High-Risk Use Cases

Critical decisions should not rely solely on AI-generated outputs.

Reliability Challenges

Reliability refers to the consistency and dependability of AI outputs.

Generative AI systems are probabilistic rather than deterministic.

This means identical prompts may produce different responses.

Examples of Reliability Issues

Inconsistent Answers

Two users asking the same question may receive slightly different responses.

Variable Quality

Some outputs may be excellent while others may require significant editing.

Missing Context

The model may misunderstand user intent.

Outdated Information

A model’s training data may not reflect recent events or changes.

Why Reliability Matters

Organizations need predictable systems for:

  • Compliance
  • Legal requirements
  • Financial reporting
  • Healthcare decisions
  • Customer communications

Low reliability can reduce:

  • User trust
  • Adoption
  • Business value

Improving Reliability

Organizations can improve reliability through:

Prompt Engineering

Well-structured prompts often produce better responses.

Human Oversight

Humans should review important outputs.

Testing and Evaluation

AI systems should be tested before deployment.

Grounding with Enterprise Data

Using RAG improves consistency by supplying current information.

Continuous Monitoring

Organizations should monitor performance after deployment.

Bias in Generative AI

Bias occurs when AI outputs unfairly favor or disadvantage certain individuals, groups, or perspectives.

Bias may appear in:

  • Recommendations
  • Language
  • Images
  • Hiring suggestions
  • Customer interactions

Sources of Bias

Training Data Bias

Models learn from large datasets that may contain historical biases.

Representation Bias

Certain populations may be underrepresented in training data.

Cultural Bias

Models may reflect assumptions from specific regions or cultures.

Human Bias

Bias can also be introduced during model development or evaluation.

Examples of Bias

An AI system might:

  • Use stereotypes.
  • Produce unbalanced recommendations.
  • Generate culturally insensitive content.
  • Favor certain demographic groups.

These outcomes may create:

  • Ethical concerns
  • Reputational risks
  • Legal risks
  • Compliance challenges

Fairness and Responsible AI

Organizations should strive to ensure that AI systems are fair and inclusive.

Responsible AI practices include:

  • Evaluating outputs for bias.
  • Testing with diverse scenarios.
  • Monitoring system behavior.
  • Incorporating human review.
  • Maintaining accountability.

Microsoft’s Responsible AI principles emphasize:

  • Fairness
  • Reliability and safety
  • Privacy and security
  • Inclusiveness
  • Transparency
  • Accountability

Privacy and Data Protection Risks

Generative AI systems may process sensitive information.

Examples include:

  • Customer data
  • Financial records
  • Intellectual property
  • Employee information

Improper use could result in:

  • Data leakage
  • Privacy violations
  • Regulatory noncompliance

Mitigation Strategies

Organizations should implement:

  • Access controls
  • Data governance policies
  • Encryption
  • Security monitoring
  • Compliance procedures

Security Risks

AI systems can introduce new attack surfaces.

Potential risks include:

Prompt Injection Attacks

Malicious instructions attempt to manipulate model behavior.

Unauthorized Access

Sensitive information could be exposed.

Data Exfiltration

Attackers may attempt to retrieve confidential information.

Abuse and Misuse

Users may intentionally exploit AI systems.

Organizations should establish strong security controls and governance processes.

Lack of Explainability

Generative AI models are often considered “black boxes.”

It can be difficult to explain:

  • Why a response was generated,
  • How conclusions were reached,
  • Which data influenced the output.

This lack of transparency may present challenges in highly regulated industries.

Dependency and Overreliance

Employees may begin trusting AI outputs without verification.

Overreliance can lead to:

  • Errors being overlooked,
  • Reduced critical thinking,
  • Poor decision-making.

AI should support human judgment rather than replace it.

Intellectual Property and Copyright Considerations

Organizations should consider:

  • Ownership of generated content,
  • Copyright implications,
  • Licensing restrictions,
  • Protection of proprietary information.

Legal and compliance teams may need to establish policies governing AI-generated content.

Ethical Considerations

AI systems can affect:

  • Customers
  • Employees
  • Society
  • Organizational reputation

Responsible use requires organizations to consider:

  • Fairness
  • Transparency
  • Accountability
  • Human impact

AI Transformation Leaders should ensure that ethical considerations are incorporated into AI strategies.

The Role of Human Oversight

Human oversight remains essential because AI:

  • Can make mistakes.
  • Can generate fabricated information.
  • Can produce biased results.
  • Cannot replace business accountability.

Humans should:

  • Review outputs.
  • Validate critical information.
  • Make final decisions.
  • Monitor system performance.

Generative AI is most effective when it augments human expertise rather than replacing it.

Common Risk Mitigation Strategies

Organizations can reduce AI risks through:

Governance Frameworks

Define policies and responsibilities.

Responsible AI Principles

Promote fairness and accountability.

Human-in-the-Loop Processes

Maintain human review.

Testing and Monitoring

Evaluate performance continuously.

Data Quality Improvements

Provide accurate and trusted information.

Employee Training

Teach users how to use AI responsibly.

Business Perspective

AI leaders should balance:

Opportunities

  • Productivity gains
  • Innovation
  • Customer experience improvements

with

Risks

  • Fabrications
  • Bias
  • Reliability concerns
  • Security threats
  • Compliance requirements

Successful AI transformation involves maximizing benefits while managing risks responsibly.

Exam Tips

For the AB-731 exam, remember:

  • Fabrications (hallucinations) occur when AI generates incorrect information that appears credible.
  • Reliability refers to consistency and dependability of outputs.
  • Bias can originate from training data and development processes.
  • Human oversight remains essential.
  • RAG can improve accuracy and reduce hallucinations.
  • Responsible AI principles help organizations mitigate risks.
  • AI systems should augment human decision-making rather than replace accountability.
  • Governance, monitoring, and testing are critical components of successful AI adoption.

Practice Exam Questions

Question 1

An AI assistant generates references to research papers that do not actually exist. Which challenge does this represent?

A. Bias
B. Security breach
C. Fabrication (hallucination)
D. Model compression

Answer: C

Explanation: Fabrications occur when AI generates plausible but incorrect or invented information, such as nonexistent citations.

Question 2

Why do generative AI systems sometimes produce inaccurate information?

A. They rely on probabilistic predictions rather than true understanding.
B. They only use structured databases.
C. They execute predefined business rules.
D. They require no training data.

Answer: A

Explanation: Generative AI predicts likely outputs based on learned patterns rather than verifying facts like a human expert.

Question 3

Which technique can help reduce hallucinations by supplying current organizational information?

A. Increasing response length
B. Retrieval-Augmented Generation (RAG)
C. Eliminating governance controls
D. Disabling monitoring

Answer: B

Explanation: RAG retrieves trusted information and provides it to the model, improving accuracy and reducing fabricated responses.

Question 4

What does reliability refer to in generative AI?

A. The amount of storage required by the model
B. The size of the training dataset
C. The speed of network connectivity
D. The consistency and dependability of outputs

Answer: D

Explanation: Reliability focuses on whether AI outputs are consistent, predictable, and trustworthy.

Question 5

Which factor is a common source of bias in AI systems?

A. Excessive hardware memory
B. Training data containing historical biases
C. Strong password policies
D. Network latency

Answer: B

Explanation: Models learn patterns from training data, and any biases present in that data may be reflected in AI outputs.

Question 6

Why is human oversight important when using generative AI?

A. Humans are required to train every model from scratch.
B. AI systems cannot generate text independently.
C. Humans must validate important outputs and make final decisions.
D. Human oversight eliminates all security risks.

Answer: C

Explanation: Humans remain accountable for reviewing AI outputs and ensuring their correctness and appropriateness.

Question 7

Which Microsoft Responsible AI principle is most directly concerned with minimizing unfair outcomes?

A. Fairness
B. Scalability
C. Profitability
D. Automation

Answer: A

Explanation: The fairness principle focuses on ensuring that AI systems treat people equitably and avoid discriminatory outcomes.

Question 8

Employees begin accepting AI-generated answers without reviewing them. What challenge does this represent?

A. Data compression
B. Prompt injection
C. Overreliance on AI
D. Fine-tuning failure

Answer: C

Explanation: Overreliance occurs when users trust AI outputs without applying human judgment or validation.

Question 9

Which risk involves malicious attempts to manipulate AI instructions?

A. Representation bias
B. Prompt injection attacks
C. Token optimization
D. Data normalization

Answer: B

Explanation: Prompt injection attacks attempt to influence or override intended AI behavior through malicious inputs.

Question 10

What is one of the primary goals of responsible AI governance?

A. Eliminate all operational costs
B. Replace human decision-making entirely
C. Prevent the need for monitoring
D. Maximize benefits while managing risks

Answer: D

Explanation: Responsible AI governance seeks to balance business value with ethical, security, reliability, and compliance considerations.

Go to the AB-731 Exam Prep Hub main page

Leave a comment