Learn all CIE question paper questions.

UNIT-1
1. Computer science, philosophical and other views of AI
2. Differentiate between Top-down and Bottom-up approach towards AI Development
3. Illustrate the characteristics of AI Agency with real world examples. (Autonomy,
Interaction, Adaptability).
4. BDI MODEL
5. Explain with a figure agent and multiagent systems
Situatedness and encapsulation
Reactive and proactive
Flexibility
6. Consider a company ABC is designing a Delivery drone identify the levels of autonomy
that can be assigned to the drone with an example scenario.
7. ML DL AND REINFORCEMENT LEARNING
8. Analyze the usage of HABA-MABA framework for any similar scenario Customer
Support Chatbot
9. Human Computer Interaction : Virtual Agents and Embodied system
10. Explain the Responsible Research and Innovation process
11. The ART of AI: Accountability, Responsibility, Transparency
12. Design for values
10. Analyze how to design a Responsible AI System integrating Design for values and
Domain Requirement of software engineering.
11. Design a responsible development life cycle for AI Systems
12. Ensuring Responsible AI in Practice

UNIT-2
1. Four key opportunities AI offers for Society :
2. What are the dual advantage of an ethical approach to AI
3. Discuss the unified framework of principles of AI for society

4. Seven essentials for achieving trustworthy AI

5) Organization ABC is planning to develop an AI-based Sentiment Analysis System.

Identify the potential ethical risks that could arise in the process, and how can the
organization apply and analyse ethical principles to mitigate these risks effectively?

To ensure that Organization ABC translates ethical principles into ethical

practices effectively while developing their AI-based Sentiment Analysis
System, they need to be aware of and address several potential risks. Here’s
an analysis of these risks:

1. Ethics Shopping:

• Risk: Selectively adopting ethical guidelines that justify their current

practices instead of genuinely aligning with comprehensive ethical standards.
• Application and Analysis: ABC should ensure they are not just cherry-
picking guidelines that are convenient. For example, they might emphasize
data privacy but ignore fairness and bias. To avoid this, they should adopt a
holistic set of ethical guidelines that covers all relevant aspects, such as the
"Ethics Guidelines for Trustworthy AI," and regularly review their
compliance with these guidelines.

2. Ethics Bluewashing:

• Risk: Making superficial claims about their commitment to ethical AI to

enhance their reputation without meaningful implementation.
• Application and Analysis: ABC should back up their ethical claims with
concrete actions. For instance, if they claim their system is unbiased, they
should provide evidence of thorough bias testing and mitigation measures.
Transparency in reporting their ethical practices can help in building genuine
trust.
 3. Ethics Lobbying:

• Risk: Influencing policymakers to create favorable regulations that benefit

the company rather than the public good.
• Application and Analysis: ABC should engage in ethical advocacy by
supporting robust, fair, and comprehensive regulations. Instead of lobbying
for lenient rules, they should advocate for standards that ensure AI systems
are safe, transparent, and fair, even if this means stricter oversight.

4. Ethics Dumping:

• Risk: Deploying their AI system in regions with weaker regulatory

frameworks to avoid strict ethical scrutiny.
• Application and Analysis: ABC should maintain consistent ethical standards
across all regions where they operate, regardless of local regulations. For
example, if they launch their sentiment analysis tool in a country with lenient
AI regulations, they should still adhere to high ethical standards to ensure the
system’s fairness and safety.

5. Ethics Shirking:

• Risk: Avoiding responsibility for ethical issues by blaming third-party

developers or users.
• Application and Analysis: ABC should take full responsibility for any ethical
issues that arise. They should have robust internal processes to address biases
and errors within their AI system, rather than deflecting blame to external
factors. Clear accountability structures should be established to ensure ethical
compliance.

Conclusion:

To effectively manage these risks, Organization ABC should:

• Adopt a comprehensive set of ethical guidelines.

• Ensure transparency and accountability in their ethical practices.
• Support robust and fair regulations.
• Maintain high ethical standards across all regions.
• Take full responsibility for ethical issues.

6) Six ethical issues raised by algorithm

Epistemic Concerns: These issues are related to knowledge and data reliability:
i) Inconclusive Evidence: Algorithms may generate results that lack certainty.
ii) Inscrutable Evidence: Algorithms can be opaque or difficult to interpret.
iii) Misguided Evidence: They may produce misleading or biased conclusions.
Normative Concerns: These are more directly related to ethical values:
Unfair Outcomes: Algorithms may result in biased or unjust outputs.
Transformative Effects: Algorithmic decisions can have substantial societal
impacts.
 Traceability: It is often challenging to trace back the decision-making process,
making it hard to assign responsibility for outcomes, particularly in cases of
(semi-) autonomous algorithms like ML.

7) Four prominent Algorithmic fairness definitions in literature

8) Participatory Design and Social Contract Models

9) Seven Essential Factors for Successful AI4SG, you can get questions on any topics from
below:
• Falsifiability and Incremental Deployment
• Safeguards Against Manipulation of Predictors
• Receiver-Contextualized Intervention
• Receiver-Contextualized Explanation and Transparent Purposes
• Privacy Protection and Data Subject Consent
• Situational Fairness
• Human-Friendly Semanticisation

UNIT-3

1. Typology
2. Challenges faced by the financial sector when adopting AI systems.
3. Fairness in the Financial Services Industry
4. Explain the below topics
1. Definition of Scope
2. Risk Identification and Assessment
3. Risk Management Plan and Control Design
4. Defined Roles and Responsibility
5. What are the requirements outlined by the FCA regarding the definition and management
of algorithmic trading, and what steps must firms take to comply with these
requirements.
5. Imagine an e-commerce company using an AI model to recommend products to users.
How can the company ensure that its AI model achieves demographic parity, ensuring
that the recommendation rate is the same across different demographic groups, such as
age and income level.

6. Steps in Risk Identification and Assessment

7. Fairness Through Unawareness explain
1. Demographic Parity
2. Counterfactual Fairness
3. Equalized Odds/Equalized Opportunity
8. Counterfactual fairness
Counterfactual fairness is an important concept in machine learning and fairness.
This model suggests that if we have a causal model (U, V, F) consisting of:
A set of observable variables (V),
A set of latent background variables (U) not caused by V,
A set of functions (F),
Then the counterfactual of belonging to a protected class is independent of the
outcome. Essentially, the protected attribute (A) should not influence the
actual outcome (Y) or the predicted outcome

9. Individual fairness

10. Equalized Odds/Equalized Opportunity

11. Imagine an organization developing a complex software system. How can it establish an
effective development and testing process, ensure strong governance and oversight,
manage documented changes efficiently, and maintain transparency and explainability
throughout the project lifecycle.

• Development and Testing Process

• Governance and Oversight
• Documented Change Management, Testing and Approval
• Transparency and Explainabilty

12. Explain the below topics

• Ensuring Solution is safe to scale

• Review the feedback mechanism
• Kill Switch and Business Continuity

13. Human AI Relationship – you might get a question, if you have learned all other topics
you can skip this bcoz of choice in each unit.

UNIT- 4

1) Six accounts of the wrongfulness of discrimination

2) Fairness Criteria in Algorithmic Audits
3) Any five Traditional Tests for discrimination
4) Taste based and statistical discrimination
5) Outcome-based tests
6) Fair considerations in applications of natural language processing
7) Demographic disparities and questionable applications of computer vision
8) Search and Recommendation system
9) Ad targeting
10) Information flow, fairness, privacy / Adversarial Test

UNIT-5
1. Need for Interpretability
2. Key Traits Enhanced by Model Interpretability
3. When Interpretability is Not Needed
4. Taxonomy of Interpretability Methods in Machine Learning
5. What are the different types of results or outputs that can be obtained from various
interpretation methods used in machine learning models.
6. Three main levels for evaluating interpretability
7. Properties of explanations
8. Properties of Individual explanations
9. What is Good Explanation / Human friendly explanation
10. Consider an e-commerce recommendation system. How does explainable machine
learning transform real-world raw customer data (such as purchase history, browsing
behavior, and customer feedback) into interpretable explanations that help marketing
analysts make informed decisions about suggesting relevant products to customers?
Explainable AI diagram and explanation - ans
11. Local Model Agnostic Methods – learn all with the equations
12. Global Model-Agnostic Methods learn all with the equations