(2nd Sem) AIML Unit I: Questions & Answers

A. Multiple Choice Questions (with Answers)Copy Anchor Link

1. The history of AI is characterized by cycles of optimism followed by reduced funding known as:

   Answer: b) AI Winters

2. Symbolic AI (GOFAI) is based on the:

   Answer: b) Physical Symbol System Hypothesis

3. Which approach is “Top-Down” where humans program every logical step?

   Answer: c) Symbolic AI

4. The main failure of Symbolic AI is called the:

   Answer: b) Brittle problem

5. Machine Learning shifted the paradigm to:

   Answer: b) Intelligence = Pattern Recognition

6. In traditional ML, humans still play a major role by:

   Answer: b) Feature Engineering

7. Deep Learning performs:

   Answer: b) Automated Feature Extraction

8. The catalyst for the Deep Learning revolution (around 2012) was the convergence of Big Data, GPUs, and:

   Answer: b) Backpropagation

9. CNNs are primarily used for:

   Answer: b) Vision and images

10. Transformers are key architectures for:

    Answer: b) Language models

11. Which is the broadest field?

    Answer: c) Artificial Intelligence

12. Human intervention is typically lowest in:

    Answer: c) DL

13. Data requirement is highest for:

    Answer: c) Deep Learning

14. NLP enables computers to:

    Answer: b) Understand, interpret and generate human language

15. Computer Vision primarily uses:

    Answer: b) CNNs

16. The two main components of an Expert System are:

    Answer: b) Knowledge Base and Inference Engine

17. Speech Recognition focuses on converting:

    Answer: b) Spoken language into text

18. In Healthcare, AI is used for:

    Answer: b) Medical imaging, drug discovery and predictive analytics

19. Algorithmic Trading is an application of AI in:

    Answer: b) Finance

20. Personalized Learning platforms that adapt difficulty in real-time belong to AI in:

    Answer: c) Education

21. Recommendation engines on Netflix/Spotify are classic examples of:

    Answer: b) Collaborative filtering / ML

22. “Garbage In, Garbage Out” refers to the problem of:

    Answer: b) Bias and Fairness

23. Explainable AI is mainly related to the ethical principle of:

    Answer: b) Transparency

24. “Human-in-the-Loop” emphasizes:

    Answer: b) Human oversight and control

25. Under risk-based AI governance, AI used in hospitals, schools or hiring is typically:

    Answer: b) High risk

26. AI-generated content (images/videos) should have a:

    Answer: b) Digital watermark / label

27. Which of the following is an example of Deep Learning?

    Answer: c) AlphaGo

28. Roomba is primarily an example of:

    Answer: b) Autonomous robot / intelligent agent

29. The paradigm of Deep Learning is:

    Answer: c) Intelligence = Representation Learning

30. Which milestone is associated with 1997?

    Answer: b) Deep Blue victory

---

B. Theory Questions with AnswersCopy Anchor Link

1. Define Artificial Intelligence. Give any three examples.

Answer:

Artificial Intelligence is the branch of computer science that focuses on creating systems capable of performing tasks that normally require human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. It mimics human intelligence and is a broad field covering many techniques.

Examples:

• Virtual Assistants – Siri
• Chatbots – Customer support bots
• Expert Systems – MYCIN (medical diagnosis)
• Game Playing Systems – IBM Deep Blue
• Autonomous Robots – Roomba

2. Differentiate between AI, Machine Learning and Deep Learning on any four parameters.

Answer:

3. What is Symbolic AI? Why is it called “brittle”?

Answer:

Symbolic AI (also called Good Old-Fashioned AI or GOFAI), dominant from the 1950s–1980s, is based on the Physical Symbol System Hypothesis. It uses high-level symbols (words, numbers) and explicit rules in a top-down approach where humans program every logical step. Key technology: Expert Systems with a Knowledge Base of if-then rules and an Inference Engine. Paradigm: Intelligence = Logic.

It is called “brittle” because it cannot handle uncertainty, fuzzy data, or common-sense reasoning. If a situation falls outside its programmed rules, the system fails or crashes.

4. Explain the shift from Symbolic AI to Machine Learning.

Answer:

Symbolic AI relied on human-coded logic and explicit rules (top-down). As computational power grew and the internet generated large amounts of data, the field shifted to a bottom-up approach. In Machine Learning (1990s–2010s), machines use statistical algorithms to find patterns in data instead of being given rules. Humans still perform feature engineering (e.g., telling the system to look for wheels and windows to identify a car). Key algorithms: Linear Regression, Random Forests, SVMs. Paradigm changed to Intelligence = Pattern Recognition, minimizing an error function through training on labeled data.

5. What is Feature Engineering? Why is it less critical in Deep Learning?

Answer:

Feature Engineering is the process in classical Machine Learning where humans manually extract and select relevant features from raw data (e.g., identifying wheels, windows, headlights for car detection).

It is less critical in Deep Learning because DL performs Automated Feature Extraction. Multi-layer neural networks learn hierarchical representations directly from raw data (e.g., pixels), deciding which features are important without human intervention.

6. List the key catalysts that enabled the Deep Learning revolution around 2012.

Answer:

• Big Data availability
• Massive Parallel Processing via GPUs
• Algorithmic breakthroughs, especially Backpropagation

7. Name the five major branches of AI and give one key task for each.

Answer:

1. Natural Language Processing (NLP) – Machine translation (Google Translate), sentiment analysis, chatbots.
2. Computer Vision – Object detection, facial recognition, medical image analysis.
3. Robotics – Industrial automation, autonomous drones, surgical robots, Mars rovers.
4. Expert Systems – Medical diagnosis, credit scoring, chemical analysis.
5. Speech Recognition – Voice-to-text, voice-controlled interfaces (Alexa, Siri).

8. What are the two main components of an Expert System? Explain their roles.

Answer:

1. Knowledge Base: A repository of facts and if-then rules provided by human experts.
2. Inference Engine: The logical component that applies the rules to known facts to deduce new information or reach conclusions.

9. List any four applications of AI in Healthcare.

Answer:

• Medical Imaging: Analyzing X-rays, MRIs, CT scans to detect tumors or fractures.
• Drug Discovery: Simulating molecular interactions to reduce time and cost of new drugs.
• Predictive Analytics: Monitoring patient vitals to predict crises such as sepsis or cardiac arrest.
• (Also: Personalized treatment recommendations, administrative automation.)

10. What is meant by “AI Ethics”? List any five key principles.

Answer:

AI Ethics refers to the moral principles and guidelines that govern the design, development, and use of artificial intelligence systems.

Key Principles:

• Transparency – Decisions should be explainable.
• Accountability – Developers/organizations are responsible for outcomes.
• Fairness – No discrimination against individuals or groups.
• Privacy – Protection of user data.
• Safety & Security – Systems should be reliable and secure.
• Human Control – Humans should remain in control.

11. Explain the concept of Bias in AI with one example of Data Bias and one of Historical Bias.

Answer:

Bias occurs when an AI system produces results that are systematically prejudiced against certain groups (based on race, gender, age, etc.). It arises because AI learns from past data (“Garbage In, Garbage Out”).

• Data Bias: A medical AI trained only on data from men may perform poorly for women.
• Historical Bias: A hiring AI trained on past successful managers (mostly men) may incorrectly learn that “being a man” is a job requirement.

12. What is AI Governance? State its main objectives.

Answer:

AI Governance is the set of rules and laws that tell companies and governments how they are allowed to build and use AI, ensuring it remains a helper rather than a threat.

Objectives:

• Ensure ethical use of AI
• Protect human rights and privacy
• Promote fairness and non-discrimination
• Maintain transparency and accountability
• Reduce risks and misuse of AI systems

13. Briefly describe the three risk categories used in modern AI regulation.

Answer:

• Unacceptable Risk (Banned): AI that tracks behaviour for social scoring or manipulates minds.
• High Risk (Strictly Regulated): AI used in hospitals, schools, or hiring – must have human oversight and an “off switch.”
• Limited Risk: AI like chatbots – must be transparent (e.g., disclose “I am an AI”).

14. What is the difference between NLP and Speech Recognition?

Answer:

NLP is the broader branch that enables computers to understand, interpret, and generate human language (both text and speech). It includes machine translation, sentiment analysis, summarization, and chatbots.

Speech Recognition is more specific – it focuses only on converting spoken language into text, involving acoustic modelling (phonemes) and language modelling, despite accents and noise.

15. Explain the paradigm shift: Logic → Pattern Recognition → Representation Learning.

Answer:

• Symbolic AI: Intelligence = Logic (human-coded rules).
• Machine Learning: Intelligence = Pattern Recognition (statistical algorithms find patterns in data).
• Deep Learning: Intelligence = Representation Learning (neural networks automatically create multi-layered abstract mathematical representations of the world).

---

C. Analytical / Application Questions with AnswersCopy Anchor Link

1. A hospital wants to deploy an AI system for detecting tumors in X-rays.

Answer:

a) Computer Vision (primarily using CNNs / Deep Learning).

b) Deep Learning model is recommended. Classical Expert Systems are brittle and cannot handle the variability and complexity of medical images well. DL performs automated feature extraction and achieves high accuracy on large image datasets.

c) Ethical concerns:

• Bias (system may perform poorly on underrepresented demographic groups if training data is skewed).
• Accountability (who is responsible if the AI misses a tumor?).
• Privacy (patient medical images must be protected).

2. A company is building a hiring tool that ranks candidates based on past successful employees’ data.

Answer:

a) Historical Bias (and possibly gender/racial bias if past data reflects past discrimination).

b) Steps to improve fairness:

• Use diverse and balanced training data.
• Regularly audit the model for disparate impact and apply fairness constraints. c) High Risk category. Requirements: human oversight, explainability, testing for bias, and an ability to intervene/override.

3. Classify the following systems:

Answer:

• MYCIN → Symbolic AI (rule-based expert system with if-then knowledge base).
• Netflix recommendation engine → Classical Machine Learning (collaborative filtering / pattern recognition from user data).
• Tesla Autopilot perception → Deep Learning (CNNs and multi-layer networks for vision and sensor fusion).
• ELIZA → Symbolic AI (early rule-based chatbot using pattern matching and scripted responses).

4. “Deep Learning has largely solved the brittle problem of Symbolic AI, but it has introduced new problems of data hunger, opacity and bias.” Critically analyze.

Answer:

Symbolic AI failed on anything outside its explicit rules (brittle). Deep Learning largely overcomes this by learning hierarchical features automatically from data and generalizing better to complex, noisy real-world inputs (e.g., AlphaGo, image recognition, self-driving perception).

However, new problems arose:

• Data hunger: Requires very large labeled datasets.
• Opacity: Multi-layer networks are “black boxes,” making decisions hard to explain (challenge for Explainable AI).
• Bias: Inherits and amplifies biases present in training data (Garbage In, Garbage Out). Thus, while brittleness is reduced, new ethical and practical challenges demand careful governance, diverse data, and human oversight.

5. Design a simple conceptual pipeline for a voice-controlled customer-support chatbot.

Answer:

• Branches used: Speech Recognition (speech → text) + NLP (understand intent, generate response) + possibly Dialogue Management.
• Data needed: Large speech corpora (different accents), conversational transcripts, domain-specific FAQs and customer history.
• Ethical safeguards:
  • Disclose that it is an AI.
  • Protect user privacy (no unnecessary data retention).
  • Provide human escalation option (Human-in-the-Loop).
  • Audit for bias in language understanding and responses.
  • Ensure transparency and accountability.

6. A city wants to use AI for dynamic traffic-signal control and facial-recognition-based surveillance.

Answer:

a) Benefits: Reduced congestion, lower emissions, improved safety, faster emergency response. Risks: Mass surveillance, privacy invasion, potential misuse, bias in facial recognition (higher error rates for certain groups).

b) Privacy: Data minimization, purpose limitation, consent or legal basis, anonymization where possible. Fairness: Regular bias audits of facial recognition models.

c) Governance measures: Classify surveillance as high-risk, require impact assessments, human oversight, transparency reports, clear legal framework, and public accountability mechanisms.

7. Why did AI experience “winters”? Relate to limitations and the 2012 breakthrough.

Answer:

AI winters occurred due to over-optimism followed by disappointment when systems failed to deliver. Symbolic AI’s brittleness (inability to handle uncertainty and common sense) and the limited scalability of early systems led to funding cuts. Early ML also struggled with feature engineering bottlenecks and computational limits.

The 2012 Deep Learning breakthrough (ImageNet success with CNNs), powered by Big Data + GPUs + better algorithms, demonstrated dramatic performance gains, ending the long winter and launching the current AI boom.

8. An AI system rejects a loan application. The applicant demands an explanation.

Answer:

a) Transparency / Explainability principle (and Accountability).

b) Modern regulations (e.g., EU-style rules) require that if an AI rejects a loan or job, the company must explain the decision in simple words.

c) Technical challenge: Deep Learning models are often black boxes; providing human-understandable explanations requires additional techniques (e.g., LIME, SHAP, attention visualization, or hybrid systems).

9. Map the milestones to the correct technical era:

Answer:

• ELIZA (1960s) → Symbolic AI – early rule-based chatbot.
• Deep Blue (1997) → Late Symbolic / Search-based AI – defeated chess champion using massive search and evaluation functions.
• ImageNet/CNN breakthrough (2012) → Deep Learning – demonstrated power of deep neural networks for vision.
• AlphaGo (2016) → Deep Learning – combined deep neural networks with search to master Go.
• GPT models (2018–2020s) → Deep Learning (Transformers) – large language models based on representation learning.

10. “AI will both destroy and create jobs.” Analyze the impact and suggest skills for engineers.

Answer:

AI automates routine/repetitive tasks (data entry, simple analysis, some manufacturing and service roles), potentially reducing demand for those jobs. At the same time, it creates new roles in AI development, data science, AI ethics, system maintenance, and human-AI collaboration, while boosting overall productivity and economic growth.

Skills engineers should develop:

• Strong foundation in AI/ML/DL concepts and Python.
• Ability to work with AI tools (prompt engineering, model evaluation).
• Domain knowledge + AI integration skills.
• Understanding of ethics, bias, and governance.
• Continuous reskilling and lifelong learning mindset.

---