Unit1 - Subjective Questions

Definition of AI: Artificial Intelligence (AI) is the branch of computer science concerned with making computers behave like humans. It is the study of how to make computers do things at which, at the moment, people are better.

Four Approaches to Defining AI:

1. Acting Humanly (The Turing Test Approach):
   • Focuses on behavior that is indistinguishable from human behavior.
   • Example: A chatbot passing the Turing Test by fooling a human interrogator.
2. Thinking Humanly (The Cognitive Modeling Approach):
   • Focuses on understanding the underlying mechanisms of the human mind and mimicking them.
   • Example: General Problem Solver (GPS) which attempted to mimic human problem-solving steps.
3. Thinking Rationally (The 'Laws of Thought' Approach):
   • Uses logic and formal inference to reach correct conclusions.
   • Example: Syllogisms (e.g., "Socrates is a man; all men are mortal; therefore, Socrates is mortal").
4. Acting Rationally (The Rational Agent Approach):
   • Focuses on agents that act to achieve the best outcome or, when there is uncertainty, the best expected outcome.
   • Example: An autonomous vacuum cleaner aiming to maximize the area cleaned while conserving battery.

AI is a multidisciplinary field built upon several foundations:

• Philosophy: Addressed questions like "Can formal rules be used to draw valid conclusions?" and "Where does knowledge come from?" It provided the basis for logic and reasoning.
• Mathematics: Provided the tools for formal logic, probability (), and computation theory (algorithms and tractability).
• Economics: Contributed decision theory and game theory, helping AI agents make rational choices to maximize utility.
• Neuroscience: Studied how the brain processes information, inspiring neural networks and deep learning models.
• Psychology: Cognitive psychology views the brain as an information-processing device, influencing AI knowledge representation.
• Computer Engineering: Provided the hardware (GPUs, TPUs) required to run complex AI applications.
• Linguistics: The study of language structure relates to Natural Language Processing (NLP) and understanding syntax and semantics.

Artificial Narrow Intelligence (ANI):

• Definition: AI systems designed and trained for a particular task.
• Capability: Specialized and operates within a limited context. It cannot transfer knowledge to other domains.
• Examples: Siri, Alexa, Google Search, AlphaGo, Self-driving cars.
• Current Status: This is the current state of AI technology.

Artificial General Intelligence (AGI):

• Definition: A theoretical form of AI where a machine possesses the ability to understand, learn, and apply knowledge across a wide variety of tasks, much like a human.
• Capability: Flexible, general-purpose intelligence capable of common sense and abstraction.
• Examples: Fictional robots like C-3PO or Data from Star Trek.
• Current Status: Hypothetical; researchers are working towards it, but it does not yet exist.

Responsible AI is a governance framework that documents how a specific organization addresses the ethical, legal, and societal challenges of artificial intelligence.

Key Principles:

1. Fairness: AI systems should treat all people fairly and avoid bias (e.g., racial or gender bias in hiring algorithms).
2. Reliability & Safety: AI should perform as intended and be resistant to errors or attacks.
3. Privacy & Security: Systems must protect user data and ensure privacy.
4. Inclusiveness: AI should be accessible to and benefit all layers of society.
5. Transparency: Users should understand when they are interacting with an AI.
6. Accountability: Organizations must be accountable for their AI systems' actions.

Importance: It is critical to build trust, prevent harm, ensure legal compliance, and avoid reinforcing societal inequalities.

The evolution of AI can be categorized into several phases:

1. The Gestation (1943-1955): McCulloch & Pitts neuron model; Turing's "Computing Machinery and Intelligence".
2. The Birth of AI (1956): Dartmouth Conference where the term "Artificial Intelligence" was coined by John McCarthy.
3. Early Enthusiasm / Golden Years (1952-1969): Development of early programs like ELIZA and the General Problem Solver. High expectations.
4. A Dose of Reality / First AI Winter (1966-1973): Realization that translation and complex problems were harder than expected due to computational limits.
5. Expert Systems (1969-1986): Rise of rule-based systems (e.g., MYCIN) used in industry. Knowledge-based approaches.
6. Second AI Winter (1987-1993): Collapse of the Lisp machine market and failure of high-cost expert systems.
7. The Return of Neural Networks (1986-Present): Backpropagation algorithm popularized.
8. AI Boom / Big Data (2011-Present): Availability of massive datasets, GPUs, and Deep Learning breakthroughs (AlexNet, Transformers, GPT).

TensorFlow:

• Developer: Google.
• Graph Type: Originally used static computation graphs (though TF 2.0 introduced eager execution).
• Deployment: Excellent for production and deployment (TensorFlow Serving, TFLite).
• Usage: widely used in large-scale industrial applications.

PyTorch:

• Developer: Facebook (Meta) AI Research.
• Graph Type: Uses dynamic computation graphs (define-by-run), making debugging easier.
• Deployment: Gaining ground with TorchServe, but historically stronger in research.
• Usage: Preferred by the research community for its pythonic nature and flexibility.

Conclusion: Both are powerful deep learning frameworks; the choice often depends on whether the goal is rapid research prototyping (PyTorch) or scalable production deployment (TensorFlow).

A Problem Space represents the environment in which the search for a solution takes place.

A problem is formally defined by four components:

1. Initial State (): The state in which the agent begins.
2. Actions (): A description of possible actions available to the agent. Given a state , returns the set of actions executable in .
3. Transition Model / Successor Function: A description of what each action does. Result() returns the state resulting from doing action in state .
   • Together, the initial state, actions, and transition model define the State Space (a graph where nodes are states and links are actions).
4. Goal Test: Determines whether a given state is a goal state.
5. Path Cost: A numeric cost assigned to each path. The problem-solving agent chooses a cost function that reflects the cost of going from state to via action .

Problem Statement: You have two jugs, a 4-gallon jug and a 3-gallon jug. Neither has any measuring markers. There is a pump that can be used to fill the jugs with water. How can you get exactly 2 gallons of water in the 4-gallon jug?

State Space Representation:

• State: Represented by an ordered pair , where is the number of gallons in the 4-gallon jug () and is the number of gallons in the 3-gallon jug ().
• Initial State: (Both jugs are empty).
• Goal State: (The 4-gallon jug contains 2 gallons, can be anything).

Operators (Rules):

1. Fill the 4-gallon jug:
2. Fill the 3-gallon jug:
3. Empty the 4-gallon jug:
4. Empty the 3-gallon jug:
5. Pour from 3-gallon to 4-gallon until full: if
6. Pour from 4-gallon to 3-gallon until full: if
7. Pour all from 3-gallon to 4-gallon: if
8. Pour all from 4-gallon to 3-gallon: if

To choose an appropriate method for a problem, we must analyze its characteristics:

1. Decomposable vs. Ignorable: Can the problem be broken down into smaller sub-problems? (e.g., Integration is decomposable; 8-puzzle is not).
2. Recoverable vs. Irrecoverable: Can steps be undone? (e.g., Chess is recoverable via backtracking; Playing cards might be irrecoverable if you cannot see previous cards).
3. Predictable vs. Unpredictable: Is the outcome of an action certain? (e.g., 8-puzzle is predictable; Bridge is unpredictable/uncertain).
4. Good solution Absolute vs. Relative: Do we need the best answer or just an answer? (e.g., Traveling Salesman requires the shortest path - relative/optimal; Answering "Who is the President?" is absolute).
5. State Knowledge: Is the knowledge consistent or inconsistent? (e.g., Logic problems vs. Legal argumentation).
6. Role of Knowledge: Does the problem require a lot of knowledge or just search? (e.g., Medical diagnosis requires vast knowledge; Chess relies heavily on search).
7. Interaction: Is the agent solitary or interacting with others? (e.g., Solitaire vs. Poker).

A typical AI workflow consists of the following iterative stages:

1. Problem Definition: Clearly defining the business or scientific problem and the desired outcome.
2. Data Collection: Gathering raw data from various sources (databases, sensors, web scraping).
3. Data Preparation (Preprocessing):
   • Cleaning: Removing noise, handling missing values.
   • Transformation: Normalization, scaling.
   • Splitting: Dividing data into training, validation, and test sets.
4. Feature Engineering: Selecting or creating relevant features that help the model learn better.
5. Modeling:
   • Selection: Choosing the algorithm (e.g., Regression, Random Forest, Neural Net).
   • Training: Feeding data into the algorithm to learn patterns.
6. Evaluation: Testing the model against unseen data using metrics (Accuracy, Precision, Recall).
7. Deployment: Integrating the model into a production environment.
8. Monitoring & Maintenance: Tracking performance over time and retraining as necessary.

Model-Centric AI:

• Focus: The primary focus is on improving the AI algorithm or model architecture.
• Approach: Keep the data fixed and iterate on the code/model to get better results.
• Use Case: Academic research where datasets (like MNIST or ImageNet) are standard benchmarks.

Data-Centric AI:

• Focus: The primary focus is on the quality of the data used to train the system.
• Approach: Keep the model/code relatively fixed and iterate on the data (cleaning, labeling, augmenting) to improve results.
• Philosophy: "Good data is better than complex models." This is increasingly vital in real-world industrial applications where data is often messy or scarce.

Solving problems with AI involves several significant challenges:

1. Data Quality and Quantity: AI requires large amounts of high-quality, labeled data. Garbage In, Garbage Out (GIGO) applies; biased or incomplete data leads to poor models.
2. Computational Power: Training complex models (like Deep Learning) requires expensive hardware (GPUs/TPUs) and high energy consumption.
3. Overfitting and Underfitting: The model might learn the training data too well (overfitting) and fail on new data, or fail to learn patterns at all (underfitting).
4. Explainability (Black Box Problem): Many advanced models (like Deep Neural Networks) act as black boxes; it is difficult to understand why they made a specific decision.
5. Generalization: Creating models that perform well on data they have never seen before.
6. Ethical and Legal Issues: Bias, privacy concerns, and liability for AI errors.

AI in Healthcare:

• Diagnostics: AI analyzes medical imaging (X-rays, MRIs) to detect diseases like cancer earlier and more accurately than humans.
• Drug Discovery: predicting molecular behavior to speed up the development of new medicines.
• Personalized Medicine: Tailoring treatment plans based on a patient's genetic makeup.
• Robotic Surgery: Assisting surgeons with precision instruments.

AI in Automation:

• Industrial Robotics: Robots in manufacturing lines performing repetitive tasks (welding, painting) with computer vision for quality control.
• RPA (Robotic Process Automation): Software bots automating clerical tasks like data entry, invoice processing, and email sorting.
• Autonomous Vehicles: Self-driving logistics trucks and warehouse robots (e.g., Amazon Kiva robots).

PEAS stands for Performance, Environment, Actuators, and Sensors. It is a framework to define the task environment for an AI agent.

Example: Autonomous Taxi

• P - Performance Measure:
  • Safety (no accidents), Speed (reaching destination quickly), Legality (obeying traffic laws), Comfort (smooth ride), Profit (minimizing fuel/wear).
• E - Environment:
  • Roads, other traffic, pedestrians, weather conditions, road signs.
• A - Actuators:
  • Steering wheel, accelerator, brake, signal lights, horn, display screen.
• S - Sensors:
  • Cameras, sonar/radar, GPS, speedometer, odometer, engine sensors, keyboard (for passenger input).

A Production System is a mechanism for solving problems based on rules. It provides the structure for an AI system to reason and make decisions.

Components:

1. The Global Database (Working Memory): Contains the current state of knowledge or facts about the world (e.g., "The patient has a fever").
2. The Set of Production Rules: A set of IF-THEN rules. The IF part (condition) checks the global database, and the THEN part (action) modifies the database.
   • Example: IF (Temperature > 100) THEN (Diagnosis = Fever).
3. The Control System (Inference Engine): The program that controls the process. It matches rules against the database, resolves conflicts (if multiple rules match), and executes the action (fires the rule).

The 8-Puzzle Problem:
It consists of a grid with 8 numbered tiles (1 to 8) and one blank space. The objective is to reach a specific configuration (goal state) from a given initial configuration by sliding tiles into the blank space.

State Space Representation:

• State: A matrix or array representing the position of the 8 tiles and the blank.
• Initial State: A random configuration.
• Goal State: Usually ordered, e.g.,
  

Operators:
Instead of moving tiles, we think of moving the Blank Space.

1. Move Blank Up
2. Move Blank Down
3. Move Blank Left
4. Move Blank Right

(Constraints: The blank cannot move out of the grid boundaries).

Computer Vision is a field of AI that enables computers and systems to derive meaningful information from digital images, videos, and other visual inputs, and take actions or make recommendations based on that information.

Applications:

1. Facial Recognition: Used in security systems, phone unlocking, and tagging photos on social media.
2. Autonomous Driving: Detecting lanes, traffic lights, pedestrians, and obstacles.
3. Medical Imaging: Analyzing X-rays and MRI scans to detect tumors or fractures.
4. Retail: Amazon Go stores (cashier-less shopping), inventory management.
5. Manufacturing: Defect detection in assembly lines.

Natural Language Processing (NLP) is a branch of AI concerned with the interaction between computers and human language. It involves programming computers to process and analyze large amounts of natural language data.

Applications:

1. Machine Translation: Google Translate (translating text from one language to another).
2. Sentiment Analysis: Analyzing customer feedback or tweets to determine if the sentiment is positive, negative, or neutral.
3. Chatbots & Virtual Assistants: Siri, Alexa, and Customer Service bots.
4. Text Summarization: Automatically generating summaries of long articles.
5. Spam Filtering: Classifying emails as spam or ham based on content.

• Agent: Anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators. An agent maps percepts to actions ().
• Sensor: A device that allows an agent to perceive its environment. (e.g., Eyes/Ears for humans, Cameras/Microphones for robots).
• Actuator: A mechanism that allows an agent to perform actions on the environment. (e.g., Hands/Legs for humans, Motors/Wheels for robots).
• Environment: The surroundings or conditions in which the agent operates. It provides the percepts to the agent and is changed by the agent's actions.