Artificial Intelligence 2022

Section II (Short Answer)

Q.2-i. Define Inference Techniques. Answer: In Artificial Intelligence, an inference technique is the logical process used by an AI system (like an Expert System) to deduce new information or conclusions from a set of existing facts and rules stored in its knowledge base. It acts as the "brain" of the system. There are two main techniques used by an inference engine:

• Forward Chaining (Data-Driven): The system starts with the known facts and applies rules to extract new facts until a specific goal is reached.
• Backward Chaining (Goal-Driven): The system starts with a goal or hypothesis and works backward to find the facts and rules that support it.

Q.2-ii. Discuss four elements of agent's environment (i.e., PEAS). Answer: When designing an AI agent, its task environment is defined using the PEAS model. PEAS stands for Performance, Environment, Actuators, and Sensors. For example, in a self-driving car agent:

• Performance Measure: The criteria used to evaluate how successful the agent is (e.g., reaching the destination safely, obeying traffic laws, saving fuel).
• Environment: The external world the agent interacts with (e.g., roads, pedestrians, other cars, weather conditions).
• Actuators: The mechanisms the agent uses to perform actions in its environment (e.g., steering wheel, brakes, accelerator).
• Sensors: The tools the agent uses to perceive its environment (e.g., cameras, GPS, radar, speedometer).

Q.2-iii. What is Heuristics? How to use it in problem solving? Answer: A heuristic is a "rule of thumb" or a mental shortcut used in AI to solve problems faster. Instead of searching every possible solution (which is slow and takes too much computing power), a heuristic guesses the most promising path to the goal based on experience or estimation. How to use it: In problem-solving algorithms like A* Search, a heuristic function (often written as h(n)) estimates the cheapest cost from the current state to the goal. For example, if you are navigating a maze, a heuristic might calculate the straight-line distance to the exit. Even if a wall is in the way, this estimate helps the AI prioritize moving in the general direction of the goal, drastically speeding up the search process.

Q.2-iv. Describe different categories of Expert System. Explain only one in detail. Answer: Expert systems are AI programs designed to solve complex problems by mimicking the decision-making ability of a human expert. The main categories include: Diagnosis, Prediction, Planning, Design, and Monitoring systems.

• Diagnosis Expert Systems (Explained): These systems are built to deduce faults, diseases, or errors from a given set of symptoms or data. They are heavily used in the medical and engineering fields. For example, MYCIN is a famous medical diagnosis expert system. A doctor inputs a patient's symptoms and blood test results, and the system uses its knowledge base of rules to identify the specific bacteria causing the infection and recommends the correct dosage of antibiotics.

Q.2-v. What are Rational Agents? Describe briefly. Answer: A rational agent is an AI entity that strives to "do the right thing" based on the information it has. Specifically, for every possible sequence of perceptions, a rational agent selects an action that is expected to maximize its performance measure, given the evidence provided by its sensors and any built-in knowledge it possesses.

• Example: A robotic vacuum cleaner is a rational agent if it chooses to suck up dirt when it senses a dirty floor, rather than randomly moving around or doing nothing. Its rationality is tied to maximizing its goal (a clean floor) based on what it can currently perceive.

Q.2-vi. What is game playing? Answer: In Artificial Intelligence, game playing refers to a competitive, multi-agent environment where agents make decisions against an opponent. These are typically modeled as "zero-sum games," meaning one player's win is the other player's loss. AI uses adversarial search algorithms, like Minimax, to calculate the best possible moves by predicting the opponent's responses.

• Key points: Game environments are highly structured, have clear rules, and define a clear winner or loser (e.g., Chess, Tic-Tac-Toe, or Checkers).

Section III (Essay Type)

Q.3- What is a Neural Network? Explain with example. Answer: An Artificial Neural Network (ANN) is a machine learning model inspired by the structure and function of the human brain. It consists of interconnected processing units called "artificial neurons" or "nodes," which work together to recognize patterns, solve complex problems, and learn from data without being explicitly programmed with rules.

Architecture of a Neural Network: A standard neural network is divided into three main types of layers:

1. Input Layer: Receives the raw data or features from the outside world.
2. Hidden Layers: One or more layers between the input and output where the actual processing, calculations, and pattern extraction happen using "weights" and "activation functions."
3. Output Layer: Delivers the final prediction or result.

How it Works (with Example): Imagine we are training a Neural Network to recognize whether an image contains an apple or an orange.

• Input: The input layer receives the pixels of the image, breaking down details like color, shape, and texture.
• Processing: The hidden layers process this data. One hidden node might trigger when it detects the color red; another might trigger when it detects a round shape. During training, the network adjusts the "weights" (importance) of these connections. It learns that red + round strongly suggests an apple.
• Output: The output layer combines the findings from the hidden layers and calculates a probability. It outputs the final decision: "This image is 95% likely to be an apple."

Q.4- What is Genetic Algorithm? Explain with example. Answer: A Genetic Algorithm (GA) is an optimization and search technique inspired by Charles Darwin’s theory of natural evolution. It belongs to the broader class of evolutionary algorithms. GAs are used to find optimal or near-optimal solutions to difficult problems by simulating the process of natural selection, where the "fittest" solutions survive and reproduce.

Core Phases of a Genetic Algorithm:

1. Initialization: Generating a random starting population of possible solutions (called "chromosomes").
2. Fitness Evaluation: A mathematical function calculates how good or "fit" each solution is at solving the problem.
3. Selection: The best (fittest) solutions are chosen to be parents for the next generation.
4. Crossover (Reproduction): The selected parents combine their traits to create new offspring solutions.
5. Mutation: Random, minor changes are applied to the offspring to maintain diversity and prevent the algorithm from getting stuck.

Example: Consider the problem of University Timetable Scheduling. Creating a timetable without overlapping classes or double-booking teachers is incredibly complex. A Genetic Algorithm solves this by generating hundreds of random timetables (the population). It then tests them: timetables with many clashes get a low fitness score, and those with fewer clashes get a high score. The best timetables are combined (Crossover) to make a new generation of timetables. Random tweaks are applied (Mutation), like swapping a math class with a physics class. Over many generations, the timetable "evolves" until a perfect, clash-free schedule is found.

Q.5- What is Task Environment of an agent? Describe different task environments briefly. Answer: The Task Environment is the specific "problem space" or external setting in which an AI agent operates. The design of an agent heavily depends on the nature of its task environment.

Task environments are categorized by several different properties:

• Fully Observable vs. Partially Observable:

• Fully: The agent's sensors give it access to the complete state of the environment at all times (e.g., Chess board).

• Partially: The agent cannot see everything due to noisy sensors or missing data (e.g., A poker game where opponents' cards are hidden).

• Deterministic vs. Stochastic:

• Deterministic: The next state of the environment is completely determined by the current state and the agent's action (e.g., Tic-Tac-Toe).

• Stochastic: The environment has elements of randomness; the exact outcome of an action cannot be predicted (e.g., Driving a car—weather or other drivers act randomly).

• Episodic vs. Sequential:

• Episodic: The agent's experience is divided into independent "episodes" or tasks. The next action doesn't depend on previous actions (e.g., An AI sorting defective parts on a conveyor belt).

• Sequential: Current decisions affect all future decisions. Long-term planning is required (e.g., Chess or pac-man).

• Static vs. Dynamic:

• Static: The environment does not change while the agent is "thinking" or computing its next move (e.g., Crossword puzzle).

• Dynamic: The environment keeps changing while the agent is making decisions (e.g., A self-driving car on a busy highway).

• Single-agent vs. Multi-agent:

• Single: Only one agent is operating in the environment (e.g., Solving a maze).

• Multi: Multiple agents are interacting, either cooperatively or competitively (e.g., A robot soccer team).

Q.6- Explain the difference between informed and uninformed search techniques. Elaborate your response using suitable examples. Answer: Search techniques in AI are algorithms used to navigate from an initial state to a goal state. They are broadly divided into two categories based on how much information they have about the problem space.

1. Uninformed Search (Blind Search): Uninformed search algorithms have no additional information about the goal other than the problem definition. They blindly search through the state space by systematically expanding nodes until they hit the goal.

• Characteristics: They are usually slower, use more memory, and do not use heuristics.
• Examples: Breadth-First Search (BFS), Depth-First Search (DFS).
• Elaboration Example: Imagine you are in a maze blindfolded. You have no idea where the exit is. An uninformed approach (like DFS) would involve walking down a single path until you hit a dead end, backing up, and trying the next path, systematically trying every route until you accidentally stumble upon the exit.

2. Informed Search (Heuristic Search): Informed search algorithms use domain-specific knowledge—called a "heuristic"—to find the goal state more efficiently. They don't just blindly explore; they actively guess which path is most likely to lead to the goal.

• Characteristics: Highly efficient, consumes less time and memory, and uses heuristic functions (h(n)) to estimate the cost to the goal.
• Examples: A* Search, Greedy Best-First Search.
• Elaboration Example: Imagine you are in a large city trying to find a specific landmark, but this time you have a compass. The compass doesn't give you the exact roads, but it tells you the general direction of the landmark (this is your heuristic). At every intersection, you choose the road that points closest to your target direction. You will find the landmark much faster than searching blindly.

Comparison Summary: