Artificial Intelligence

Chapter 1, 2, 9, 10

What is the definition of Artificial Intelligence (AI)?

Simulation of human intelligence in machines, enabling them to perform tasks that typically require human cognitive functions such as learning, reasoning, problem-solving, and understanding natural language

• also defined as the theory and development of computer systems capable of performing tasks that historically required human intelligence

What is the goal of Artificial Intelligence as a science?

To make machines do things that would require intelligence if done by humans

Name the 4 main approaches to defining AI based on dimensions like 'thought vs. behaviour' and 'human-like vs. rational’.

1. Systems that Act/Behave Like Humans (The Turing Test Approach)

2. Systems that Think Like Humans (The Cognitive Modeling Approach)

3. Systems that Think Rationally (The Laws of Thought Approach)

4. Systems that Act Rationally (The Rational Agent Approach)

Briefly describe the Turing Test Approach to AI.

The Turing Test Approach suggests that a computer is said to be intelligent if its response to an unrestricted textual natural-language conversation cannot be distinguished from that of a human’s.

What capabilities would a computer need to pass the Turing Test?

• Natural language processing (to communicate)

• Knowledge representation (to store information)

• Automated reasoning (to answer questions and draw conclusions)

• Machine learning (to adapt and detect patterns)

Explain "Rationality" in the context of AI.

In AI, rationality refers to maximally achieving pre-defined goals.

It concerns what decisions are made, not the thought process behind them. Being rational means maximising your expected utility.

List 2 ADVANTAGES of Artificial Intelligence.

• High precision, accuracy, and speed with low error rates compared to humans.

• Suitable for repetitive tasks & boring jobs.

List 2 DISADVANTAGES of Artificial Intelligence.

• High costs of creation

• Potential for making humans lazy and contributing to unemployment

What is an AI Agent?

An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.

What does the acronym PEAS stand for in the context of intelligent agents, and what is its purpose?

PEAS stands for Performance measure, Environment, Actuators, and Sensors. It helps to define the task environment for an intelligent agent.

Differentiate between a "fully observable" and a "partially observable" environment for an agent, with examples.

• Fully Observable: An environment where an agent's sensors can access the complete state of the environment at each point in time (e.g., Chess).

• Partially Observable: An environment where the agent's sensors cannot access the complete state (e.g., Driving, as what's around the corner is unknown).

Explain the difference between "deterministic" and "stochastic" environments, with examples.

• Deterministic: An environment where a unique current state completely determines the next state of the agent (e.g., Chess).

• Stochastic: An environment that is random in nature, where the next state cannot be completely determined by the agent (e.g., Self-Driving Cars, where actions vary over time).

What are the 4 basic types of agents in increasing order of generality and ability?

The four basic types of agents are:

1. Simple reflex agents.

2. Model-based reflex agents.

3. Goal-based agents.

4. Utility-based agents. All of these can be turned into Learning agents.

What is a "Simple Reflex Agent" and what are its limitations?

A Simple Reflex Agent is the simplest type, making decisions based only on current percepts and ignoring past percept history. They only succeed in fully observable environments and work on a Condition-action rule (mapping current state to action).

Limitations include limited intelligence, no knowledge of non-perceptual parts of the state, and lack of adaptability.

What key factors define a "Model-Based Reflex Agent"?

A Model-Based Reflex Agent can work in a partially observable environment and tracks the situation.

It has two key factors: a Model (knowledge about "how things happen in the world") and an Internal State (a representation of the current state based on percept history).

How do "Goal-Based Agents" differ from "Model-Based Agents"?

Goal-Based Agents expand on model-based agents by incorporating "goal" information. They choose actions specifically to achieve a desirable situation (goal), which may involve considering sequences of actions through searching and planning.

What makes "Utility-Based Agents" distinct from "Goal-Based Agents"?

Utility-Based Agents are similar to goal-based agents but include an extra component of utility measurement. This provides a measure of success at a given state, allowing the agent to choose the best action when there are multiple possible alternatives, not just whether a goal is achieved.

What are the 4 conceptual components of a "Learning Agent"?

The four conceptual components of a Learning Agent are:

1. Learning element: Responsible for making improvements by learning from the environment.

2. Critic: Provides feedback to the learning element on how well the agent is performing.

3. Performance element: Responsible for selecting external actions.

4. Problem generator: Suggests actions that lead to new and informative experiences.

What is "Probabilistic Reasoning" in AI?

Probabilistic reasoning is a way of knowledge representation where the concept of probability is applied to indicate uncertainty in knowledge.

It combines probability theory with logic to handle situations where predicates are not certain (e.g., "It will rain today").

What is the range of values for probability, and what do the extreme values indicate?

The value of probability always remains between 0 and 1.

• P(A) = 0 indicates total uncertainty in an event A.

• P(A) = 1 indicates total certainty in an event A.

Define "Conditional Probability".

Conditional probability is the probability of an event occurring when another event has already happened.

It is denoted as P(A|B), meaning the probability of A under the condition of B.

What is Bayes' Theorem used for?

Bayes' Theorem (also known as Bayes' rule or Bayesian reasoning) is used to determine the probability of an event with uncertain knowledge.

It relates the conditional probability and marginal probabilities of two random events and provides a way to calculate P(B|A) with the knowledge of P(A|B).

What is the formula for Bayes' Theorem?

P(A|B) = [P(B|A) * P(A)] / P(B).

List some applications of Bayes' Theorem in AI.

Bayes' Theorem is applied in AI for:

• Calculating the next step of a robot when a previous step is known.

• Weather forecasting.

• Solving the Monty Hall problem (a mathematical problem).

• Test diagnosis in the medical field.

What is the primary significance of an Artificial Neural Network (ANN)?

Its ability to learn from its environment and improve its performance through learning

How is the learning process of a Neural Network often viewed?

The learning process can be viewed as reshaping a sheet of metal (representing the output range of the function being mapped), where the training set acts as the energy to bend the metal to pass through predefined points.

The network seeks a low energy (flat/non-wrinkled) configuration that satisfies the training data constraints.

What is an Artificial Neural Network (ANN)?

A method of computing based on the interaction of multiple connected processing elements.

It consists of a number of very simple processors (neurons), analogous to biological neurons, which are connected by weighted links passing signals.

What are the key analogies between biological and artificial neural networks?

• Soma (Biological) is analogous to Neuron (Artificial).

• Dendrite (Biological) is analogous to Input (Artificial).

• Axon (Biological) is analogous to Output (Artificial).

• Synapse (Biological) is analogous to Weight (Artificial).

What are the typical layers of an Artificial Neural Network?

3 layers: an Input Layer, Hidden Layer(s), and an Output Layer

What is a Perceptron and who introduced it?

A Perceptron is the simplest form of a neural network, consisting of a single neuron with adjustable synaptic weights and a hard limiter.

It was introduced by Frank Rosenblatt in 1958.

What is the aim of a Perceptron?

To classify inputs into one of two classes (e.g., A1 and A2).

It produces an output of +1 if its input is positive and −1 if it is negative, based on a weighted sum and a hard limiter.

Explain the concept of "backpropagation" in neural networks.

Backpropagation is the process where, after an initial mistake is made by the computer, this information is backtracked through the algorithm.

Connections supporting the incorrect answer become weaker, while connections supporting the right answer become stronger. This iterative adjustment helps the network learn.

What do "Learning Rate" and "Momentum" refer to in ANNs?

Learning Rate: Determines how much the current step's weight and outcome affect the biases.

Momentum: Determines how much past outcomes should affect the weights and biases.