2024 - 2025 CI406 Introduction to Artificial Intelligence (no Python)

Applications of AI

Robotics - Humanoid Robots, Robotic Surgery, Boston Dynamics
Healthcare - Medical Imaging diagnosis, Disease diagnosis, Drug discovery
Anomaly detection - Assisted Daily Living, Credit Card Fraud Detection
Self-Driving cars
Finance - Stock market predictions
E-commerce - Recommendations, personalised services, Marketing, Market basket analysis
Entertainment - game design, Film, and music streaming Platform

What are neural networks?

Algorithms modelled after the human brain, drawing inspiration from neuroscience, are supervised learning algorithms. These algorithms learn from training data, identifying patterns as they process information through their layers. They are commonly used in applications such as image recognition, generative AI, and speech recognition.

What is Generative AI?

A type of artificial intelligence (AI) that can create new content, such as text, images, audio, and video. It does this by learning from existing data and then using that knowledge to generate new and unique outputs.

What is edge detection?

• This is a technique of image processing used to identify points in a digital image with discontinuities, ~ sharp changes in the image brightness.
  -It is also a method of segmenting an image into discontinuity regions.
• This can be used in Image Classification as it is easier for an AI to process what is being displayed when there are fewer variables at play

What happened to Garry Kasparov in 1997?

In 1997, world chess champion Garry Kasparov faced off against an AI called Deep Blue created by IBM. Although the first match ended with Garry winning (4-2), the AI won in the rematch (3½-2½).

When was the First general-purpose, purpose-built robot?

1966-1972

What was the name of the First general-purpose, purpose-built robot?

Shakey the robot

What did the First general-purpose, purpose-built robot do?

Computer vision and natural language processing technologies

What is the DaVinci Robot?

A robotic surgical system used by surgeons to perform robotic-assisted surgery, extending the capabilities of their eyes and hands. More than 76,000 surgeons around the world have been trained on da Vinci systems and have completed more than 14 million surgical procedures.

What is the biggest challenege for humanoid robots?

How can robots be integrated into the human environment?

• How do we put a robot in a primary school to clean up?
• How can we put a robot into our kitchen to cook?
• Robots are employed doing complicated functionalities - eg building aeroplanes, robots on Mars

Who coined the term Artificial Intelligence?

John McCarthy in 1955

What did John McCarthy define Ai as?

"The goal of AI is to develop machines that behave as though they were intelligent"

Is this intelligent behaviour according to McCarthy's definition?

• Fifteen robotic vehicles moving on a 4mx4m surface
• Various patterns of behaviour observed
• Some vehicles form small groups, and relatively little movement
• Some follow a leader
• Others move peacefully through the space
• Aggressive behaviour was also observed

Yes, as he thought of AI as developed machines that behave as though they were intelligent.

How does McCarthy's definition of AI hold up with the Baritenberg vehicle thought experiment?

His definition no longer holds true, the experiment was that Braitenberg vehicles have two wheels, each driven by an independent motor and have light sensors

• Vehicle 1 moves towards the light sensor; Vehicle 2 moves away from the light sensor.
• The more light hits the sensor, the faster the vehicle travels
  So that would not contend for intelligence, rather following commands

How does Encyclopaedia Britannica define AI?

"AI is the ability of digital computers or computercontrolled robots to solve problems that arenormally associated with the higher intellectualprocessing capabilities of humans"

Why is the Encyclopaedia Britannica's definition of AI poor?

Consider processing the following numbers by a human and a computer. Who would compute it faster?
12345676542 x 243562829262=?
Does this mean computers are intelligent enough to compute this faster than humans? Are all computers intelligent?

How did Alaine Rich (1983) define AI?

"AI is the study of how to make computers do things at which, at the moment, people are better"

Why is Alaine Rich's defintion of AI good?

It acknowledges that tasks such as executing many computations in a short amount of time are possible with digital computers. On the other hand, humans are superior in other areas— time is a factor, for example, someone entering a room recognising familiar surroundings, making decisions, and planning actions quickly. This task is too demanding for autonomous robots.

What are the four categories AI falls under?

• Thinking humanly
• Thinking rationally
• Acting humanly
• Acting rationally

What does thinking humanly: cognitive modelling mean?

Understand how the human brain works and model or simulate it on a computer. Cognitive science is devoted to research into human thinking at a higher level. Neural networks are an area in AI that stems from brain science related to neuroscience.

What does thinking Thinking rationally: "laws of thought" mean?

The Greek philosopher Aristotle was the first to establish the principles of "right thinking." For instance, he stated that "Socrates is a man," "all men are mortals," and "Socrates is mortal." These principles of thought were intended to guide the functioning of the mind and laid the foundation for the field of logic. In the 19th century, logicians developed a precise notation system for representing various objects and their relationships. By 1965, programs were created that could solve problems expressed in this logical notation, such as Prolog.

What was the Turing Test?

A machine will try to fool an interrogator into believing that the AI system is a human while the human is not. The human will act normally.

When did Alan Turing predict that machines may have a 30% chance of fooling a layperson ( a non-professional in a specific field) for five minutes

By 2000

What was Alan Turing's paper, written in 1950, called?

Computing Machinery and Intelligence

What are the major components of the Turing test?

Natural Language Processing: to be able to communicate.

• Knowledge Representation: to store what it already knows or hears
• Automated Reasoning: to answer questions using the stored information.
• Machine Learning: to adapt to new circumstances and detect and explore patterns.

What issues/limitations are there with the Turing test?

• It is not reproducible and cannot be analysed mathematically or algorithmically, as it is based on the opinion of the interrogator rather than defined characteristics.
• avoided physical interaction because at the time it would have been too obvious; however, even nowadays getting the minute movements of a person in modern society is difficult
• uncanny valley with what an AI may look like, due to humans' ability to define what a human looks like very specifically

What was the Loebner prize?

A competition to create an AI that best completed the Turing Test.

When was the Loebner Prize started?

In 1991

Who created the Loebner Prize?

Whilst basically being a competition for the Turing Test, the competition itself was created by Hugh Loebner and the Cambridge Centre for Behavioural Studies.

What was the 2015 winner of the Loebner prize?

Eugene Goostman, an algorithm in the form of a chat box pretending to be a 13-year-old boy, fooling ten of the 30 judges.

What were the criticisms of the Turing Test?

AI researchers question the validity and scientific basis of the test, which devalues real AI research and action. Pioneering technologies that employ sophisticated AI technology, such as those used in healthcare, are more important than the Turing Test. Chatbots that fool humans and make it appear like science fiction are not considered real AI.
Think, Planes were built once the Wright brothers stopped imitating birds. Once they started work on aerodynamics

What are Agents?

Anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators

What do Robiotic agents have?

• cameras and infrared range finders for sensors;
• Various motors for actuators

What are Software agents?

Agents may use keystrokes, file contents, etc., as sensory inputs and act on the environment by writing to files and displaying information on the screen.

What is Percept sequence?

A complete history of everything that the agenthas perceived

What is Agent function?

A mathematical function that maps the given percept-sequence/percept-histories into an action.
[f: P* → A]

What is the agent program's relation ot the architecture?

The agent program runs on the architecture to produce the agent.
Agent = architecture + program

What is a vacuum cleaner world?

A vacuum cleaner world is how we describe how everything works. For example, a world may have only two locations: A and B. A vacuum cleaner agent perceives which square it is in and whether it has dirt; its actions will include left, right, clean_dirt, or complete. Its Percepts are location and contents, e.g., [A, Dirty]

What is a rational agent?

An agent does the right thing. An agent should strive to "do the right thing", based on what it can perceive and the actions it can perform. The correct action is the one that will cause the agent to be most successful

What is a Performance measure in an agent?

An objective criterion for the success of an agent's behaviour. For example, a vacuum cleaner agent could be the amount of dirt cleaned up, the amount of time taken, the amount of electricity consumed, the amount of noise generated, etc.

How is rationality described for an Agent?

• The performance measure that defines the success criterion.
• The agent's prior knowledge of the environment.
• The actions that the agent can perform
• The agent's percept sequence to date

True or False: Rationality is distinct from omniscience?

True.

True or False: Rationality is not distinct from omniscience?

False

What does autonomous mean in an agent?

If its behaviour is determined by its own experience (with the ability to learn and adapt)

What does PEAS stand for?

Performance measure, Environment, Actuators, Sensors

True or False: We must first specify the setting for intelligent agent design

True.

What would the PEAS be for an Automated Taxi Driver?

Performance Measure - Safe, fast, legal, comfortable trip, maximise profits
Environment - Roads, other traffic, pedestrians, customers
Actuators - Steering, accelerator, signal, brake, horn, display
Sensors - Cameras, sonar, speedometer, GPS, engine sensor, accelerometer

What would the PEAS be for a Part-picking robot?

Performance Measure - Percentage of parts in correct bins
Environment - Conveyor belt with parts and bins
Actuators - Jointed arm and hand
Sensors - Camera, joint angle sensors

What does fully observable mean to an agent?

If an agent's sensors give it access to the complete state of the environment at each point in time, the agent has all the information to make the optimal decision, such as playing backgammon or chess.

What does partially observable mean to an agent?

The agent only has some information about the environment. In these cases, the agent requires internal memory to store past actions to make the best possible decisions. For example, when playing poker, not all cards are visible. The agent needs to"memorise" past moves, i.e keep information in internal memory.

True or False: An agent in a fully observable environment means the sensors see the entire state.

True

True or False: An agent in a partially observable environment means the sensors see the entire state.

False

True or false: An agent in a partially observable environment means the sensors can only see part of the environment.

True

True or false: An agent in a fully observable environment means the sensors can only see part of the environment.

False

Is the vacuum world a fully or partially observable environment?

Partially observable

What does it mean if an environment is deterministic?

If the next state of the environment is completely determined by the current state and the action executed by the agent.
Example - crossword puzzle

What does it mean if an environment is stochastic?

We are not able to predict what will happen next. For example, an automated taxi driving agent and games that use dice cannot predict what will happen.

What does it mean if an environment is Episodic?

The agent's experience is divided into atomic episodes in the episodic environment. The agent receives a percept in each episode and performs a single action. The next episode does not depend on the actions in the previous episodes. For example, a defective part picking robot - the outcome of one episode is not dependent on the next episode

What does it mean if an environment is sequential?

When the current decision can affect future decisions, episodic environments are much simpler than sequential environments. For example, playing chess, or an automated taxi driving agent

What does it mean if an environment is static?

If the environment is unchanged while an agent is thinking.

What does it mean if an environment is dynamic?

If the environment is constantly changing. The agent will need to monitor the changes in the environment.

What does it mean if an environment is discrete?

When an agent has a limited number of distinct, clearly defined percepts and actions. For example, chess, because there is a finite set of moves available.

What does it mean if an environment is continuous?

When an agent has a range of percepts and actions that may blend together, for example, a taxi driver agent, variables such as speed and location can change fluidly instead of being confined to a set of distinct options.

How would we describe a Crossword puzzle agent?
Fully/Partially Observable?
Agent/Multi-Agent?
Deterministic/Stochastic?
Episodic/Sequential?
Static/Dynamic?
Discrete/Continuous?

Fully
Single Agent
Deterministic
Sequential
Static
Descrete

How would we describe an Automated Taxi Driving agent?
Fully/Partially Observable?
Agent/Multi-Agent?
Deterministic/Stochastic?
Episodic/Sequential?
Static/Dynamic?
Discrete/Continuous?

Partially
Multi-Agent
Stochastic
Sequential
Dynamic
Continuous

List agent tpyes in order of increasing generality
Goal-based agents
Learning Agent
Table driven agents
Simple reflex agents
Utility-based agents
Model-based reflex agents

• Table driven agents
• Simple reflex agents
• Model-based reflex agents
• Goal-based agents
• Utility-based agents
• Learning Agent

What is a Table-driven agent?

An agent that uses a percept sequence/action table in memory to find the next action. They are implemented by a (large) lookup table.

What are the problems with Table-driven agents?

• Too big to generate and to store (Chess has about 10120 states, for example)
• Not adaptive to changes in the environment; requires the table to be updated if changes occur

What are Simple reflex agents?

Agents that ignore the percept history. It will select an action based on the current percept.

What does this mean:
function REFLEX-VACUM-AGENT ([percepts,status]) returns action
If status=dirty then return clean_dirt
else if location =A then return right
else if location =B then return left

Decision based on whether the current square contains dirt and not whether it is in A or B.

How could you code a Simple reflex agents in python?

import random

locA, locB=(0,0),(1,0)

environment={
locA:random.choice(['clean','dirty']), locB:random.choice(['clean','dirty'])
}

startlocation=random.choice([locA,loc_B])

def ReflexAgent(location,status): if status=='dirty': action='cleandirt'

else:

move right

if location==loc_A:
action='Right'

move left

if location==loc_B:
action='Left'
return action

ReflexAgent(startlocation, environment[start_location])

What are Model-based reflex agents?

To handle partial observability, the agent must keep track of the world it cannot observe. The agent should maintain an internal state that depends on the percept history, but can infer the unobserved state. For example, a taxi driving agent may keep track of where the other cars are on the road; closer when other cars are overtaking

What are goal-based agents?

Knowing something about the environment is not enough for the agent to decide what to do. The agent needs to have information about its goal

What are Utility-based agents?

When goals are not enough to create high-quality behaviour in most environments, this means that the Agent chooses actions to maximise its utility ( "the quality of being useful") function

What are Learning agents?

When an agent becomes more competent, it perceives more knowledge. It can be divided into four conceptual components: learning element, which is responsible for making improvements, performance element, which is responsible for selecting actions, critic, which can give feedback to the learning element regarding the performance, and problem generator, which suggests actions that will lead to new experiences

What are uninformed search strategies?

Search strategies that search through the search space without having any additional knowledge about the search space.

What are informed search strategies?

Search strategies that have additional knowledge, such as how far we are from the goal, path cost, how to reach the goal node, etc. This additional knowledge of the search space helps the intelligent agent explore less of the search space and find the goal node more efficiently. This means that they are more useful for large state spaces.
As well as that, this search strategy uses the idea of a heuristic function.

What does the Heuristic function mean?

A method of finding the most promising path. It takes the current state of the agents as input and produces an estimate of how close the node is to the goal.
Although it may not always give the best solution, it is guaranteed to find a good solution in a reasonable time.

How is a heuristic function represented?

h(n)

What is Best-First Search?

When using the evaluation function f(n) for each node to estimate the desirability. Then it will expand the most desirable (promising) unexpanded nodes.

What is greedy best-first search?

This is when the evaluation function "f(n)" equals h(n) (the heuristic)
This means that this search is the estimated cost of n to the goal. This search expands the node that appears to be closest to the goal.

In greedy best-first search, what does hSLD(n) mean?

straight-line distance from n to the goal

What is the A* / A-star Algorithm?

f(n) = g(n) + h(n)
where
g(n) = cost so far to reach n
h(n) = estimated cost from n to goal
f(n) = estimated total cost of path through n to goal (evaluation function)

What does Admissible Heuristic mean?

A heuristic h(n) is admissible if for every node n,
h(n) ≤ h(n), where h(n) is the true cost to reach the goal state from n.
In layman's terms, this means that an admissible heuristic never overestimates the cost to reach the goal, i.e., it is optimistic

Example: hSLD(n) (never overestimates the actual roaddistance)

For admissible heuristics, let's say we have an eight block puzzle. This is the number sliding game to get the numbers in the right order let's say that
h1(n) = number of misplaced tiles
h2(n) = total Manhattan distance
what would that mean?

h1(n) = if the tile is in the wrong location, count 1
h2(n). = How many squares is each title away from its desired location

Using the A* algorthm, how can we use it for the 8block puzzle? remember:
f(n) = g(n) + h(n)
h1(n) = number of misplaced tiles
h2(n) = total Manhattan distance

g(n) = 1 (step cost of 1)
h(n) = h1 (h1 heuristic - number of misplaced tiles)

True or False: The h(x) can be the Euclidean distance(straight line distance) or the Manhattan distance.

True

What would the Euclidean Distance between two points x and y be if
The hypotenuse (the long line) - d(x,y)
The opposite (opposite the angle) - y1-y2
the adjacent (opposite to the hypotenuse) - x1-x2
If the point between the hypotenuse and the adjacent is (y1,y2)
and if the point between the hypotenuse and opposite is (x1,x2)
(The point between opposite and adjacent is a right angle, but unimportant.)?

d(x,y) = square root( (x1-x2)^2 + (y1-y2)^2 )

What would the Manhattan Distance between two points x and y be if
The hypotenuse (the long line) - d(x,y)
The opposite (opposite the angle) - y1-y2
the adjacent (opposite to the hypotenuse) - x1-x2
If the point between the hypotenuse and the adjacent is (y1,y2)
and if the point between the hypotenuse and opposite is (x1,x2)
(The point between opposite and adjacent is a right angle, but unimportant.)?

d(x,y) = | x1 - x2 | + | y1 - y2 |

What would the Minkowski Distance between two points x and y be if
The hypotenuse (the long line) - d(x,y)
The opposite (opposite the angle) - y1-y2
the adjacent (opposite to the hypotenuse) - x1-x2
If the point between the hypotenuse and the adjacent is (y1,y2)
and if the point between the hypotenuse and opposite is (x1,x2)
(The point between opposite and adjacent is a right angle, but unimportant.)?

d(x,y) = ( | x1 - x2 |^k + | y1 - y2 |^k )^1/k

Let's say that:

• Point A is (2,3),
• Point B is (7,6)
  What is the Euclidean distance between them?

d(x,y) = square root( (x1-x2)^2 + (y1-y2)^2 )
d(a,b) = square root( (2-7)^2 + (3-6)^2 )
d(a,b) = 5.830951895
d(a,b) = 5.83

Let's say that:

• Point A is (2,3),
• Point B is (7,6)
  What is the Manhattan distance between them?

d(x,y) = | x1 - x2 | + | y1 - y2 |
d(a,b) = | 2 - 7 | + | 3 - 6 |
d(a,b) = | -5 | + | -3 |
d(a,b) = 5+3

What is machine learning?

algorithms which are able to perform certain tasks without explicitly being programmed.

What are the three ways machine learning can be divided into?

Supervised
Unsupervised
Reinforcement

What is supervised learning?

When the algorithm is trained on labelled data, meaning data for which the target answer is known, for example, if you are shown a cat picture, you are told it is a cat. This means that the algorithm learns to associate the labels with the features of the image

What is labelled data?

Data that has been assigned what it is, for example, labelled data would tell the algorithm that this picture of a cat is a cat

What is unlabelled data?

Data for which the target answer is unknown. For example, if you are shown an image but are not given any information about the image description, eg you are shown an image of a cat but are unsure what it is as you have no label.

How can supervised learning be divided into two types?

Classification
Regression

What is the supervised learning classification type?

When the output can be divided into two or more classes, for example, positive or negative, pass or fail, win or lose

What is the supervised learning regression type?

When the output variable is continuous or real-valued. For example, predicting someone's salary based on work experience