Please enable JavaScript.
Coggle requires JavaScript to display documents.
LU5 - ARTIFICIAL INTELLIGENCE - Coggle Diagram
LU5 - ARTIFICIAL INTELLIGENCE
Connectionism & the Neural Network
Biological Neural Network
Neurons
Nerve cells specialized to receive and transmit in the nervous system
Communicate through stimulation of electric impulses.
Dendrite
Receive input and carry the to the cell soma
Cell body
Main body of the cell and contains the nucleus
Axon
Transmit signals
Artificial Neural Network (ANN)
Biological neural networks or an emulation of biological neural system
Neural networks = connections to produce dynamic memory similar to those that important in human learning
Models the human brain
Designed based on a similar principle of neurons within the human brain.
Consists of nodes, connected by input and output links.
The neural net's version of the brain synapses (artificial neural net by the node, input and output link)
Idea
Links
Weights
Networks (activation)
Set each link at a certain weight
Link is activated, weight is increased
Node will be either on or off
On
Set to fire at a particular time; send a signal to its neighbors via links (some links will conduct and some won't)
Off
Receiving signal will turn on and sending node, having fired, will turn off
Brain as learning system
Learning from repetition
Forgetting from disuse
Idea is based on:
Unpredictability
Another important property of human brain which is modelled as
well
Include a probability of firing into neural net design
Associative learning
Each time this sequence takes place, the likelihood that it will be repeated increases
The connection between node 1 and node 2 becomes stronger
A node that fire causes the firing of another which connects with the conducting link
Repetition
More repetition, more conductance
Conductivity increase by little
The tendency to conduct may be compared to human STM
Less use, less conductance
Conductivity increases a great deal
The tendency to conduct may be compared to human LTM
Early developments in neural net research
Interdisciplinary nature of cognitive science
Donald Hebb's work is important in the conceptualisation of neural networks
His principle of cell assemblies and his concept of connectionism
Perceptrons
Hypothetical nervous system realised as a neural net
Difference between perceptrons and the earlier artificial neural networks
Detects information about the world around it
How information is stored
How stored information influences recognition and behaviour
Stored information (experience)
Uses a simple form of supervised learning
Machine capable of learning, could be trained to produce a desired output for each input pattern.
The differences between Artificial and Biological Neural Networks
The beginnings
Ancient history
Renowned attempt at AI
Frankenstein - published by Mary Shelley, in 1818, describes the attempts of true scientist, Victor Frankenstein, to create life
found in Greek
mythology
Invention of human-like
artifacts
Modern history
1956 - John McCarthy, AI as the topic of the Dartmouth Conference, the first coference devoted to the subject
What is Artificial Intelligence
Focus of Intelligence in AI
the identification of the signifcance, interpretation or explanation of certain data or information (the ability to employ knowledge)
the ability to generate new ideas or concieve new perspectives on existing ideas - produncing ideas which are original and potentialy useful
the ability of drawing conclusions appropriate to the situation in hand
the process of acquiring knowledge, skills, experiences or values by study, experiences or training
the inner knowledge, without rational process-sixth sense
Understanding AI (1995)
Thought processes and reasoning
1.Human-like performance - systems that think like humans
2.Ideal performance (rationality) - systems that think rationally
System thinking rationally
Function by the rules
Laws of thought - logic
If the rukes are followed, does not make mistakes
Logical reasoning
Interruptions
Distractions
Emotions
System that thinks rationally = Intelligence machines
Deep Blue
They do not think. They simply calculate
System thinking humanly
Not about getting the right answer but
how it gets the answer.
Need to get inside the actual workings of
human minds.
Once have a sufficiently precise theory of the mind (human thought processes), it becomes poosible to express the theory as a computer program
System that thinks humanly = Intelligent Machine
In the realm of Cognitive
Science
try to reconstruct theories of how
the human mind works
create a model of intelligent
human behavior
Stimulate the model on a machine/computer to deterimine if it exhibits the same intelligent behavior as humans
Behavior
1.Human-like performance -systems that think like humans
Ideal performance (rationality) - systems that act rationally
System acting humanly
To do what HUMANS do, regardless of
whether it is done the same way as humans
Must behave according to certain normal conventions of human interaction in order to make themselves understood (speech recognition, robotics)
Should pass the Turing Test – distinguishing intelligent
entities from unintelligent ones
System that can act like human do= Intelligent Machine
System acting rationally
acting so as to achieve one's
goals, given one's beliefs.
An agent is just something
that perceives and acts.
Agent = anything that makes decisions, typically a person, firm,
An agent gets percepts one at a time, and maps this percept
sequence to actions (one action at a time)
Properties :
1.Autonomous
2.Interacts with other agents plues enviroment
3.Reactive to the environment
4.Pro-active (goal-directed)
System that acts rationally
= Intelligent Machine
AI strives to build intelligent entities as well to understand humans
The study of how to make computers/
machines do things, which, at the moment, people are better
speech recognition
smell
intuition
inferencing
learning new skills
decisions making
abstract thinking
Turing Test
Proposed by Alan Turing (1950)
Intelligence = the ability to achieve human-level performance in all cognitive tasks, sufficient to fool an interrogator
To determine if a computer program has intelligence (A computer is intelligent if machine output is indistinguishable from human output)
Cognitive Skills (Human-Like Capabilities)
Required in intelligent machine
Natural language processing
Knowledge representation
Automated reasoning
Machine learning
Computer vision (perceive objects)
Robotics (to move)
Historical Background (19th-20th Century)
The Jacquard Loom
Ancestor of Computer
A system which allow any pattern, no matter how complex, to be automatically woven onto fabric
Using a set of punched cards to govern the warp thread control system
Analytical Engine
Idea for giant calculating machine
1812: Babbage formed the idea of advanced calculating machine to calculate and print mathematical tables but was never built
1944: Howard Aiken develop Mark 1, the world's first program-controlled calculator
Turing's Proposal: Automated Computing Engine (1946)
Essential aspects of computer
Erasable memory
Convert binary form to decimal form of input & output
Logical control
Central arithmetic part to carry out fundamental arithmetic process
Advantages
Speed
Avoid human error
Carry out complicated tasks
Electronic Numeral Integrator & Calculator (ENIAC)
Von Neumann Architecture
During war: increased interest in speeding up computation to be able to calculate speed and trajectory of bombs and missiles
To calculate artillery tables making it possible to aim artillery at specific targets
Artificial Intelligence: Modeling the Brain
Natural Language Processing
Making computers talk
Systems that translate ordinary human natural language into language computers can understand and act on
Challenge
Providing the computer with a sense of context that humans have
Human language filled with ambiguity & require context for understanding/interpretation
The complexity of programming a computer to “do” human language reflects the complexity of human language
Attempt at NLP: ELIZA (1965)
The ELIZA program simulated a conversation between a patient and a psychotherapist by using a person's responses to shape the computer's replies.
When the original ELIZA first appeared in the 60's, some people actually mistook her for human.
Understanding Language
Parsers
An attempt to address the limitations of programmes such as ELIZA
Programmes that assign grammatical categories to the words in a sentence and group them into phrases
Based on the contributions of contemporary linguistics
Semantic Information Processing
Terry Winograd
Taken into account parsing, storage, and other cognitive functions crucial to understanding and using natural language
SHRDLU
A simple virtual/simulated robot.
Has natural language processing capabilities.
based on the expert system idea.
SHRDLU contains
A dictionary for syntax, semantics, and grammar.
A parser
Set of programmes to control the responses made by SHRDLU.
Script Applier Mechanism (SAM)
A computer program designed to understand stories that rely heavily on scripts
SAM makes inferences about stories
SAM – deal only with highly restricted scripts
If a machine were ever to understand a real-life story, it would have to be able to deal with enormous complexity
No rival to Human Intelligence
Attempts made to develop programs that can carry out some cognitive functions of the Human Brain
Computer programs do not have represented within their boundaries the amount of knowledge necessary to solve thetypes of problems human face with and confront every day.
Linguistics & Artificial Intelligence
Preliminary Insights
Development of computer technology
Programmed to perform human task
Related questions
Could we ask them to do more?
Would they ever reach a stage where they could also exhibit intelligence like we do?
Do we know enough about Human Intelligence to enable us to create intelligent machines?
Machine Translation
1960s – machine perform simple linguistic task
Automatic translation
Difficulty with translation - Meaning
Machine translation works best only when vocabulary involved limited to very specific domain
Natural Language Processing
Computer technology that handles human natural language
Programs in which sentence pattern frames stored as data
Translate ordinary human commands into language computers can understand & act on
The goal is “to achieve human-like language processing”.
Renown program = ELIZA
Matching input to fixed pattern
Complexity of programming a computer to ‘do’ human language reflects the vast complexity of human language
Other Language-Related Tasks
Due to its suitability in storing & retrieving large amount if data, computer technology can undertake tasks which would be daunting for human to attempt
Concordance
Speech recognition
Robotics
A system that contains sensors, control systems, manipulators, power supplies and software all working together to perform a task.
Essential features
Movement
Energy (Power)
Sensors
Intelligence
Artificial Life : Animats
Based on behaviour
E.g. : insect (simple behaviours)
A simulated animal or a robot whose structure and functionalities are inspired from current biological knowledge
Cog Project
Human intelligence more complex than of simpler life forms
Predicted on the notion that these attributes of human intelligence can be exploited in building a robotic system
Building a Humanoid Robot = resembling human
Relies on knowledge from many fields
Psychology
Physiology and psychology
Linguistics
Neuroscience, etc.
The Machine Performs
Human Logic
Human exercise reasoning when solving problems
Logical steps
Vast & complex world knowledge as background
To program computer so that it can engage in reasoning like that of human, the vast, complex world must be simplified
Logic Theorist Program
First program deliberately engineered to mimic the problem solving skills of a human being and is called "the first artificial intelligence program” (Newell, Shaw & Simon, 1956)
The program succeeded in proving thirty-eight of the first fifty-two theorems presented in Chapter 2 of Principia Mathematica
the program also found a proof for one theorem which was more elegant than the one provided by the text’s writers.
Example of a theorem :Given that either X or Y is true, and given further that Y is in fact false, it follows that X is true
The General Problem Solver (GPS)
Certain types of problem, whose solutions lend themselves to a very explicit and clear steps, can be formulated for the GPS
Searching among and following a series of elementary reasoning steps, similar to those carried out by human
The only difference is that GPS does NOT have access to information outside of the domain that it is programmed to have
But researchers soon realized that developing such general purpose systems was too difficult - better to focus on systems for limited domains.
Expert Systems
We all depend on HUMAN expert assistance in our
lives
doctors, attorneys, automobile mechanics, computer
repairmen, etc
Can expert assistance be given by computer
programs?
Prospector: identify sites for drilling of oil
Dendral: suggest the chemical structure of unknown
compounds
MYCIN: choose appropriate antibiotics for patients with
severe bacterial infections
what is an expert system?
Systems which encode human expertise in limited
domains.
a computer program designed to hold the accumulated knowledge of one or more domain experts