MelissaA
Brenton
Shilpa Karhik
A bluffer's guide to AI
July 6th 2006 01:33
Well in a similar vein to last weeks post, today's post is going to be a detailed post about artificial intelligence.
What is AI?
Artificial intelligence is basically intelligence exhibited by an artificial construct. Now, I realise that that definition involves using both the words in the phrase artificial intelligence itself, but it is a good summary.
Artificial intelligence is normally referred to when we talk about artificial intelligence research, which aims to produce such a form of research. This form of research has historically looked at many issues, including machine vision, pattern recognition and robotics.
What are some divisions in AI?
There are two major types of divisions in the study of artificial intelligence: the division between computational and conventional AI, and that between strong and weak AI.
Conventional AI relies on machine learning techniques and often focuses on developing the reasoning skills of the computer, whereas computational AI uses iterative learning techniques as such as evolutionary learning.
The weak vs strong AI division is based on the type of AI that is aimed for. Proponents of weak AI say that consciousness can never be achieved by an AI, and instead they are just imitating humanity. Strong AI disagrees with this, for a number of reasons.
Uses of AI
One of the main uses of artificial intelligence these days is in predicting trends. This is used by businesses to help them run efficiently. AI is also useful in developing autonomous control, which helps with such applications as developing unmanned military planes.
Major techniques of AI (case study)
I will look at some of the major techniques used in automated planning. Automated planning is the attempt to help AI's take in information about the environment and then use this to help achieve a goal.
This demonstrates a key technique in AI, which is trial and error learning. Such a technique can be achieved through reinforcement, which gives rewards for behaviour that works, and punishment for behaviour that doesn't, and by combinatorial optimisation, which helps an AI choose from a large number of options.
Automated planning is generally achieved through either forward or backward chaining. Forward chaining uses inference to move from knowledge about the state of the world, to a solution to the problem. Backwards chaining moves backwards from a goal, searching the conditions that could make this true.
For example, if the aim of an AI was to discover where Mr. Toombie lived, given that the man wearing black lives near the Sydney Opera house, and it had the following rules in its head:
1. If the man wearing black lives near the Sydney Opera House then the man who is wearing black lives in Sydney
2. If man who is wearing black lives in Sydney Then Mr Toombie lives in Sydney
then the two techniques would both reach the correct answer in different ways.
Forward chaining would search for an if statement that matched what it had been given (that the man wearing black lives near the Sydney Opera House) and would find statement one. It would then search for an if statement that matched statement ones then statement and would come up with statement two. Forward chaining has now given us our answer.
Backwards chaining would search the then statements for the answer we were looking for and would find statement 2. Next, it would search for a then statement that matched statement 2's if statement. It would find rule 1. As the if clause is true the conclusion is known to be true.
A brief history of artificial intelligence
The early stages of artificial intelligence occurred before any computers existed and by the 1930's some important AI theory had been developed. The most important of this problem came from Godel and Turing.
Godel said that there were some maths theorems that we know to be true, but that can't be proved. This was taken as evidence that humans could see truth, which machines would never be able to deduce logically.
Turing conceived of the idea of a Universal Turing Machine, which could mimic any other computational device. While he thought such a machine would have limits, he still believed it would be able to think. The Church-Turing thesis, which he and Church came up with independently states that any human solvable problem can be reduced to an algorithm. Later, Turing developed the Turing test.
In the 1950's the first AI was developed, called Logical Theorist, this used rules to satisfy goals. This decade saw the emergence of AI as an important field and led to the goals of improving response to trial and error learning problems and making computers learn without author input.
In the 1960's a huge number of AI projects began and these led to substantial increases in knowledge. MIT's AI Laboratory improved computer vision and ability to process natural languages. These were major steps forward. This development continued in the 1970's.
In the 1980's AI began to gain more and more commercial applications and by the end of this decade there was widespread use of AI in some industries and this process accelerated in the 1990's when AI became extremely widespread across the business world.
Deep Blue's 1997 defeat of the world chess champion, Gary Kasparov was symbolic of the rapid improvement of AI systems.
Well that's it for now
Adam
What is AI?
Artificial intelligence is basically intelligence exhibited by an artificial construct. Now, I realise that that definition involves using both the words in the phrase artificial intelligence itself, but it is a good summary.
Artificial intelligence is normally referred to when we talk about artificial intelligence research, which aims to produce such a form of research. This form of research has historically looked at many issues, including machine vision, pattern recognition and robotics.
What are some divisions in AI?
There are two major types of divisions in the study of artificial intelligence: the division between computational and conventional AI, and that between strong and weak AI.
Conventional AI relies on machine learning techniques and often focuses on developing the reasoning skills of the computer, whereas computational AI uses iterative learning techniques as such as evolutionary learning.
The weak vs strong AI division is based on the type of AI that is aimed for. Proponents of weak AI say that consciousness can never be achieved by an AI, and instead they are just imitating humanity. Strong AI disagrees with this, for a number of reasons.
Uses of AI
One of the main uses of artificial intelligence these days is in predicting trends. This is used by businesses to help them run efficiently. AI is also useful in developing autonomous control, which helps with such applications as developing unmanned military planes.
Major techniques of AI (case study)
I will look at some of the major techniques used in automated planning. Automated planning is the attempt to help AI's take in information about the environment and then use this to help achieve a goal.
This demonstrates a key technique in AI, which is trial and error learning. Such a technique can be achieved through reinforcement, which gives rewards for behaviour that works, and punishment for behaviour that doesn't, and by combinatorial optimisation, which helps an AI choose from a large number of options.
Automated planning is generally achieved through either forward or backward chaining. Forward chaining uses inference to move from knowledge about the state of the world, to a solution to the problem. Backwards chaining moves backwards from a goal, searching the conditions that could make this true.
For example, if the aim of an AI was to discover where Mr. Toombie lived, given that the man wearing black lives near the Sydney Opera house, and it had the following rules in its head:
1. If the man wearing black lives near the Sydney Opera House then the man who is wearing black lives in Sydney
2. If man who is wearing black lives in Sydney Then Mr Toombie lives in Sydney
then the two techniques would both reach the correct answer in different ways.
Forward chaining would search for an if statement that matched what it had been given (that the man wearing black lives near the Sydney Opera House) and would find statement one. It would then search for an if statement that matched statement ones then statement and would come up with statement two. Forward chaining has now given us our answer.
Backwards chaining would search the then statements for the answer we were looking for and would find statement 2. Next, it would search for a then statement that matched statement 2's if statement. It would find rule 1. As the if clause is true the conclusion is known to be true.
A brief history of artificial intelligence
The early stages of artificial intelligence occurred before any computers existed and by the 1930's some important AI theory had been developed. The most important of this problem came from Godel and Turing.
Godel said that there were some maths theorems that we know to be true, but that can't be proved. This was taken as evidence that humans could see truth, which machines would never be able to deduce logically.
Turing conceived of the idea of a Universal Turing Machine, which could mimic any other computational device. While he thought such a machine would have limits, he still believed it would be able to think. The Church-Turing thesis, which he and Church came up with independently states that any human solvable problem can be reduced to an algorithm. Later, Turing developed the Turing test.
In the 1950's the first AI was developed, called Logical Theorist, this used rules to satisfy goals. This decade saw the emergence of AI as an important field and led to the goals of improving response to trial and error learning problems and making computers learn without author input.
In the 1960's a huge number of AI projects began and these led to substantial increases in knowledge. MIT's AI Laboratory improved computer vision and ability to process natural languages. These were major steps forward. This development continued in the 1970's.
In the 1980's AI began to gain more and more commercial applications and by the end of this decade there was widespread use of AI in some industries and this process accelerated in the 1990's when AI became extremely widespread across the business world.
Deep Blue's 1997 defeat of the world chess champion, Gary Kasparov was symbolic of the rapid improvement of AI systems.
Well that's it for now
Adam
| 61 |
| Vote |
