Views and History of AI
Views of AI
Typical AI Application Areas
- Natural Language Processing
- Computer Vision
- Robotics
- Theorem Proving
- Speech Recognition
- Game Playing
- Sematic Web Search
- Diagnosis
What is AI
The goal of AI is to build machines that perform tasks that normally require human intelligence. This can also be construed as what is new in computing as the computers performing the task can be seen as intelligent.
AI is both science and engineering as it is:
- the science of understanding intelligent entities: developing theories/models which attempt to explain and predict the nature of such entities.
- the engineering of such entities.
Four Views of AI
- Systems that think like humans
- Systems that act like humans
- Systems that think rationally (think the correct way)
- Systems that act rationally (act optimal to achieve a goal)
Thinking Humanly vs Acting Humanly
Computer science is concerned with acting intelligently rather than simulating a brain like a cognitive scientist might want. An abstract of this is whether they act intelligently. This is because although an entity might not think intelligently it can appear intelligent by acting intelligently.
Turing Test 1950
A human should not be able to tell whether an entity is a human or a machine.
No system has yet passed the test. In addition, it is not always the most practical or foolproof as those who are the most successful rely on tricks.
Thinking and Acting Rationally
This is the focus on making systems that act how we think that they should act. To do this we can use techniques from logic and probability theory to create machines that can reason correctly.
- Acting Rationally - Acting in such a way to achieve one’s goals optimally and given one’s beliefs.
As a result of this view, we can use techniques from economics/game theory to investigate and create machines that act rationally.
History of AI
Combinatorial Explosion - Chess
For a simple game it is possible to write a program that will select the best possible move from all possible moves. However in chess there are an exponential number of moves after each move.
The fact that a program can find a solution in principle does not mean that the program will be able to find it in practice.
Expert systems
- As general purpose brute force techniques don’t work we should use knowledge rich solutions.
- These are very specialised systems with vast amounts of knowledge about a tightly focused domain.
The main problems with expert systems are:
- The knowledge elicitation bottleneck
- Finding knowledge from experts is very time consuming and expensive
- Lack of trust in recommendations given by expert systems
- The main problem as you don’t know how the system has came to the solution.