Overview of Artificial Intelligence
1. Origin
In the 1940s, various computers were invented, initially for numerical calculation, though they were also used in non-numerical computation or symbol manipulation, as anticipated in theoretical models like Turing machine.
Several visionaries had noticed the important similarities between the human brain/mind and the machine:
AI as a research field started in 1956 at the Dartmouth meeting, and was strongly influenced by
John McCarthy,
Marvin Minsky,
Allen Newell,
Herbert A. Simon.
2. The big picture
The goal of AI is to build computer systems that work like the human mind. Consequently, AI can be considered both as a branch of science (related to cognitive science) and a branch of engineering (related to computer engineering and software engineering). An AI project often has results on three levels:
- a theory of intelligence (described in a human language)
- a model of the theory (specified in a symbolic/mathematical language)
- an implementation of the model (realized in a programming language or a piece of hardware)
Basic questions about the whole field:
- What is AI?
- Can AI be built?
- How to build AI?
- Should AI be built?
There are still different opinions on each of them.
3. Working definitions
In which aspect is an AI similar to a human mind? The answers can be clustered into the following groups:
-
[Structure]
Rationale: Human intelligence is produced by the human brain.
Background: Neuroscience, biology, etc.
Challenge: There may be biological details that are neither possible nor necessary to be reproduced in AI systems.
-
[Behavior]
Rationale: Human intelligence is displayed in how the human beings behave.
Background: Psychology, linguistics, etc.
Challenge: There may be psychological or social factors that are neither possible nor necessary to be reproduced in AI systems.
-
[Capability]
Rationale: Human intelligence is evaluated by problem-solving capability.
Background: Computer application guided by domain knowledge
Challenge: There is no defining problems of intelligence, and the special-purpose solutions lack generality and flexibility.
-
[Function]
Rationale: Human intelligence is associated to a collection of cognitive functionality.
Background: Computer science
Challenge: The AI techniques developed so far are highly fragmented and rigid, and it is hard for them to work together on novel problems.
-
[Principle]
Rationale: Human intelligence represents a form of rationality.
Background: Logic, mathematics, etc.
Challenge: There are too many phenomena in intelligence and cognition to be explained and reproduced by a simple theory.
These approaches are related, though correspond to different levels of abstraction of human intelligence. See a detailed analysis and the following discussion.
4. The agent framework
A widespread approach takes intelligence as a set of cognitive functions integrated in an intelligent agent framework, consisting of
- interfaces with the environment: natural language (understanding and generating), perception (vision, hearing, touch, etc.), and action (robotics)
- a common substance: knowledge representation and organization
- cognitive functions: searching, reasoning, learning, planning, predicting, decision making, problem solving, etc.
In recent years, deep learning and large language model have achieved great successes. However, they still cannot replace other AI techniques.
Reading
- Poole and Mackworth, Chapters 1 & 2
- Russell and Norvig, Chapters 1 & 2
- Luger, Chapters 1