A Definition of Artificial Intelligence And Its History As A Practical Research Field
Since the dawn of science fiction, the concept of man-made helpers, thinking machines that can do undirected work, have been a staple of adventure fiction, ranging from Mary Shelley's Frankenstein to the friendly droids of Star Wars. The process by which programmatic entities gain the ability to take input, adapt their behaviors and solve problems is a practical definition of artificial intelligence.
A Definition of Artificial Intelligence
Working from a definition of artificial intelligence to something that has real world applications has been a process with lots of false starts and programmatic blind alleys. The initial research into artificial intelligence tried to work from formal problem solving methods, using deductive and inductive logic processes.
Initial progress, working from Alan Turing's insight on the binary nature of mathematics, was quite rapid - the initial foundations of artificial intelligence research was founded in the 1950s, and many of the luminaries of the field got started there, from John McCarthy to Marvin Mnsky and Allen Newell. The great founding of AI laboratories in the 1950s promised that within twenty years, a computer would be able to do anything a human mind could.
What hindered this is that human brains don't actually use deductive problem solving; they use inexact neural mapping techniques.
Attempting to brute force cognition with AI software results in an explosion of options and considerations that need to be iteratively searched, and rapidly outpaced the ability of computers in the '50s and '60s to handle it. Even within their limitations, the early AI programs made seemingly astonishing progress, with the ability to solve word problems algebraicly and proving logical theorems.
This 'amazing advance' showcases one of the primary differences between human beings and computers: It's much easier to get computers to do things logically than it is to get humans to knuckle down and apply formal logic structures to problems.
Even so, the expansion of the field in the 1960s lead to widespread government funding...most of which went away when the operational definition of artificial intelligence showcased its limits (driven by limitations of computation architectures) in the 1970s.
The 1980s saw a brief resurgence of AI research when computers got powerful enough and cheap enough to run expert system databases, which used a limited form of some of the artificial intelligence techniques to allow a user to ask questions of a computer within a very narrowly defined field and get responses that 'made sense'.
This definition of artificial intelligence resulted in AI 'savants' of sorts, and began the field of Bayesian stochastic research which underlies all the modern search engine technologies.
The 1990s saw major advances in computational capabilities, and algorithms that found shortcuts for the brute force applications of AI-style research.
These algorithms can be thought of as logical filters or process filters, and allow AI researchers to work with incomplete or 'fuzzy' data, allowing the definition of artificial intelligence to become a bit more humanistic in what it can be expected to handle.
