Development of AI

• Charles Babbage and the Analytical Engine (1842)
• Boole, An investigation into the Laws of Thought (1815-64)
• Russell and Wittgenstein, work on logic and language (1889-1951)
• Turing proposed that machines could be programmed to exhibit intelligence (1950, Turing Test).
• Claude Shannon and playing chess by computer (1950) .
• Dartmouth Conference and John McCarthy, Herb Simon, John Newell, Marvin Minsky (1956).
• Evans and the ANALOGY program (1964) to solve the architypal geometrical IQ tests (see later in this lecture).
• Symbolic integration programs leading to packages such as MACSYMA, MAPLE, SMP and MATHEMATICA.
• Waltz's line-labelling. Roberts (1965) produced an object recognition system which could recognize simple wedge and cube shapes from imagery.
• Rule-based expert systems in 1970's included MYCIN, RENDRAL, XCON, INTERNIST, PROSPECTOR.

``In considering any new subject, there is frequently a tendency first to overstate what we already find to be interesting or remarkable; and secondly by a sort of natural reaction to undervalue the true state of the case when we do discover that our notions have surpassed those that were really tenable.''

The Essential Ingredients of AI

1. Intelligence needs knowledge.
   
2. Knowledge needs representation.
   
3. Inter- and intra-representation changes require reasoning processes
   
4. AI systems need to learn.

Emerging Principals from AI Research

1. A Physical Symbol System is Necessary and Sufficient for Intelligent Action?
2. Bounded Rationality implies Opportunistic Search.
3. Search Compensates for Lack of Knowledge.
4. Knowledge compensates for Lack of Search.
5. Expertise comes in chunks of 70,000 ± 20,000!

When will we know when we have made it?

Turing suggested in his article, 'Computing Machinery and Intelligence' (1950) that a truly intelligent machine will become indistinguishable from a human in conversation. He proposed to demonstarte this via an interrogation test whereby an interrogator would set questions to be answered by a human and a computer, both attempting to prove themselves as the human. Printed copies of their responses would be returned to the interragator, when the machine could not be reliably identified it would be truly intelligent.
For Turing's article in full visit -

http://www.oxy.edu/departments/cog-sci/courses/1998/cs101/texts/Computing-machinery.html

The following questions suggest what we require.

Is the task clearly defined?
Is there an implemented procedure for the task?
Is there a set of regularities or constraints from which the procedure derives its power?
Is the procedure generic?
Does it degrade gracefully in the presence of noise?

Recently Newell suggested the following attributes for an AI system :

operates in real time
exploits vast amounts of knowledge
tolerates errorful, unknown and unexpected inputs
uses symbols and abstractions
communicates using natural language
learns from the environment
exhibits adaptive goal-oriented behaviours

<<<Previous Next>>>