Artificial Intelligence

By: The FHE Team

By Bradley LaChance

Enough is known about Artificial Intelligence today such that it’s not necessary to “hype” it to get people excited about the very real prospect of machine intelligence.  In this section of the website, we’re going to provide some straight-forward information on exactly what Artificial Intelligence is and do some reporting on where leaders in the field stand on its progress today.

In later pages we will employ some futuring techniques to explore the various assumptions and implications of what it may mean to actually live with such entities in the very near future and what impact it may have on the future of human evolution.

The Goals of Artificial Intelligence

Stuart Russell and Peter Norvig in their runaway best-selling text book “Artificial Intelligence: A Modern Approach” now adopted by over 1200 universities around the globe, have synthesized a definition, or more precisely the goals of artificial intelligence as being the branch of science concerned with creating machines capable of human-like thought and/or behavior, or rational thought and/or behavior.  This matrix, they say, captures the vast majority of what people in the field have aimed to do over the years.

The Goals of Artificial Intelligence

The largest distinction being made is human vs. rational. It is not to say that humans are necessarily irrational, but we are not exactly pure logical machines either; we have specific ways of thinking developed over unknowable stretches of evolutionary development and we are subject to a large number of built-in cognitive biases which presumably served some purpose in our history or which evolution has not yet had a chance to eliminate.

Their thinking vs. behavior distinction boils down simply: a machine “thinking” about a problem merely takes an input, processes it (either like a human would or somehow entirely different) and produces an answer, perhaps on a computer screen.  A machine that must behave has to “do” something; maneuver about the three-dimensional world to achieve its goal/reach its target.

Thinking

Human Thinking – Some scientists, both those interested in producing a human-like artificial intelligence and those interested in better understanding how the human mind operates, aren’t satisfied with just developing a reasoning machine but strive to develop one that uses human thought patterns and “subroutines”.  The interdisciplinary field of cognitive science brings together computer models from AI and experimental techniques from psychology to construct precise and testable theories of the human mind.

Rational Thinking – The goal of creating a pure rational-thinking artificial intelligence is one of applying pure logical processing in between the input and the output.  This “logicist” approach is marked by the need to leverage formal logical notation (symbols denoting things and their relationship(s) originally developed in the 19th century) to distinctly represent the world.  The major drawback to this approach is that the world is neither as formal nor distinct as a pure logical approach/notation requires.

Behavior

Human Behavior – The goal of human behavior is best summed up by the now well-known Turing test in which an interviewer is not supposed to be able to distinguish between a human or machine interviewee as communicated through a keyboard.  Presumably this type of artificial intelligence would have all the characteristics of a human including a sense of humor, distinctive personality traits, and perhaps some idiosyncrasies- real ones, not imitated. Given our limited knowledge of the human psyche let alone how to create one, this goal remains in the realm of science fiction, though each day research makes it more plausible.

Rational Behavior – Here’s where the rubber meets the road, so to speak.  Main-stream (though by no means all of) artificial intelligence education, research, and application are geared toward the creation of a “rational agent”.  That is, an entity that can take a goal, can perceive, learn, and store information about its environment, then devise and take a course of action intended to produce the best available outcome given the information available at the time.

Artificial Intelligence's Rational Agent

While this “rational agent” objective does not by any means make the task of accomplishing artificial intelligence simple, it does eliminate the need to force the world into a black and white representational system (the logicist approach), as well as the need to represent little-understood thought patterns born of evolution.  An example Russell and Norvig use to illustrate the correctness of this approach is the invention of flight: only when man decided to quit trying to emulate a bird and start studying aerodynamics and using wind tunnels, did we learn to fly.

Measuring Progress toward True Artificial intelligence

Regardless of whether we are pursuing an artificial intelligence that behaves like a human, or one that behaves rationally, the six subfields of study below remain the same and constitute the bulk of modern artificial intelligence activity.

  • Natural language processing to enable it to communicate successfully in human language;
  • Knowledge representation to store information provided before or during an attempt at accomplishing a given goal;
  • Automated reasoning to use the stored information to answer questions and to draw new conclusions;
  • Machine learning to adapt to new circumstances and to detect and extrapolate patterns.

For an AI to get around the physical environment you’d have to tack on:

  • Computer vision (which does not have to be within humans’ visual range and could include sonar/radar capability), and
  • Robotics for locomotion and manipulation of matter (objects).

One additional field being pursued toward developing human-like intelligence is:

  • Neural Networks that try to emulate the physical construct of the brain’s complex of neurons.

As we continue to build and update the material in this section of the website, we’ll use the subfields as defined above to benchmark where state-of-the-art development stands against the most daunting challenges faced in each field.